program(1.3)
[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3520.4.1"}, {"coremlc-version", "3520.5.1"}})]
{
    func main<ios18>(tensor<fp32, [1]> current_step, tensor<fp32, [1, 1, ?]> latent_mask, tensor<fp32, [1, 144, ?]> noisy_latent, tensor<fp32, [1, 50, 256]> style_ttl, tensor<fp32, [1, 256, ?]> text_emb, tensor<fp32, [1, 1, ?]> text_mask, tensor<fp32, [1]> total_step) [FlexibleShapeInformation = tuple<tuple<string, dict<string, tensor<int32, [?]>>>, tuple<string, dict<string, list<tensor<int32, [2]>, ?>>>>((("DefaultShapes", {{"latent_mask", [1, 1, 24]}, {"noisy_latent", [1, 144, 24]}, {"text_emb", [1, 256, 24]}, {"text_mask", [1, 1, 24]}}), ("RangeDims", {{"latent_mask", [[1, 1], [1, 1], [17, 512]]}, {"noisy_latent", [[1, 1], [144, 144], [17, 512]]}, {"text_emb", [[1, 1], [256, 256], [17, 512]]}, {"text_mask", [[1, 1], [1, 1], [17, 512]]}})))] {
            string text_emb_to_fp16_dtype_0 = const()[name = string("text_emb_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1, 256, ?]> text_emb_to_fp16 = cast(dtype = text_emb_to_fp16_dtype_0, x = text_emb)[name = string("cast_19")];
            tensor<int32, [3]> var_34_shape_cast_fp16 = shape(x = text_emb_to_fp16)[name = string("op_34_shape_cast_fp16")];
            int32 gather_1_axis_0 = const()[name = string("gather_1_axis_0"), val = int32(0)];
            int32 gather_1_batch_dims_0 = const()[name = string("gather_1_batch_dims_0"), val = int32(0)];
            bool gather_1_validate_indices_0 = const()[name = string("gather_1_validate_indices_0"), val = bool(false)];
            string var_34_shape_cast_fp16_to_int16_dtype_0 = const()[name = string("op_34_shape_cast_fp16_to_int16_dtype_0"), val = string("int16")];
            uint16 gather_1_indices_0_to_uint16 = const()[name = string("gather_1_indices_0_to_uint16"), val = uint16(2)];
            tensor<int16, [3]> var_34_shape_cast_fp16_to_int16 = cast(dtype = var_34_shape_cast_fp16_to_int16_dtype_0, x = var_34_shape_cast_fp16)[name = string("cast_18")];
            int16 gather_1_cast_uint16 = gather(axis = gather_1_axis_0, batch_dims = gather_1_batch_dims_0, indices = gather_1_indices_0_to_uint16, validate_indices = gather_1_validate_indices_0, x = var_34_shape_cast_fp16_to_int16)[name = string("gather_1_cast_uint16")];
            string gather_1_cast_uint16_to_int32_dtype_0 = const()[name = string("gather_1_cast_uint16_to_int32_dtype_0"), val = string("int32")];
            int32 var_38 = const()[name = string("op_38"), val = int32(-1)];
            int32 var_39 = const()[name = string("op_39"), val = int32(0)];
            int32 gather_2 = const()[name = string("gather_2"), val = int32(1)];
            int32 concat_0_axis_0 = const()[name = string("concat_0_axis_0"), val = int32(0)];
            bool concat_0_interleave_0 = const()[name = string("concat_0_interleave_0"), val = bool(false)];
            int32 gather_1_cast_uint16_to_int32 = cast(dtype = gather_1_cast_uint16_to_int32_dtype_0, x = gather_1_cast_uint16)[name = string("cast_17")];
            tensor<int32, [3]> concat_0 = concat(axis = concat_0_axis_0, interleave = concat_0_interleave_0, values = (gather_2, var_38, gather_1_cast_uint16_to_int32))[name = string("concat_0")];
            tensor<int32, [3]> shape_0 = const()[name = string("shape_0"), val = tensor<int32, [3]>([1, 256, 1])];
            int32 equal_0_y_0 = const()[name = string("equal_0_y_0"), val = int32(-1)];
            tensor<bool, [3]> equal_0 = equal(x = concat_0, y = equal_0_y_0)[name = string("equal_0")];
            tensor<int32, [3]> select_0 = select(a = shape_0, b = concat_0, cond = equal_0)[name = string("select_0")];
            tensor<int32, [3]> real_div_0 = real_div(x = select_0, y = shape_0)[name = string("real_div_0")];
            tensor<fp16, [1, 256, 1]> uncond_masker_text_special_token_to_fp16 = const()[name = string("uncond_masker_text_special_token_to_fp16"), val = tensor<fp16, [1, 256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))];
            tensor<fp16, [?, ?, ?]> text_uncond_cast_fp16 = tile(reps = real_div_0, x = uncond_masker_text_special_token_to_fp16)[name = string("text_uncond_cast_fp16")];
            bool text_emb_interleave_0 = const()[name = string("text_emb_interleave_0"), val = bool(false)];
            tensor<fp16, [?, 256, ?]> text_emb_cast_fp16 = concat(axis = var_39, interleave = text_emb_interleave_0, values = (text_emb_to_fp16, text_uncond_cast_fp16))[name = string("text_emb_cast_fp16")];
            bool input_83_interleave_0 = const()[name = string("input_83_interleave_0"), val = bool(false)];
            string style_ttl_to_fp16_dtype_0 = const()[name = string("style_ttl_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1, 50, 256]> uncond_masker_style_value_special_token_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [1, 50, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(640))), scale = tensor<fp16, [1, 1, 1]>([[[0x1.bb4p-9]]]))[name = string("uncond_masker_style_value_special_token_to_fp16_quantized")];
            tensor<fp16, [1, 50, 256]> style_ttl_to_fp16 = cast(dtype = style_ttl_to_fp16_dtype_0, x = style_ttl)[name = string("cast_16")];
            tensor<fp16, [2, 50, 256]> input_83_cast_fp16 = concat(axis = var_39, interleave = input_83_interleave_0, values = (style_ttl_to_fp16, uncond_masker_style_value_special_token_to_fp16_quantized))[name = string("input_83_cast_fp16")];
            int32 var_64 = const()[name = string("op_64"), val = int32(0)];
            bool input_5_interleave_0 = const()[name = string("input_5_interleave_0"), val = bool(false)];
            string noisy_latent_to_fp16_dtype_0 = const()[name = string("noisy_latent_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1, 144, ?]> noisy_latent_to_fp16 = cast(dtype = noisy_latent_to_fp16_dtype_0, x = noisy_latent)[name = string("cast_15")];
            tensor<fp16, [2, 144, ?]> input_5_cast_fp16 = concat(axis = var_64, interleave = input_5_interleave_0, values = (noisy_latent_to_fp16, noisy_latent_to_fp16))[name = string("input_5_cast_fp16")];
            int32 var_67 = const()[name = string("op_67"), val = int32(0)];
            bool latent_mask_b_interleave_0 = const()[name = string("latent_mask_b_interleave_0"), val = bool(false)];
            string latent_mask_to_fp16_dtype_0 = const()[name = string("latent_mask_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1, 1, ?]> latent_mask_to_fp16 = cast(dtype = latent_mask_to_fp16_dtype_0, x = latent_mask)[name = string("cast_14")];
            tensor<fp16, [2, 1, ?]> latent_mask_b_cast_fp16 = concat(axis = var_67, interleave = latent_mask_b_interleave_0, values = (latent_mask_to_fp16, latent_mask_to_fp16))[name = string("latent_mask_b_cast_fp16")];
            int32 var_70 = const()[name = string("op_70"), val = int32(0)];
            bool text_mask_interleave_0 = const()[name = string("text_mask_interleave_0"), val = bool(false)];
            string text_mask_to_fp16_dtype_0 = const()[name = string("text_mask_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1, 1, ?]> text_mask_to_fp16 = cast(dtype = text_mask_to_fp16_dtype_0, x = text_mask)[name = string("cast_13")];
            tensor<fp16, [2, 1, ?]> text_mask_cast_fp16 = concat(axis = var_70, interleave = text_mask_interleave_0, values = (text_mask_to_fp16, text_mask_to_fp16))[name = string("text_mask_cast_fp16")];
            string current_step_to_fp16_dtype_0 = const()[name = string("current_step_to_fp16_dtype_0"), val = string("fp16")];
            string total_step_to_fp16_dtype_0 = const()[name = string("total_step_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1]> current_step_to_fp16 = cast(dtype = current_step_to_fp16_dtype_0, x = current_step)[name = string("cast_11")];
            tensor<fp16, [1]> total_step_to_fp16 = cast(dtype = total_step_to_fp16_dtype_0, x = total_step)[name = string("cast_12")];
            tensor<fp16, [1]> t_1_cast_fp16 = real_div(x = current_step_to_fp16, y = total_step_to_fp16)[name = string("t_1_cast_fp16")];
            int32 var_74 = const()[name = string("op_74"), val = int32(0)];
            bool t_3_interleave_0 = const()[name = string("t_3_interleave_0"), val = bool(false)];
            tensor<fp16, [2]> t_3_cast_fp16 = concat(axis = var_74, interleave = t_3_interleave_0, values = (t_1_cast_fp16, t_1_cast_fp16))[name = string("t_3_cast_fp16")];
            int32 var_78 = const()[name = string("op_78"), val = int32(-1)];
            tensor<int32, [2]> var_85 = const()[name = string("op_85"), val = tensor<int32, [2]>([-1, 1])];
            tensor<fp16, [2, 1]> var_86_cast_fp16 = reshape(shape = var_85, x = t_3_cast_fp16)[name = string("op_86_cast_fp16")];
            fp16 time_encoder_time_scale_to_fp16 = const()[name = string("time_encoder_time_scale_to_fp16"), val = fp16(0x1.f4p+9)];
            tensor<fp16, [2, 1]> t_scaled_cast_fp16 = mul(x = var_86_cast_fp16, y = time_encoder_time_scale_to_fp16)[name = string("t_scaled_cast_fp16")];
            tensor<fp16, [1, 32]> time_encoder_omegas_to_fp16 = const()[name = string("time_encoder_omegas_to_fp16"), val = tensor<fp16, [1, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13504)))];
            tensor<fp16, [2, 32]> phase_cast_fp16 = mul(x = t_scaled_cast_fp16, y = time_encoder_omegas_to_fp16)[name = string("phase_cast_fp16")];
            tensor<fp16, [2, 32]> var_89_cast_fp16 = sin(x = phase_cast_fp16)[name = string("op_89_cast_fp16")];
            tensor<fp16, [2, 32]> var_90_cast_fp16 = cos(x = phase_cast_fp16)[name = string("op_90_cast_fp16")];
            bool input_1_interleave_0 = const()[name = string("input_1_interleave_0"), val = bool(false)];
            tensor<fp16, [2, 64]> input_1_cast_fp16 = concat(axis = var_78, interleave = input_1_interleave_0, values = (var_89_cast_fp16, var_90_cast_fp16))[name = string("input_1_cast_fp16")];
            tensor<fp16, [256, 64]> time_encoder_mlp_0_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [256, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13632))), scale = tensor<fp16, [256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30080))))[name = string("time_encoder_mlp_0_linear_weight_to_fp16_quantized")];
            tensor<fp16, [256]> time_encoder_mlp_0_linear_bias_to_fp16 = const()[name = string("time_encoder_mlp_0_linear_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30656)))];
            tensor<fp16, [2, 256]> linear_0_cast_fp16 = linear(bias = time_encoder_mlp_0_linear_bias_to_fp16, weight = time_encoder_mlp_0_linear_weight_to_fp16_quantized, x = input_1_cast_fp16)[name = string("linear_0_cast_fp16")];
            tensor<fp16, [2, 256]> var_97_cast_fp16 = softplus(x = linear_0_cast_fp16)[name = string("op_97_cast_fp16")];
            tensor<fp16, [2, 256]> var_98_cast_fp16 = tanh(x = var_97_cast_fp16)[name = string("op_98_cast_fp16")];
            tensor<fp16, [2, 256]> input_3_cast_fp16 = mul(x = linear_0_cast_fp16, y = var_98_cast_fp16)[name = string("input_3_cast_fp16")];
            tensor<fp16, [64, 256]> time_encoder_mlp_2_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [64, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31232))), scale = tensor<fp16, [64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47680))))[name = string("time_encoder_mlp_2_linear_weight_to_fp16_quantized")];
            tensor<fp16, [64]> time_encoder_mlp_2_linear_bias_to_fp16 = const()[name = string("time_encoder_mlp_2_linear_bias_to_fp16"), val = tensor<fp16, [64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47872)))];
            tensor<fp16, [2, 64]> linear_1_cast_fp16 = linear(bias = time_encoder_mlp_2_linear_bias_to_fp16, weight = time_encoder_mlp_2_linear_weight_to_fp16_quantized, x = input_3_cast_fp16)[name = string("linear_1_cast_fp16")];
            tensor<int32, [1]> time_emb_axes_0 = const()[name = string("time_emb_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2, 64, 1]> time_emb_cast_fp16 = expand_dims(axes = time_emb_axes_0, x = linear_1_cast_fp16)[name = string("time_emb_cast_fp16")];
            string var_115_pad_type_0 = const()[name = string("op_115_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_115_strides_0 = const()[name = string("op_115_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_115_pad_0 = const()[name = string("op_115_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_115_dilations_0 = const()[name = string("op_115_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_115_groups_0 = const()[name = string("op_115_groups_0"), val = int32(1)];
            tensor<fp16, [512, 144, 1]> proj_in_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 144, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48064))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121856))))[name = string("proj_in_weight_to_fp16_quantized")];
            tensor<fp16, [2, 512, ?]> var_115_cast_fp16 = conv(dilations = var_115_dilations_0, groups = var_115_groups_0, pad = var_115_pad_0, pad_type = var_115_pad_type_0, strides = var_115_strides_0, weight = proj_in_weight_to_fp16_quantized, x = input_5_cast_fp16)[name = string("op_115_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_3_cast_fp16 = mul(x = var_115_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_3_cast_fp16")];
            int32 var_126 = const()[name = string("op_126"), val = int32(-1)];
            tensor<fp16, [2, 512, ?]> input_7_cast_fp16 = mul(x = x_3_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("input_7_cast_fp16")];
            tensor<int32, [6]> input_9_pad_0 = const()[name = string("input_9_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_9_mode_0 = const()[name = string("input_9_mode_0"), val = string("replicate")];
            fp16 const_0_to_fp16 = const()[name = string("const_0_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_9_cast_fp16 = pad(constant_val = const_0_to_fp16, mode = input_9_mode_0, pad = input_9_pad_0, x = input_7_cast_fp16)[name = string("input_9_cast_fp16")];
            string h_3_pad_type_0 = const()[name = string("h_3_pad_type_0"), val = string("valid")];
            int32 h_3_groups_0 = const()[name = string("h_3_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_3_strides_0 = const()[name = string("h_3_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_3_pad_0 = const()[name = string("h_3_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_3_dilations_0 = const()[name = string("h_3_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 5]> main_blocks_0_convnext_0_0_dwconv__conv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122944))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125568))))[name = string("main_blocks_0_convnext_0_0_dwconv__conv_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_0_convnext_0_0_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_0_0_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126656)))];
            tensor<fp16, [2, 512, ?]> h_3_cast_fp16 = conv(bias = main_blocks_0_convnext_0_0_dwconv__conv_bias_to_fp16, dilations = h_3_dilations_0, groups = h_3_groups_0, pad = h_3_pad_0, pad_type = h_3_pad_type_0, strides = h_3_strides_0, weight = main_blocks_0_convnext_0_0_dwconv__conv_weight_to_fp16_quantized, x = input_9_cast_fp16)[name = string("h_3_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_5_cast_fp16 = mul(x = h_3_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_5_cast_fp16")];
            tensor<int32, [3]> input_11_perm_0 = const()[name = string("input_11_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_179_axes_0 = const()[name = string("op_179_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_0_convnext_0_0_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_0_convnext_0_0_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127744)))];
            tensor<fp16, [512]> main_blocks_0_convnext_0_0_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_0_0_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128832)))];
            fp16 var_138_to_fp16 = const()[name = string("op_138_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [2, ?, 512]> input_11_cast_fp16 = transpose(perm = input_11_perm_0, x = x_5_cast_fp16)[name = string("transpose_117")];
            tensor<fp16, [2, ?, 512]> var_179_cast_fp16 = layer_norm(axes = var_179_axes_0, beta = main_blocks_0_convnext_0_0_norm_norm_bias_to_fp16, epsilon = var_138_to_fp16, gamma = main_blocks_0_convnext_0_0_norm_norm_weight_to_fp16, x = input_11_cast_fp16)[name = string("op_179_cast_fp16")];
            tensor<int32, [3]> input_13_perm_0 = const()[name = string("input_13_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_5_pad_type_0 = const()[name = string("h_5_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_5_strides_0 = const()[name = string("h_5_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_5_pad_0 = const()[name = string("h_5_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_5_dilations_0 = const()[name = string("h_5_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_5_groups_0 = const()[name = string("h_5_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_0_convnext_0_0_pwconv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129920))), scale = tensor<fp16, [2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1178560))))[name = string("main_blocks_0_convnext_0_0_pwconv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> main_blocks_0_convnext_0_0_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_0_0_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1182720)))];
            tensor<fp16, [2, 512, ?]> input_13_cast_fp16 = transpose(perm = input_13_perm_0, x = var_179_cast_fp16)[name = string("transpose_116")];
            tensor<fp16, [2, 2048, ?]> h_5_cast_fp16 = conv(bias = main_blocks_0_convnext_0_0_pwconv1_bias_to_fp16, dilations = h_5_dilations_0, groups = h_5_groups_0, pad = h_5_pad_0, pad_type = h_5_pad_type_0, strides = h_5_strides_0, weight = main_blocks_0_convnext_0_0_pwconv1_weight_to_fp16_quantized, x = input_13_cast_fp16)[name = string("h_5_cast_fp16")];
            string input_15_mode_0 = const()[name = string("input_15_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_15_cast_fp16 = gelu(mode = input_15_mode_0, x = h_5_cast_fp16)[name = string("input_15_cast_fp16")];
            string h_7_pad_type_0 = const()[name = string("h_7_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_7_strides_0 = const()[name = string("h_7_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_7_pad_0 = const()[name = string("h_7_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_7_dilations_0 = const()[name = string("h_7_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_7_groups_0 = const()[name = string("h_7_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> op_196_weight_0_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1186880))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2235520))))[name = string("op_196_weight_0_to_fp16_quantized")];
            tensor<fp16, [512]> var_196_bias_0_to_fp16 = const()[name = string("op_196_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2236608)))];
            tensor<fp16, [2, 512, ?]> var_196_cast_fp16 = conv(bias = var_196_bias_0_to_fp16, dilations = h_7_dilations_0, groups = h_7_groups_0, pad = h_7_pad_0, pad_type = h_7_pad_type_0, strides = h_7_strides_0, weight = op_196_weight_0_to_fp16_quantized, x = input_15_cast_fp16)[name = string("op_196_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_1_cast_fp16 = add(x = input_7_cast_fp16, y = var_196_cast_fp16)[name = string("out_1_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_7_cast_fp16 = mul(x = out_1_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_7_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_17_cast_fp16 = mul(x = x_7_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("input_17_cast_fp16")];
            tensor<int32, [6]> input_19_pad_0 = const()[name = string("input_19_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_19_mode_0 = const()[name = string("input_19_mode_0"), val = string("replicate")];
            fp16 const_1_to_fp16 = const()[name = string("const_1_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_19_cast_fp16 = pad(constant_val = const_1_to_fp16, mode = input_19_mode_0, pad = input_19_pad_0, x = input_17_cast_fp16)[name = string("input_19_cast_fp16")];
            string h_9_pad_type_0 = const()[name = string("h_9_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_9_dilations_0 = const()[name = string("h_9_dilations_0"), val = tensor<int32, [1]>([2])];
            int32 h_9_groups_0 = const()[name = string("h_9_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_9_strides_0 = const()[name = string("h_9_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_9_pad_0 = const()[name = string("h_9_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [512, 1, 5]> main_blocks_0_convnext_0_1_dwconv__conv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2237696))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2240320))))[name = string("main_blocks_0_convnext_0_1_dwconv__conv_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_0_convnext_0_1_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_0_1_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2241408)))];
            tensor<fp16, [2, 512, ?]> h_9_cast_fp16 = conv(bias = main_blocks_0_convnext_0_1_dwconv__conv_bias_to_fp16, dilations = h_9_dilations_0, groups = h_9_groups_0, pad = h_9_pad_0, pad_type = h_9_pad_type_0, strides = h_9_strides_0, weight = main_blocks_0_convnext_0_1_dwconv__conv_weight_to_fp16_quantized, x = input_19_cast_fp16)[name = string("h_9_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_9_cast_fp16 = mul(x = h_9_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_9_cast_fp16")];
            tensor<int32, [3]> input_21_perm_0 = const()[name = string("input_21_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_221_axes_0 = const()[name = string("op_221_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_0_convnext_0_1_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_0_convnext_0_1_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2242496)))];
            tensor<fp16, [512]> main_blocks_0_convnext_0_1_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_0_1_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2243584)))];
            tensor<fp16, [2, ?, 512]> input_21_cast_fp16 = transpose(perm = input_21_perm_0, x = x_9_cast_fp16)[name = string("transpose_115")];
            tensor<fp16, [2, ?, 512]> var_221_cast_fp16 = layer_norm(axes = var_221_axes_0, beta = main_blocks_0_convnext_0_1_norm_norm_bias_to_fp16, epsilon = var_138_to_fp16, gamma = main_blocks_0_convnext_0_1_norm_norm_weight_to_fp16, x = input_21_cast_fp16)[name = string("op_221_cast_fp16")];
            tensor<int32, [3]> input_23_perm_0 = const()[name = string("input_23_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_11_pad_type_0 = const()[name = string("h_11_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_11_strides_0 = const()[name = string("h_11_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_11_pad_0 = const()[name = string("h_11_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_11_dilations_0 = const()[name = string("h_11_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_11_groups_0 = const()[name = string("h_11_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_0_convnext_0_1_pwconv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2244672))), scale = tensor<fp16, [2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3293312))))[name = string("main_blocks_0_convnext_0_1_pwconv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> main_blocks_0_convnext_0_1_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_0_1_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3297472)))];
            tensor<fp16, [2, 512, ?]> input_23_cast_fp16 = transpose(perm = input_23_perm_0, x = var_221_cast_fp16)[name = string("transpose_114")];
            tensor<fp16, [2, 2048, ?]> h_11_cast_fp16 = conv(bias = main_blocks_0_convnext_0_1_pwconv1_bias_to_fp16, dilations = h_11_dilations_0, groups = h_11_groups_0, pad = h_11_pad_0, pad_type = h_11_pad_type_0, strides = h_11_strides_0, weight = main_blocks_0_convnext_0_1_pwconv1_weight_to_fp16_quantized, x = input_23_cast_fp16)[name = string("h_11_cast_fp16")];
            string input_25_mode_0 = const()[name = string("input_25_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_25_cast_fp16 = gelu(mode = input_25_mode_0, x = h_11_cast_fp16)[name = string("input_25_cast_fp16")];
            string h_13_pad_type_0 = const()[name = string("h_13_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_13_strides_0 = const()[name = string("h_13_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_13_pad_0 = const()[name = string("h_13_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_13_dilations_0 = const()[name = string("h_13_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_13_groups_0 = const()[name = string("h_13_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> op_238_weight_0_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3301632))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4350272))))[name = string("op_238_weight_0_to_fp16_quantized")];
            tensor<fp16, [512]> var_238_bias_0_to_fp16 = const()[name = string("op_238_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4351360)))];
            tensor<fp16, [2, 512, ?]> var_238_cast_fp16 = conv(bias = var_238_bias_0_to_fp16, dilations = h_13_dilations_0, groups = h_13_groups_0, pad = h_13_pad_0, pad_type = h_13_pad_type_0, strides = h_13_strides_0, weight = op_238_weight_0_to_fp16_quantized, x = input_25_cast_fp16)[name = string("op_238_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_3_cast_fp16 = add(x = input_17_cast_fp16, y = var_238_cast_fp16)[name = string("out_3_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_11_cast_fp16 = mul(x = out_3_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_11_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_27_cast_fp16 = mul(x = x_11_cast_fp16, y = latent_mask_b_cast_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, 8, 8])];
            string input_29_mode_0 = const()[name = string("input_29_mode_0"), val = string("replicate")];
            fp16 const_2_to_fp16 = const()[name = string("const_2_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_29_cast_fp16 = pad(constant_val = const_2_to_fp16, mode = input_29_mode_0, pad = input_29_pad_0, x = input_27_cast_fp16)[name = string("input_29_cast_fp16")];
            string h_15_pad_type_0 = const()[name = string("h_15_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_15_dilations_0 = const()[name = string("h_15_dilations_0"), val = tensor<int32, [1]>([4])];
            int32 h_15_groups_0 = const()[name = string("h_15_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_15_strides_0 = const()[name = string("h_15_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_15_pad_0 = const()[name = string("h_15_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [512, 1, 5]> main_blocks_0_convnext_0_2_dwconv__conv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4352448))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4355072))))[name = string("main_blocks_0_convnext_0_2_dwconv__conv_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_0_convnext_0_2_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_0_2_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4356160)))];
            tensor<fp16, [2, 512, ?]> h_15_cast_fp16 = conv(bias = main_blocks_0_convnext_0_2_dwconv__conv_bias_to_fp16, dilations = h_15_dilations_0, groups = h_15_groups_0, pad = h_15_pad_0, pad_type = h_15_pad_type_0, strides = h_15_strides_0, weight = main_blocks_0_convnext_0_2_dwconv__conv_weight_to_fp16_quantized, x = input_29_cast_fp16)[name = string("h_15_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_13_cast_fp16 = mul(x = h_15_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_13_cast_fp16")];
            tensor<int32, [3]> input_31_perm_0 = const()[name = string("input_31_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_263_axes_0 = const()[name = string("op_263_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_0_convnext_0_2_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_0_convnext_0_2_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4357248)))];
            tensor<fp16, [512]> main_blocks_0_convnext_0_2_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_0_2_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4358336)))];
            tensor<fp16, [2, ?, 512]> input_31_cast_fp16 = transpose(perm = input_31_perm_0, x = x_13_cast_fp16)[name = string("transpose_113")];
            tensor<fp16, [2, ?, 512]> var_263_cast_fp16 = layer_norm(axes = var_263_axes_0, beta = main_blocks_0_convnext_0_2_norm_norm_bias_to_fp16, epsilon = var_138_to_fp16, gamma = main_blocks_0_convnext_0_2_norm_norm_weight_to_fp16, x = input_31_cast_fp16)[name = string("op_263_cast_fp16")];
            tensor<int32, [3]> input_33_perm_0 = const()[name = string("input_33_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_17_pad_type_0 = const()[name = string("h_17_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_17_strides_0 = const()[name = string("h_17_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_17_pad_0 = const()[name = string("h_17_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_17_dilations_0 = const()[name = string("h_17_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_17_groups_0 = const()[name = string("h_17_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_0_convnext_0_2_pwconv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4359424))), scale = tensor<fp16, [2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5408064))))[name = string("main_blocks_0_convnext_0_2_pwconv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> main_blocks_0_convnext_0_2_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_0_2_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5412224)))];
            tensor<fp16, [2, 512, ?]> input_33_cast_fp16 = transpose(perm = input_33_perm_0, x = var_263_cast_fp16)[name = string("transpose_112")];
            tensor<fp16, [2, 2048, ?]> h_17_cast_fp16 = conv(bias = main_blocks_0_convnext_0_2_pwconv1_bias_to_fp16, dilations = h_17_dilations_0, groups = h_17_groups_0, pad = h_17_pad_0, pad_type = h_17_pad_type_0, strides = h_17_strides_0, weight = main_blocks_0_convnext_0_2_pwconv1_weight_to_fp16_quantized, x = input_33_cast_fp16)[name = string("h_17_cast_fp16")];
            string input_35_mode_0 = const()[name = string("input_35_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_35_cast_fp16 = gelu(mode = input_35_mode_0, x = h_17_cast_fp16)[name = string("input_35_cast_fp16")];
            string h_19_pad_type_0 = const()[name = string("h_19_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_19_strides_0 = const()[name = string("h_19_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_19_pad_0 = const()[name = string("h_19_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_19_dilations_0 = const()[name = string("h_19_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_19_groups_0 = const()[name = string("h_19_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> op_280_weight_0_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5416384))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6465024))))[name = string("op_280_weight_0_to_fp16_quantized")];
            tensor<fp16, [512]> var_280_bias_0_to_fp16 = const()[name = string("op_280_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6466112)))];
            tensor<fp16, [2, 512, ?]> var_280_cast_fp16 = conv(bias = var_280_bias_0_to_fp16, dilations = h_19_dilations_0, groups = h_19_groups_0, pad = h_19_pad_0, pad_type = h_19_pad_type_0, strides = h_19_strides_0, weight = op_280_weight_0_to_fp16_quantized, x = input_35_cast_fp16)[name = string("op_280_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_5_cast_fp16 = add(x = input_27_cast_fp16, y = var_280_cast_fp16)[name = string("out_5_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_15_cast_fp16 = mul(x = out_5_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_15_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_37_cast_fp16 = mul(x = x_15_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("input_37_cast_fp16")];
            tensor<int32, [6]> input_39_pad_0 = const()[name = string("input_39_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 16, 16])];
            string input_39_mode_0 = const()[name = string("input_39_mode_0"), val = string("replicate")];
            fp16 const_3_to_fp16 = const()[name = string("const_3_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_39_cast_fp16 = pad(constant_val = const_3_to_fp16, mode = input_39_mode_0, pad = input_39_pad_0, x = input_37_cast_fp16)[name = string("input_39_cast_fp16")];
            string h_21_pad_type_0 = const()[name = string("h_21_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_21_dilations_0 = const()[name = string("h_21_dilations_0"), val = tensor<int32, [1]>([8])];
            int32 h_21_groups_0 = const()[name = string("h_21_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_21_strides_0 = const()[name = string("h_21_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_21_pad_0 = const()[name = string("h_21_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [512, 1, 5]> main_blocks_0_convnext_0_3_dwconv__conv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6467200))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6469824))))[name = string("main_blocks_0_convnext_0_3_dwconv__conv_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_0_convnext_0_3_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_0_3_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6470912)))];
            tensor<fp16, [2, 512, ?]> h_21_cast_fp16 = conv(bias = main_blocks_0_convnext_0_3_dwconv__conv_bias_to_fp16, dilations = h_21_dilations_0, groups = h_21_groups_0, pad = h_21_pad_0, pad_type = h_21_pad_type_0, strides = h_21_strides_0, weight = main_blocks_0_convnext_0_3_dwconv__conv_weight_to_fp16_quantized, x = input_39_cast_fp16)[name = string("h_21_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_17_cast_fp16 = mul(x = h_21_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_17_cast_fp16")];
            tensor<int32, [3]> input_41_perm_0 = const()[name = string("input_41_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_305_axes_0 = const()[name = string("op_305_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_0_convnext_0_3_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_0_convnext_0_3_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6472000)))];
            tensor<fp16, [512]> main_blocks_0_convnext_0_3_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_0_3_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6473088)))];
            tensor<fp16, [2, ?, 512]> input_41_cast_fp16 = transpose(perm = input_41_perm_0, x = x_17_cast_fp16)[name = string("transpose_111")];
            tensor<fp16, [2, ?, 512]> var_305_cast_fp16 = layer_norm(axes = var_305_axes_0, beta = main_blocks_0_convnext_0_3_norm_norm_bias_to_fp16, epsilon = var_138_to_fp16, gamma = main_blocks_0_convnext_0_3_norm_norm_weight_to_fp16, x = input_41_cast_fp16)[name = string("op_305_cast_fp16")];
            tensor<int32, [3]> input_43_perm_0 = const()[name = string("input_43_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_23_pad_type_0 = const()[name = string("h_23_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_23_strides_0 = const()[name = string("h_23_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_23_pad_0 = const()[name = string("h_23_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_23_dilations_0 = const()[name = string("h_23_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_23_groups_0 = const()[name = string("h_23_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_0_convnext_0_3_pwconv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6474176))), scale = tensor<fp16, [2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7522816))))[name = string("main_blocks_0_convnext_0_3_pwconv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> main_blocks_0_convnext_0_3_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_0_3_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7526976)))];
            tensor<fp16, [2, 512, ?]> input_43_cast_fp16 = transpose(perm = input_43_perm_0, x = var_305_cast_fp16)[name = string("transpose_110")];
            tensor<fp16, [2, 2048, ?]> h_23_cast_fp16 = conv(bias = main_blocks_0_convnext_0_3_pwconv1_bias_to_fp16, dilations = h_23_dilations_0, groups = h_23_groups_0, pad = h_23_pad_0, pad_type = h_23_pad_type_0, strides = h_23_strides_0, weight = main_blocks_0_convnext_0_3_pwconv1_weight_to_fp16_quantized, x = input_43_cast_fp16)[name = string("h_23_cast_fp16")];
            string input_45_mode_0 = const()[name = string("input_45_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_45_cast_fp16 = gelu(mode = input_45_mode_0, x = h_23_cast_fp16)[name = string("input_45_cast_fp16")];
            string h_25_pad_type_0 = const()[name = string("h_25_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_25_strides_0 = const()[name = string("h_25_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_25_pad_0 = const()[name = string("h_25_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_25_dilations_0 = const()[name = string("h_25_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_25_groups_0 = const()[name = string("h_25_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> op_322_weight_0_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7531136))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8579776))))[name = string("op_322_weight_0_to_fp16_quantized")];
            tensor<fp16, [512]> var_322_bias_0_to_fp16 = const()[name = string("op_322_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8580864)))];
            tensor<fp16, [2, 512, ?]> var_322_cast_fp16 = conv(bias = var_322_bias_0_to_fp16, dilations = h_25_dilations_0, groups = h_25_groups_0, pad = h_25_pad_0, pad_type = h_25_pad_type_0, strides = h_25_strides_0, weight = op_322_weight_0_to_fp16_quantized, x = input_45_cast_fp16)[name = string("op_322_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_7_cast_fp16 = add(x = input_37_cast_fp16, y = var_322_cast_fp16)[name = string("out_7_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_19_cast_fp16 = mul(x = out_7_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_19_cast_fp16")];
            tensor<int32, [1]> input_47_axes_0 = const()[name = string("input_47_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2, 64]> input_47_cast_fp16 = squeeze(axes = input_47_axes_0, x = time_emb_cast_fp16)[name = string("input_47_cast_fp16")];
            tensor<fp16, [512, 64]> main_blocks_0_time_cond_linear_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8581952))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8614784))))[name = string("main_blocks_0_time_cond_linear_linear_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_0_time_cond_linear_linear_bias_to_fp16 = const()[name = string("main_blocks_0_time_cond_linear_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8615872)))];
            tensor<fp16, [2, 512]> linear_2_cast_fp16 = linear(bias = main_blocks_0_time_cond_linear_linear_bias_to_fp16, weight = main_blocks_0_time_cond_linear_linear_weight_to_fp16_quantized, x = input_47_cast_fp16)[name = string("linear_2_cast_fp16")];
            tensor<int32, [1]> t_5_axes_0 = const()[name = string("t_5_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2, 512, 1]> t_5_cast_fp16 = expand_dims(axes = t_5_axes_0, x = linear_2_cast_fp16)[name = string("t_5_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_332_cast_fp16 = add(x = x_19_cast_fp16, y = t_5_cast_fp16)[name = string("op_332_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_21_cast_fp16 = mul(x = var_332_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_21_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_49_cast_fp16 = mul(x = x_21_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("input_49_cast_fp16")];
            tensor<int32, [6]> input_51_pad_0 = const()[name = string("input_51_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_51_mode_0 = const()[name = string("input_51_mode_0"), val = string("replicate")];
            fp16 const_4_to_fp16 = const()[name = string("const_4_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_51_cast_fp16 = pad(constant_val = const_4_to_fp16, mode = input_51_mode_0, pad = input_51_pad_0, x = input_49_cast_fp16)[name = string("input_51_cast_fp16")];
            string h_27_pad_type_0 = const()[name = string("h_27_pad_type_0"), val = string("valid")];
            int32 h_27_groups_0 = const()[name = string("h_27_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_27_strides_0 = const()[name = string("h_27_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_27_pad_0 = const()[name = string("h_27_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_27_dilations_0 = const()[name = string("h_27_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 5]> main_blocks_0_convnext_1_0_dwconv__conv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8616960))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8619584))))[name = string("main_blocks_0_convnext_1_0_dwconv__conv_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_0_convnext_1_0_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_1_0_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8620672)))];
            tensor<fp16, [2, 512, ?]> h_27_cast_fp16 = conv(bias = main_blocks_0_convnext_1_0_dwconv__conv_bias_to_fp16, dilations = h_27_dilations_0, groups = h_27_groups_0, pad = h_27_pad_0, pad_type = h_27_pad_type_0, strides = h_27_strides_0, weight = main_blocks_0_convnext_1_0_dwconv__conv_weight_to_fp16_quantized, x = input_51_cast_fp16)[name = string("h_27_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_23_cast_fp16 = mul(x = h_27_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_23_cast_fp16")];
            tensor<int32, [3]> input_53_perm_0 = const()[name = string("input_53_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_356_axes_0 = const()[name = string("op_356_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_0_convnext_1_0_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_0_convnext_1_0_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8621760)))];
            tensor<fp16, [512]> main_blocks_0_convnext_1_0_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_1_0_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8622848)))];
            tensor<fp16, [2, ?, 512]> input_53_cast_fp16 = transpose(perm = input_53_perm_0, x = x_23_cast_fp16)[name = string("transpose_109")];
            tensor<fp16, [2, ?, 512]> var_356_cast_fp16 = layer_norm(axes = var_356_axes_0, beta = main_blocks_0_convnext_1_0_norm_norm_bias_to_fp16, epsilon = var_138_to_fp16, gamma = main_blocks_0_convnext_1_0_norm_norm_weight_to_fp16, x = input_53_cast_fp16)[name = string("op_356_cast_fp16")];
            tensor<int32, [3]> input_55_perm_0 = const()[name = string("input_55_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_29_pad_type_0 = const()[name = string("h_29_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_29_strides_0 = const()[name = string("h_29_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_29_pad_0 = const()[name = string("h_29_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_29_dilations_0 = const()[name = string("h_29_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_29_groups_0 = const()[name = string("h_29_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_0_convnext_1_0_pwconv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8623936))), scale = tensor<fp16, [2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9672576))))[name = string("main_blocks_0_convnext_1_0_pwconv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> main_blocks_0_convnext_1_0_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_1_0_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9676736)))];
            tensor<fp16, [2, 512, ?]> input_55_cast_fp16 = transpose(perm = input_55_perm_0, x = var_356_cast_fp16)[name = string("transpose_108")];
            tensor<fp16, [2, 2048, ?]> h_29_cast_fp16 = conv(bias = main_blocks_0_convnext_1_0_pwconv1_bias_to_fp16, dilations = h_29_dilations_0, groups = h_29_groups_0, pad = h_29_pad_0, pad_type = h_29_pad_type_0, strides = h_29_strides_0, weight = main_blocks_0_convnext_1_0_pwconv1_weight_to_fp16_quantized, x = input_55_cast_fp16)[name = string("h_29_cast_fp16")];
            string input_57_mode_0 = const()[name = string("input_57_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_57_cast_fp16 = gelu(mode = input_57_mode_0, x = h_29_cast_fp16)[name = string("input_57_cast_fp16")];
            string h_31_pad_type_0 = const()[name = string("h_31_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_31_strides_0 = const()[name = string("h_31_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_31_pad_0 = const()[name = string("h_31_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_31_dilations_0 = const()[name = string("h_31_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_31_groups_0 = const()[name = string("h_31_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> op_373_weight_0_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9680896))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10729536))))[name = string("op_373_weight_0_to_fp16_quantized")];
            tensor<fp16, [512]> var_373_bias_0_to_fp16 = const()[name = string("op_373_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10730624)))];
            tensor<fp16, [2, 512, ?]> var_373_cast_fp16 = conv(bias = var_373_bias_0_to_fp16, dilations = h_31_dilations_0, groups = h_31_groups_0, pad = h_31_pad_0, pad_type = h_31_pad_type_0, strides = h_31_strides_0, weight = op_373_weight_0_to_fp16_quantized, x = input_57_cast_fp16)[name = string("op_373_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_9_cast_fp16 = add(x = input_49_cast_fp16, y = var_373_cast_fp16)[name = string("out_9_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_25_cast_fp16 = mul(x = out_9_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_25_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_m_1_cast_fp16 = mul(x = x_25_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_m_1_cast_fp16")];
            tensor<int32, [3]> var_383_shape_cast_fp16 = shape(x = x_25_cast_fp16)[name = string("op_383_shape_cast_fp16")];
            int32 gather_4_axis_0 = const()[name = string("gather_4_axis_0"), val = int32(0)];
            int32 gather_4_batch_dims_0 = const()[name = string("gather_4_batch_dims_0"), val = int32(0)];
            bool gather_4_validate_indices_0 = const()[name = string("gather_4_validate_indices_0"), val = bool(false)];
            string var_383_shape_cast_fp16_to_int16_dtype_0 = const()[name = string("op_383_shape_cast_fp16_to_int16_dtype_0"), val = string("int16")];
            uint16 gather_4_indices_0_to_uint16 = const()[name = string("gather_4_indices_0_to_uint16"), val = uint16(2)];
            tensor<int16, [3]> var_383_shape_cast_fp16_to_int16 = cast(dtype = var_383_shape_cast_fp16_to_int16_dtype_0, x = var_383_shape_cast_fp16)[name = string("cast_10")];
            int16 gather_4_cast_uint16 = gather(axis = gather_4_axis_0, batch_dims = gather_4_batch_dims_0, indices = gather_4_indices_0_to_uint16, validate_indices = gather_4_validate_indices_0, x = var_383_shape_cast_fp16_to_int16)[name = string("gather_4_cast_uint16")];
            string gather_4_cast_uint16_to_int32_dtype_0 = const()[name = string("gather_4_cast_uint16_to_int32_dtype_0"), val = string("int32")];
            tensor<int32, [3]> var_385_shape_cast_fp16 = shape(x = text_emb_cast_fp16)[name = string("op_385_shape_cast_fp16")];
            int32 gather_5_axis_0 = const()[name = string("gather_5_axis_0"), val = int32(0)];
            int32 gather_5_batch_dims_0 = const()[name = string("gather_5_batch_dims_0"), val = int32(0)];
            bool gather_5_validate_indices_0 = const()[name = string("gather_5_validate_indices_0"), val = bool(false)];
            string var_385_shape_cast_fp16_to_uint16_dtype_0 = const()[name = string("op_385_shape_cast_fp16_to_uint16_dtype_0"), val = string("uint16")];
            uint16 gather_5_indices_0_to_uint16 = const()[name = string("gather_5_indices_0_to_uint16"), val = uint16(2)];
            tensor<uint16, [3]> var_385_shape_cast_fp16_to_uint16 = cast(dtype = var_385_shape_cast_fp16_to_uint16_dtype_0, x = var_385_shape_cast_fp16)[name = string("cast_9")];
            uint16 gather_5_cast_uint16 = gather(axis = gather_5_axis_0, batch_dims = gather_5_batch_dims_0, indices = gather_5_indices_0_to_uint16, validate_indices = gather_5_validate_indices_0, x = var_385_shape_cast_fp16_to_uint16)[name = string("gather_5_cast_uint16")];
            string gather_5_cast_uint16_to_int32_dtype_0 = const()[name = string("gather_5_cast_uint16_to_int32_dtype_0"), val = string("int32")];
            tensor<int32, [3]> input_59_perm_0 = const()[name = string("input_59_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [512, 512]> main_blocks_0_text_attn_W_query_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10731712))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10993920))))[name = string("main_blocks_0_text_attn_W_query_linear_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_0_text_attn_W_query_linear_bias_to_fp16 = const()[name = string("main_blocks_0_text_attn_W_query_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10995008)))];
            tensor<fp16, [2, ?, 512]> input_59_cast_fp16 = transpose(perm = input_59_perm_0, x = x_25_cast_fp16)[name = string("transpose_107")];
            tensor<fp16, [2, ?, 512]> linear_3_cast_fp16 = linear(bias = main_blocks_0_text_attn_W_query_linear_bias_to_fp16, weight = main_blocks_0_text_attn_W_query_linear_weight_to_fp16_quantized, x = input_59_cast_fp16)[name = string("linear_3_cast_fp16")];
            tensor<int32, [4]> concat_4x = const()[name = string("concat_4x"), val = tensor<int32, [4]>([2, -1, 8, 64])];
            tensor<fp16, [2, ?, 8, 64]> var_392_cast_fp16 = reshape(shape = concat_4x, x = linear_3_cast_fp16)[name = string("op_392_cast_fp16")];
            tensor<int32, [4]> x_27_perm_0 = const()[name = string("x_27_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> input_61_perm_0 = const()[name = string("input_61_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [512, 256]> main_blocks_0_text_attn_W_key_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10996096))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11127232))))[name = string("main_blocks_0_text_attn_W_key_linear_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_0_text_attn_W_key_linear_bias_to_fp16 = const()[name = string("main_blocks_0_text_attn_W_key_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11128320)))];
            tensor<fp16, [?, ?, 256]> input_61_cast_fp16 = transpose(perm = input_61_perm_0, x = text_emb_cast_fp16)[name = string("transpose_106")];
            tensor<fp16, [?, ?, 512]> linear_4_cast_fp16 = linear(bias = main_blocks_0_text_attn_W_key_linear_bias_to_fp16, weight = main_blocks_0_text_attn_W_key_linear_weight_to_fp16_quantized, x = input_61_cast_fp16)[name = string("linear_4_cast_fp16")];
            tensor<int32, [4]> concat_5x = const()[name = string("concat_5x"), val = tensor<int32, [4]>([2, -1, 8, 64])];
            tensor<fp16, [2, ?, 8, 64]> var_400_cast_fp16 = reshape(shape = concat_5x, x = linear_4_cast_fp16)[name = string("op_400_cast_fp16")];
            tensor<int32, [4]> x_29_perm_0 = const()[name = string("x_29_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [512, 256]> main_blocks_0_text_attn_W_value_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11129408))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11260544))))[name = string("main_blocks_0_text_attn_W_value_linear_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_0_text_attn_W_value_linear_bias_to_fp16 = const()[name = string("main_blocks_0_text_attn_W_value_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11261632)))];
            tensor<fp16, [?, ?, 512]> linear_5_cast_fp16 = linear(bias = main_blocks_0_text_attn_W_value_linear_bias_to_fp16, weight = main_blocks_0_text_attn_W_value_linear_weight_to_fp16_quantized, x = input_61_cast_fp16)[name = string("linear_5_cast_fp16")];
            tensor<int32, [4]> concat_6x = const()[name = string("concat_6x"), val = tensor<int32, [4]>([2, -1, 8, 64])];
            tensor<fp16, [2, ?, 8, 64]> var_408_cast_fp16 = reshape(shape = concat_6x, x = linear_5_cast_fp16)[name = string("op_408_cast_fp16")];
            tensor<int32, [4]> v_1_perm_0 = const()[name = string("v_1_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            int32 concat_7_values0_0 = const()[name = string("concat_7_values0_0"), val = int32(1)];
            int32 concat_7_values2_0 = const()[name = string("concat_7_values2_0"), val = int32(1)];
            int32 concat_7_axis_0 = const()[name = string("concat_7_axis_0"), val = int32(0)];
            bool concat_7_interleave_0 = const()[name = string("concat_7_interleave_0"), val = bool(false)];
            int32 gather_4_cast_uint16_to_int32 = cast(dtype = gather_4_cast_uint16_to_int32_dtype_0, x = gather_4_cast_uint16)[name = string("cast_8")];
            tensor<int32, [3]> concat_7 = concat(axis = concat_7_axis_0, interleave = concat_7_interleave_0, values = (concat_7_values0_0, gather_4_cast_uint16_to_int32, concat_7_values2_0))[name = string("concat_7")];
            tensor<int32, [3]> var_411_begin_0 = const()[name = string("op_411_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<bool, [3]> var_411_end_mask_0 = const()[name = string("op_411_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 1000, 1]> main_blocks_0_text_attn_increments_to_fp16 = const()[name = string("main_blocks_0_text_attn_increments_to_fp16"), val = tensor<fp16, [1, 1000, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11262720)))];
            tensor<fp16, [1, ?, 1]> var_411_cast_fp16 = slice_by_index(begin = var_411_begin_0, end = concat_7, end_mask = var_411_end_mask_0, x = main_blocks_0_text_attn_increments_to_fp16)[name = string("op_411_cast_fp16")];
            tensor<int32, [3]> concat_8 = const()[name = string("concat_8"), val = tensor<int32, [3]>([2, -1, -1])];
            tensor<int32, [3]> shape_1_cast_fp16 = shape(x = var_411_cast_fp16)[name = string("shape_1_cast_fp16")];
            tensor<bool, [3]> equal_1 = const()[name = string("equal_1"), val = tensor<bool, [3]>([false, true, true])];
            tensor<int32, [3]> select_1 = select(a = shape_1_cast_fp16, b = concat_8, cond = equal_1)[name = string("select_1")];
            tensor<int32, [3]> real_div_1 = real_div(x = select_1, y = shape_1_cast_fp16)[name = string("real_div_1")];
            tensor<fp16, [?, ?, ?]> var_414_cast_fp16 = tile(reps = real_div_1, x = var_411_cast_fp16)[name = string("op_414_cast_fp16")];
            int32 concat_9_values0_0 = const()[name = string("concat_9_values0_0"), val = int32(1)];
            int32 concat_9_values2_0 = const()[name = string("concat_9_values2_0"), val = int32(1)];
            int32 concat_9_axis_0 = const()[name = string("concat_9_axis_0"), val = int32(0)];
            bool concat_9_interleave_0 = const()[name = string("concat_9_interleave_0"), val = bool(false)];
            int32 gather_5_cast_uint16_to_int32 = cast(dtype = gather_5_cast_uint16_to_int32_dtype_0, x = gather_5_cast_uint16)[name = string("cast_7")];
            tensor<int32, [3]> concat_9 = concat(axis = concat_9_axis_0, interleave = concat_9_interleave_0, values = (concat_9_values0_0, gather_5_cast_uint16_to_int32, concat_9_values2_0))[name = string("concat_9")];
            tensor<int32, [3]> var_417_begin_0 = const()[name = string("op_417_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<bool, [3]> var_417_end_mask_0 = const()[name = string("op_417_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, ?, 1]> var_417_cast_fp16 = slice_by_index(begin = var_417_begin_0, end = concat_9, end_mask = var_417_end_mask_0, x = main_blocks_0_text_attn_increments_to_fp16)[name = string("op_417_cast_fp16")];
            tensor<int32, [3]> concat_10 = const()[name = string("concat_10"), val = tensor<int32, [3]>([2, -1, -1])];
            tensor<int32, [3]> shape_2_cast_fp16 = shape(x = var_417_cast_fp16)[name = string("shape_2_cast_fp16")];
            tensor<bool, [3]> equal_2 = const()[name = string("equal_2"), val = tensor<bool, [3]>([false, true, true])];
            tensor<int32, [3]> select_2 = select(a = shape_2_cast_fp16, b = concat_10, cond = equal_2)[name = string("select_2")];
            tensor<int32, [3]> real_div_2 = real_div(x = select_2, y = shape_2_cast_fp16)[name = string("real_div_2")];
            tensor<fp16, [?, ?, ?]> var_420_cast_fp16 = tile(reps = real_div_2, x = var_417_cast_fp16)[name = string("op_420_cast_fp16")];
            tensor<int32, [2]> divisor_1_axes_0 = const()[name = string("divisor_1_axes_0"), val = tensor<int32, [2]>([1, 2])];
            bool divisor_1_keep_dims_0 = const()[name = string("divisor_1_keep_dims_0"), val = bool(false)];
            tensor<fp16, [2]> divisor_1_cast_fp16 = reduce_sum(axes = divisor_1_axes_0, keep_dims = divisor_1_keep_dims_0, x = latent_mask_b_cast_fp16)[name = string("divisor_1_cast_fp16")];
            tensor<int32, [2]> divisor_3_axes_0 = const()[name = string("divisor_3_axes_0"), val = tensor<int32, [2]>([1, 2])];
            bool divisor_3_keep_dims_0 = const()[name = string("divisor_3_keep_dims_0"), val = bool(false)];
            tensor<fp16, [2]> divisor_3_cast_fp16 = reduce_sum(axes = divisor_3_axes_0, keep_dims = divisor_3_keep_dims_0, x = text_mask_cast_fp16)[name = string("divisor_3_cast_fp16")];
            tensor<int32, [3]> var_426 = const()[name = string("op_426"), val = tensor<int32, [3]>([-1, 1, 1])];
            tensor<fp16, [2, 1, 1]> var_427_cast_fp16 = reshape(shape = var_426, x = divisor_1_cast_fp16)[name = string("op_427_cast_fp16")];
            tensor<fp16, [2, ?, ?]> scaled_1_cast_fp16 = real_div(x = var_414_cast_fp16, y = var_427_cast_fp16)[name = string("scaled_1_cast_fp16")];
            tensor<fp16, [1, 1, 32]> main_blocks_0_text_attn_theta_to_fp16 = const()[name = string("main_blocks_0_text_attn_theta_to_fp16"), val = tensor<fp16, [1, 1, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11264832)))];
            tensor<fp16, [2, ?, 32]> angles_1_cast_fp16 = mul(x = scaled_1_cast_fp16, y = main_blocks_0_text_attn_theta_to_fp16)[name = string("angles_1_cast_fp16")];
            tensor<fp16, [2, ?, 32]> var_430_cast_fp16 = cos(x = angles_1_cast_fp16)[name = string("op_430_cast_fp16")];
            tensor<int32, [1]> cos_1_axes_0 = const()[name = string("cos_1_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, ?, 32]> cos_1_cast_fp16 = expand_dims(axes = cos_1_axes_0, x = var_430_cast_fp16)[name = string("cos_1_cast_fp16")];
            tensor<fp16, [2, ?, 32]> var_432_cast_fp16 = sin(x = angles_1_cast_fp16)[name = string("op_432_cast_fp16")];
            tensor<int32, [1]> sin_1_axes_0 = const()[name = string("sin_1_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, ?, 32]> sin_1_cast_fp16 = expand_dims(axes = sin_1_axes_0, x = var_432_cast_fp16)[name = string("sin_1_cast_fp16")];
            tensor<int32, [4]> x_a_1_begin_0 = const()[name = string("x_a_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x_a_1_end_0 = const()[name = string("x_a_1_end_0"), val = tensor<int32, [4]>([2, 8, 0, 32])];
            tensor<bool, [4]> x_a_1_end_mask_0 = const()[name = string("x_a_1_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 8, ?, 64]> x_27_cast_fp16 = transpose(perm = x_27_perm_0, x = var_392_cast_fp16)[name = string("transpose_105")];
            tensor<fp16, [2, 8, ?, 32]> x_a_1_cast_fp16 = slice_by_index(begin = x_a_1_begin_0, end = x_a_1_end_0, end_mask = x_a_1_end_mask_0, x = x_27_cast_fp16)[name = string("x_a_1_cast_fp16")];
            tensor<int32, [4]> x_b_1_begin_0 = const()[name = string("x_b_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x_b_1_end_0 = const()[name = string("x_b_1_end_0"), val = tensor<int32, [4]>([2, 8, 0, 64])];
            tensor<bool, [4]> x_b_1_end_mask_0 = const()[name = string("x_b_1_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [2, 8, ?, 32]> x_b_1_cast_fp16 = slice_by_index(begin = x_b_1_begin_0, end = x_b_1_end_0, end_mask = x_b_1_end_mask_0, x = x_27_cast_fp16)[name = string("x_b_1_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_436_cast_fp16 = mul(x = x_a_1_cast_fp16, y = cos_1_cast_fp16)[name = string("op_436_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_437_cast_fp16 = mul(x = x_b_1_cast_fp16, y = sin_1_cast_fp16)[name = string("op_437_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_a_1_cast_fp16 = sub(x = var_436_cast_fp16, y = var_437_cast_fp16)[name = string("rot_a_1_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_439_cast_fp16 = mul(x = x_a_1_cast_fp16, y = sin_1_cast_fp16)[name = string("op_439_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_440_cast_fp16 = mul(x = x_b_1_cast_fp16, y = cos_1_cast_fp16)[name = string("op_440_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_b_1_cast_fp16 = add(x = var_439_cast_fp16, y = var_440_cast_fp16)[name = string("rot_b_1_cast_fp16")];
            bool q_1_interleave_0 = const()[name = string("q_1_interleave_0"), val = bool(false)];
            tensor<fp16, [2, 8, ?, 64]> q_1_cast_fp16 = concat(axis = var_126, interleave = q_1_interleave_0, values = (rot_a_1_cast_fp16, rot_b_1_cast_fp16))[name = string("q_1_cast_fp16")];
            tensor<int32, [3]> var_444 = const()[name = string("op_444"), val = tensor<int32, [3]>([-1, 1, 1])];
            tensor<fp16, [2, 1, 1]> var_445_cast_fp16 = reshape(shape = var_444, x = divisor_3_cast_fp16)[name = string("op_445_cast_fp16")];
            tensor<fp16, [2, ?, ?]> scaled_3_cast_fp16 = real_div(x = var_420_cast_fp16, y = var_445_cast_fp16)[name = string("scaled_3_cast_fp16")];
            tensor<fp16, [2, ?, 32]> angles_3_cast_fp16 = mul(x = scaled_3_cast_fp16, y = main_blocks_0_text_attn_theta_to_fp16)[name = string("angles_3_cast_fp16")];
            tensor<fp16, [2, ?, 32]> var_448_cast_fp16 = cos(x = angles_3_cast_fp16)[name = string("op_448_cast_fp16")];
            tensor<int32, [1]> cos_3_axes_0 = const()[name = string("cos_3_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, ?, 32]> cos_3_cast_fp16 = expand_dims(axes = cos_3_axes_0, x = var_448_cast_fp16)[name = string("cos_3_cast_fp16")];
            tensor<fp16, [2, ?, 32]> var_450_cast_fp16 = sin(x = angles_3_cast_fp16)[name = string("op_450_cast_fp16")];
            tensor<int32, [1]> sin_3_axes_0 = const()[name = string("sin_3_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, ?, 32]> sin_3_cast_fp16 = expand_dims(axes = sin_3_axes_0, x = var_450_cast_fp16)[name = string("sin_3_cast_fp16")];
            tensor<int32, [4]> x_a_3_begin_0 = const()[name = string("x_a_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x_a_3_end_0 = const()[name = string("x_a_3_end_0"), val = tensor<int32, [4]>([2, 8, 0, 32])];
            tensor<bool, [4]> x_a_3_end_mask_0 = const()[name = string("x_a_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 8, ?, 64]> x_29_cast_fp16 = transpose(perm = x_29_perm_0, x = var_400_cast_fp16)[name = string("transpose_104")];
            tensor<fp16, [2, 8, ?, 32]> x_a_3_cast_fp16 = slice_by_index(begin = x_a_3_begin_0, end = x_a_3_end_0, end_mask = x_a_3_end_mask_0, x = x_29_cast_fp16)[name = string("x_a_3_cast_fp16")];
            tensor<int32, [4]> x_b_3_begin_0 = const()[name = string("x_b_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x_b_3_end_0 = const()[name = string("x_b_3_end_0"), val = tensor<int32, [4]>([2, 8, 0, 64])];
            tensor<bool, [4]> x_b_3_end_mask_0 = const()[name = string("x_b_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [2, 8, ?, 32]> x_b_3_cast_fp16 = slice_by_index(begin = x_b_3_begin_0, end = x_b_3_end_0, end_mask = x_b_3_end_mask_0, x = x_29_cast_fp16)[name = string("x_b_3_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_454_cast_fp16 = mul(x = x_a_3_cast_fp16, y = cos_3_cast_fp16)[name = string("op_454_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_455_cast_fp16 = mul(x = x_b_3_cast_fp16, y = sin_3_cast_fp16)[name = string("op_455_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_a_3_cast_fp16 = sub(x = var_454_cast_fp16, y = var_455_cast_fp16)[name = string("rot_a_3_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_457_cast_fp16 = mul(x = x_a_3_cast_fp16, y = sin_3_cast_fp16)[name = string("op_457_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_458_cast_fp16 = mul(x = x_b_3_cast_fp16, y = cos_3_cast_fp16)[name = string("op_458_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_b_3_cast_fp16 = add(x = var_457_cast_fp16, y = var_458_cast_fp16)[name = string("rot_b_3_cast_fp16")];
            bool k_1_interleave_0 = const()[name = string("k_1_interleave_0"), val = bool(false)];
            tensor<fp16, [2, 8, ?, 64]> k_1_cast_fp16 = concat(axis = var_126, interleave = k_1_interleave_0, values = (rot_a_3_cast_fp16, rot_b_3_cast_fp16))[name = string("k_1_cast_fp16")];
            bool var_463_transpose_x_1 = const()[name = string("op_463_transpose_x_1"), val = bool(false)];
            bool var_463_transpose_y_1 = const()[name = string("op_463_transpose_y_1"), val = bool(true)];
            tensor<fp16, [2, 8, ?, ?]> var_463_cast_fp16 = matmul(transpose_x = var_463_transpose_x_1, transpose_y = var_463_transpose_y_1, x = q_1_cast_fp16, y = k_1_cast_fp16)[name = string("op_463_cast_fp16")];
            fp16 _inversed_scores_1_y_0_to_fp16 = const()[name = string("_inversed_scores_1_y_0_to_fp16"), val = fp16(0x1p-4)];
            tensor<fp16, [2, 8, ?, ?]> _inversed_scores_1_cast_fp16 = mul(x = var_463_cast_fp16, y = _inversed_scores_1_y_0_to_fp16)[name = string("_inversed_scores_1_cast_fp16")];
            tensor<int32, [1]> var_466_axes_0 = const()[name = string("op_466_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [2, 1, 1, ?]> var_466_cast_fp16 = expand_dims(axes = var_466_axes_0, x = text_mask_cast_fp16)[name = string("op_466_cast_fp16")];
            fp16 var_125_to_fp16 = const()[name = string("op_125_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [2, 1, 1, ?]> var_467_cast_fp16 = sub(x = var_125_to_fp16, y = var_466_cast_fp16)[name = string("op_467_cast_fp16")];
            fp16 var_468_to_fp16 = const()[name = string("op_468_to_fp16"), val = fp16(0x1.388p+13)];
            tensor<fp16, [2, 1, 1, ?]> var_469_cast_fp16 = mul(x = var_467_cast_fp16, y = var_468_to_fp16)[name = string("op_469_cast_fp16")];
            tensor<fp16, [2, 8, ?, ?]> input_65_cast_fp16 = sub(x = _inversed_scores_1_cast_fp16, y = var_469_cast_fp16)[name = string("input_65_cast_fp16")];
            tensor<fp16, [2, 8, ?, ?]> attn_1_cast_fp16 = softmax(axis = var_126, x = input_65_cast_fp16)[name = string("attn_1_cast_fp16")];
            bool var_472_transpose_x_0 = const()[name = string("op_472_transpose_x_0"), val = bool(false)];
            bool var_472_transpose_y_0 = const()[name = string("op_472_transpose_y_0"), val = bool(false)];
            tensor<fp16, [2, 8, ?, 64]> v_1_cast_fp16 = transpose(perm = v_1_perm_0, x = var_408_cast_fp16)[name = string("transpose_103")];
            tensor<fp16, [2, 8, ?, 64]> var_472_cast_fp16 = matmul(transpose_x = var_472_transpose_x_0, transpose_y = var_472_transpose_y_0, x = attn_1_cast_fp16, y = v_1_cast_fp16)[name = string("op_472_cast_fp16")];
            tensor<int32, [4]> var_473_perm_0 = const()[name = string("op_473_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_11x = const()[name = string("concat_11x"), val = tensor<int32, [3]>([2, -1, 512])];
            tensor<fp16, [2, ?, 8, 64]> var_473_cast_fp16 = transpose(perm = var_473_perm_0, x = var_472_cast_fp16)[name = string("transpose_102")];
            tensor<fp16, [2, ?, 512]> input_67_cast_fp16 = reshape(shape = concat_11x, x = var_473_cast_fp16)[name = string("input_67_cast_fp16")];
            tensor<fp16, [512, 512]> main_blocks_0_text_attn_out_fc_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11264960))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11527168))))[name = string("main_blocks_0_text_attn_out_fc_linear_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_0_text_attn_out_fc_linear_bias_to_fp16 = const()[name = string("main_blocks_0_text_attn_out_fc_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11528256)))];
            tensor<fp16, [2, ?, 512]> linear_6_cast_fp16 = linear(bias = main_blocks_0_text_attn_out_fc_linear_bias_to_fp16, weight = main_blocks_0_text_attn_out_fc_linear_weight_to_fp16_quantized, x = input_67_cast_fp16)[name = string("linear_6_cast_fp16")];
            tensor<int32, [3]> out_11_perm_0 = const()[name = string("out_11_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> out_11_cast_fp16 = transpose(perm = out_11_perm_0, x = linear_6_cast_fp16)[name = string("transpose_101")];
            tensor<fp16, [2, 512, ?]> var_481_cast_fp16 = mul(x = out_11_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_481_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_31_cast_fp16 = add(x = x_m_1_cast_fp16, y = var_481_cast_fp16)[name = string("x_31_cast_fp16")];
            tensor<int32, [3]> input_69_perm_0 = const()[name = string("input_69_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_488_axes_0 = const()[name = string("op_488_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_0_text_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_0_text_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11529344)))];
            tensor<fp16, [512]> main_blocks_0_text_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_0_text_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11530432)))];
            tensor<fp16, [2, ?, 512]> input_69_cast_fp16 = transpose(perm = input_69_perm_0, x = x_31_cast_fp16)[name = string("transpose_100")];
            tensor<fp16, [2, ?, 512]> var_488_cast_fp16 = layer_norm(axes = var_488_axes_0, beta = main_blocks_0_text_norm_norm_bias_to_fp16, epsilon = var_138_to_fp16, gamma = main_blocks_0_text_norm_norm_weight_to_fp16, x = input_69_cast_fp16)[name = string("op_488_cast_fp16")];
            tensor<int32, [3]> var_489_perm_0 = const()[name = string("op_489_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> var_489_cast_fp16 = transpose(perm = var_489_perm_0, x = var_488_cast_fp16)[name = string("transpose_99")];
            tensor<fp16, [2, 512, ?]> x_33_cast_fp16 = mul(x = var_489_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_33_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_71_cast_fp16 = mul(x = x_33_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("input_71_cast_fp16")];
            tensor<int32, [6]> input_73_pad_0 = const()[name = string("input_73_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_73_mode_0 = const()[name = string("input_73_mode_0"), val = string("replicate")];
            fp16 const_5_to_fp16 = const()[name = string("const_5_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_73_cast_fp16 = pad(constant_val = const_5_to_fp16, mode = input_73_mode_0, pad = input_73_pad_0, x = input_71_cast_fp16)[name = string("input_73_cast_fp16")];
            string h_33_pad_type_0 = const()[name = string("h_33_pad_type_0"), val = string("valid")];
            int32 h_33_groups_0 = const()[name = string("h_33_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_33_strides_0 = const()[name = string("h_33_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_33_pad_0 = const()[name = string("h_33_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_33_dilations_0 = const()[name = string("h_33_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 5]> main_blocks_0_convnext_2_0_dwconv__conv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11531520))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11534144))))[name = string("main_blocks_0_convnext_2_0_dwconv__conv_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_0_convnext_2_0_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_2_0_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11535232)))];
            tensor<fp16, [2, 512, ?]> h_33_cast_fp16 = conv(bias = main_blocks_0_convnext_2_0_dwconv__conv_bias_to_fp16, dilations = h_33_dilations_0, groups = h_33_groups_0, pad = h_33_pad_0, pad_type = h_33_pad_type_0, strides = h_33_strides_0, weight = main_blocks_0_convnext_2_0_dwconv__conv_weight_to_fp16_quantized, x = input_73_cast_fp16)[name = string("h_33_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_35_cast_fp16 = mul(x = h_33_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_35_cast_fp16")];
            tensor<int32, [3]> input_75_perm_0 = const()[name = string("input_75_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_513_axes_0 = const()[name = string("op_513_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_0_convnext_2_0_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_0_convnext_2_0_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11536320)))];
            tensor<fp16, [512]> main_blocks_0_convnext_2_0_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_2_0_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11537408)))];
            tensor<fp16, [2, ?, 512]> input_75_cast_fp16 = transpose(perm = input_75_perm_0, x = x_35_cast_fp16)[name = string("transpose_98")];
            tensor<fp16, [2, ?, 512]> var_513_cast_fp16 = layer_norm(axes = var_513_axes_0, beta = main_blocks_0_convnext_2_0_norm_norm_bias_to_fp16, epsilon = var_138_to_fp16, gamma = main_blocks_0_convnext_2_0_norm_norm_weight_to_fp16, x = input_75_cast_fp16)[name = string("op_513_cast_fp16")];
            tensor<int32, [3]> input_77_perm_0 = const()[name = string("input_77_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_35_pad_type_0 = const()[name = string("h_35_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_35_strides_0 = const()[name = string("h_35_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_35_pad_0 = const()[name = string("h_35_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_35_dilations_0 = const()[name = string("h_35_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_35_groups_0 = const()[name = string("h_35_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_0_convnext_2_0_pwconv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11538496))), scale = tensor<fp16, [2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12587136))))[name = string("main_blocks_0_convnext_2_0_pwconv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> main_blocks_0_convnext_2_0_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_0_convnext_2_0_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12591296)))];
            tensor<fp16, [2, 512, ?]> input_77_cast_fp16 = transpose(perm = input_77_perm_0, x = var_513_cast_fp16)[name = string("transpose_97")];
            tensor<fp16, [2, 2048, ?]> h_35_cast_fp16 = conv(bias = main_blocks_0_convnext_2_0_pwconv1_bias_to_fp16, dilations = h_35_dilations_0, groups = h_35_groups_0, pad = h_35_pad_0, pad_type = h_35_pad_type_0, strides = h_35_strides_0, weight = main_blocks_0_convnext_2_0_pwconv1_weight_to_fp16_quantized, x = input_77_cast_fp16)[name = string("h_35_cast_fp16")];
            string input_79_mode_0 = const()[name = string("input_79_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_79_cast_fp16 = gelu(mode = input_79_mode_0, x = h_35_cast_fp16)[name = string("input_79_cast_fp16")];
            string h_37_pad_type_0 = const()[name = string("h_37_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_37_strides_0 = const()[name = string("h_37_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_37_pad_0 = const()[name = string("h_37_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_37_dilations_0 = const()[name = string("h_37_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_37_groups_0 = const()[name = string("h_37_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> op_530_weight_0_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12595456))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13644096))))[name = string("op_530_weight_0_to_fp16_quantized")];
            tensor<fp16, [512]> var_530_bias_0_to_fp16 = const()[name = string("op_530_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13645184)))];
            tensor<fp16, [2, 512, ?]> var_530_cast_fp16 = conv(bias = var_530_bias_0_to_fp16, dilations = h_37_dilations_0, groups = h_37_groups_0, pad = h_37_pad_0, pad_type = h_37_pad_type_0, strides = h_37_strides_0, weight = op_530_weight_0_to_fp16_quantized, x = input_79_cast_fp16)[name = string("op_530_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_13_cast_fp16 = add(x = input_71_cast_fp16, y = var_530_cast_fp16)[name = string("out_13_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_37_cast_fp16 = mul(x = out_13_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_37_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_m_3_cast_fp16 = mul(x = x_37_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_m_3_cast_fp16")];
            tensor<int32, [3]> input_81_perm_0 = const()[name = string("input_81_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [256, 512]> main_blocks_0_style_attn_W_query_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [256, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13646272))), scale = tensor<fp16, [256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13777408))))[name = string("main_blocks_0_style_attn_W_query_linear_weight_to_fp16_quantized")];
            tensor<fp16, [256]> main_blocks_0_style_attn_W_query_linear_bias_to_fp16 = const()[name = string("main_blocks_0_style_attn_W_query_linear_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13777984)))];
            tensor<fp16, [2, ?, 512]> input_81_cast_fp16 = transpose(perm = input_81_perm_0, x = x_37_cast_fp16)[name = string("transpose_96")];
            tensor<fp16, [2, ?, 256]> linear_7_cast_fp16 = linear(bias = main_blocks_0_style_attn_W_query_linear_bias_to_fp16, weight = main_blocks_0_style_attn_W_query_linear_weight_to_fp16_quantized, x = input_81_cast_fp16)[name = string("linear_7_cast_fp16")];
            tensor<int32, [4]> concat_12x = const()[name = string("concat_12x"), val = tensor<int32, [4]>([2, -1, 2, 128])];
            tensor<fp16, [2, ?, 2, 128]> var_547_cast_fp16 = reshape(shape = concat_12x, x = linear_7_cast_fp16)[name = string("op_547_cast_fp16")];
            tensor<int32, [4]> q_3_perm_0 = const()[name = string("q_3_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [256, 256]> main_blocks_0_style_attn_W_value_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [256, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13778560))), scale = tensor<fp16, [256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13844160))))[name = string("main_blocks_0_style_attn_W_value_linear_weight_to_fp16_quantized")];
            tensor<fp16, [256]> main_blocks_0_style_attn_W_value_linear_bias_to_fp16 = const()[name = string("main_blocks_0_style_attn_W_value_linear_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13844736)))];
            tensor<fp16, [2, 50, 256]> linear_9_cast_fp16 = linear(bias = main_blocks_0_style_attn_W_value_linear_bias_to_fp16, weight = main_blocks_0_style_attn_W_value_linear_weight_to_fp16_quantized, x = input_83_cast_fp16)[name = string("linear_9_cast_fp16")];
            tensor<int32, [4]> var_560 = const()[name = string("op_560"), val = tensor<int32, [4]>([2, 50, 2, 128])];
            tensor<fp16, [2, 50, 2, 128]> var_561_cast_fp16 = reshape(shape = var_560, x = linear_9_cast_fp16)[name = string("op_561_cast_fp16")];
            tensor<int32, [4]> v_3_perm_0 = const()[name = string("v_3_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool var_565_transpose_x_0 = const()[name = string("op_565_transpose_x_0"), val = bool(false)];
            bool var_565_transpose_y_0 = const()[name = string("op_565_transpose_y_0"), val = bool(false)];
            tensor<fp16, [2, 2, 128, 50]> op_564_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2, 2, 128, 50]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13845312))), scale = tensor<fp16, [1, 1, 1, 50]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13870976))))[name = string("op_564_to_fp16_quantized")];
            tensor<fp16, [2, 2, ?, 128]> q_3_cast_fp16 = transpose(perm = q_3_perm_0, x = var_547_cast_fp16)[name = string("transpose_95")];
            tensor<fp16, [2, 2, ?, 50]> var_565_cast_fp16 = matmul(transpose_x = var_565_transpose_x_0, transpose_y = var_565_transpose_y_0, x = q_3_cast_fp16, y = op_564_to_fp16_quantized)[name = string("op_565_cast_fp16")];
            fp16 _inversed_input_85_y_0_to_fp16 = const()[name = string("_inversed_input_85_y_0_to_fp16"), val = fp16(0x1p-4)];
            tensor<fp16, [2, 2, ?, 50]> _inversed_input_85_cast_fp16 = mul(x = var_565_cast_fp16, y = _inversed_input_85_y_0_to_fp16)[name = string("_inversed_input_85_cast_fp16")];
            tensor<fp16, [2, 2, ?, 50]> attn_3_cast_fp16 = softmax(axis = var_126, x = _inversed_input_85_cast_fp16)[name = string("attn_3_cast_fp16")];
            tensor<int32, [3]> var_569_perm_0 = const()[name = string("op_569_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> mask_1_axes_0 = const()[name = string("mask_1_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, ?, 1]> var_569_cast_fp16 = transpose(perm = var_569_perm_0, x = latent_mask_b_cast_fp16)[name = string("transpose_94")];
            tensor<fp16, [2, 1, ?, 1]> mask_1_cast_fp16 = expand_dims(axes = mask_1_axes_0, x = var_569_cast_fp16)[name = string("mask_1_cast_fp16")];
            tensor<fp16, [2, 2, ?, 50]> attn_5_cast_fp16 = mul(x = attn_3_cast_fp16, y = mask_1_cast_fp16)[name = string("attn_5_cast_fp16")];
            bool var_572_transpose_x_0 = const()[name = string("op_572_transpose_x_0"), val = bool(false)];
            bool var_572_transpose_y_0 = const()[name = string("op_572_transpose_y_0"), val = bool(false)];
            tensor<fp16, [2, 2, 50, 128]> v_3_cast_fp16 = transpose(perm = v_3_perm_0, x = var_561_cast_fp16)[name = string("transpose_93")];
            tensor<fp16, [2, 2, ?, 128]> var_572_cast_fp16 = matmul(transpose_x = var_572_transpose_x_0, transpose_y = var_572_transpose_y_0, x = attn_5_cast_fp16, y = v_3_cast_fp16)[name = string("op_572_cast_fp16")];
            tensor<int32, [4]> var_573_perm_0 = const()[name = string("op_573_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_13x = const()[name = string("concat_13x"), val = tensor<int32, [3]>([2, -1, 256])];
            tensor<fp16, [2, ?, 2, 128]> var_573_cast_fp16 = transpose(perm = var_573_perm_0, x = var_572_cast_fp16)[name = string("transpose_92")];
            tensor<fp16, [2, ?, 256]> input_87_cast_fp16 = reshape(shape = concat_13x, x = var_573_cast_fp16)[name = string("input_87_cast_fp16")];
            tensor<fp16, [512, 256]> main_blocks_0_style_attn_out_fc_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13871168))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14002304))))[name = string("main_blocks_0_style_attn_out_fc_linear_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_0_style_attn_out_fc_linear_bias_to_fp16 = const()[name = string("main_blocks_0_style_attn_out_fc_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14003392)))];
            tensor<fp16, [2, ?, 512]> linear_10_cast_fp16 = linear(bias = main_blocks_0_style_attn_out_fc_linear_bias_to_fp16, weight = main_blocks_0_style_attn_out_fc_linear_weight_to_fp16_quantized, x = input_87_cast_fp16)[name = string("linear_10_cast_fp16")];
            tensor<int32, [3]> out_15_perm_0 = const()[name = string("out_15_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> out_15_cast_fp16 = transpose(perm = out_15_perm_0, x = linear_10_cast_fp16)[name = string("transpose_91")];
            tensor<fp16, [2, 512, ?]> var_581_cast_fp16 = mul(x = out_15_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_581_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_39_cast_fp16 = add(x = x_m_3_cast_fp16, y = var_581_cast_fp16)[name = string("x_39_cast_fp16")];
            tensor<int32, [3]> input_89_perm_0 = const()[name = string("input_89_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_588_axes_0 = const()[name = string("op_588_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_0_style_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_0_style_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14004480)))];
            tensor<fp16, [512]> main_blocks_0_style_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_0_style_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14005568)))];
            tensor<fp16, [2, ?, 512]> input_89_cast_fp16 = transpose(perm = input_89_perm_0, x = x_39_cast_fp16)[name = string("transpose_90")];
            tensor<fp16, [2, ?, 512]> var_588_cast_fp16 = layer_norm(axes = var_588_axes_0, beta = main_blocks_0_style_norm_norm_bias_to_fp16, epsilon = var_138_to_fp16, gamma = main_blocks_0_style_norm_norm_weight_to_fp16, x = input_89_cast_fp16)[name = string("op_588_cast_fp16")];
            tensor<int32, [3]> var_589_perm_0 = const()[name = string("op_589_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> var_589_cast_fp16 = transpose(perm = var_589_perm_0, x = var_588_cast_fp16)[name = string("transpose_89")];
            tensor<fp16, [2, 512, ?]> x_41_cast_fp16 = mul(x = var_589_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_41_cast_fp16")];
            int32 var_600 = const()[name = string("op_600"), val = int32(-1)];
            tensor<fp16, [2, 512, ?]> input_91_cast_fp16 = mul(x = x_41_cast_fp16, y = latent_mask_b_cast_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, 2, 2])];
            string input_93_mode_0 = const()[name = string("input_93_mode_0"), val = string("replicate")];
            fp16 const_7_to_fp16 = const()[name = string("const_7_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_93_cast_fp16 = pad(constant_val = const_7_to_fp16, mode = input_93_mode_0, pad = input_93_pad_0, x = input_91_cast_fp16)[name = string("input_93_cast_fp16")];
            string h_39_pad_type_0 = const()[name = string("h_39_pad_type_0"), val = string("valid")];
            int32 h_39_groups_0 = const()[name = string("h_39_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_39_strides_0 = const()[name = string("h_39_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_39_pad_0 = const()[name = string("h_39_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_39_dilations_0 = const()[name = string("h_39_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 5]> main_blocks_1_convnext_0_0_dwconv__conv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14006656))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14009280))))[name = string("main_blocks_1_convnext_0_0_dwconv__conv_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_1_convnext_0_0_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_0_0_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14010368)))];
            tensor<fp16, [2, 512, ?]> h_39_cast_fp16 = conv(bias = main_blocks_1_convnext_0_0_dwconv__conv_bias_to_fp16, dilations = h_39_dilations_0, groups = h_39_groups_0, pad = h_39_pad_0, pad_type = h_39_pad_type_0, strides = h_39_strides_0, weight = main_blocks_1_convnext_0_0_dwconv__conv_weight_to_fp16_quantized, x = input_93_cast_fp16)[name = string("h_39_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_43_cast_fp16 = mul(x = h_39_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_43_cast_fp16")];
            tensor<int32, [3]> input_95_perm_0 = const()[name = string("input_95_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_653_axes_0 = const()[name = string("op_653_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_1_convnext_0_0_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_1_convnext_0_0_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14011456)))];
            tensor<fp16, [512]> main_blocks_1_convnext_0_0_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_0_0_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14012544)))];
            fp16 var_612_to_fp16 = const()[name = string("op_612_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [2, ?, 512]> input_95_cast_fp16 = transpose(perm = input_95_perm_0, x = x_43_cast_fp16)[name = string("transpose_88")];
            tensor<fp16, [2, ?, 512]> var_653_cast_fp16 = layer_norm(axes = var_653_axes_0, beta = main_blocks_1_convnext_0_0_norm_norm_bias_to_fp16, epsilon = var_612_to_fp16, gamma = main_blocks_1_convnext_0_0_norm_norm_weight_to_fp16, x = input_95_cast_fp16)[name = string("op_653_cast_fp16")];
            tensor<int32, [3]> input_97_perm_0 = const()[name = string("input_97_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_41_pad_type_0 = const()[name = string("h_41_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_41_strides_0 = const()[name = string("h_41_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_41_pad_0 = const()[name = string("h_41_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_41_dilations_0 = const()[name = string("h_41_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_41_groups_0 = const()[name = string("h_41_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_1_convnext_0_0_pwconv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14013632))), scale = tensor<fp16, [2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15062272))))[name = string("main_blocks_1_convnext_0_0_pwconv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> main_blocks_1_convnext_0_0_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_0_0_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15066432)))];
            tensor<fp16, [2, 512, ?]> input_97_cast_fp16 = transpose(perm = input_97_perm_0, x = var_653_cast_fp16)[name = string("transpose_87")];
            tensor<fp16, [2, 2048, ?]> h_41_cast_fp16 = conv(bias = main_blocks_1_convnext_0_0_pwconv1_bias_to_fp16, dilations = h_41_dilations_0, groups = h_41_groups_0, pad = h_41_pad_0, pad_type = h_41_pad_type_0, strides = h_41_strides_0, weight = main_blocks_1_convnext_0_0_pwconv1_weight_to_fp16_quantized, x = input_97_cast_fp16)[name = string("h_41_cast_fp16")];
            string input_99_mode_0 = const()[name = string("input_99_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_99_cast_fp16 = gelu(mode = input_99_mode_0, x = h_41_cast_fp16)[name = string("input_99_cast_fp16")];
            string h_43_pad_type_0 = const()[name = string("h_43_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_43_strides_0 = const()[name = string("h_43_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_43_pad_0 = const()[name = string("h_43_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_43_dilations_0 = const()[name = string("h_43_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_43_groups_0 = const()[name = string("h_43_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> op_670_weight_0_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15070592))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16119232))))[name = string("op_670_weight_0_to_fp16_quantized")];
            tensor<fp16, [512]> var_670_bias_0_to_fp16 = const()[name = string("op_670_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16120320)))];
            tensor<fp16, [2, 512, ?]> var_670_cast_fp16 = conv(bias = var_670_bias_0_to_fp16, dilations = h_43_dilations_0, groups = h_43_groups_0, pad = h_43_pad_0, pad_type = h_43_pad_type_0, strides = h_43_strides_0, weight = op_670_weight_0_to_fp16_quantized, x = input_99_cast_fp16)[name = string("op_670_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_17_cast_fp16 = add(x = input_91_cast_fp16, y = var_670_cast_fp16)[name = string("out_17_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_45_cast_fp16 = mul(x = out_17_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_45_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_101_cast_fp16 = mul(x = x_45_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("input_101_cast_fp16")];
            tensor<int32, [6]> input_103_pad_0 = const()[name = string("input_103_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_103_mode_0 = const()[name = string("input_103_mode_0"), val = string("replicate")];
            fp16 const_8_to_fp16 = const()[name = string("const_8_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_103_cast_fp16 = pad(constant_val = const_8_to_fp16, mode = input_103_mode_0, pad = input_103_pad_0, x = input_101_cast_fp16)[name = string("input_103_cast_fp16")];
            string h_45_pad_type_0 = const()[name = string("h_45_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_45_dilations_0 = const()[name = string("h_45_dilations_0"), val = tensor<int32, [1]>([2])];
            int32 h_45_groups_0 = const()[name = string("h_45_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_45_strides_0 = const()[name = string("h_45_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_45_pad_0 = const()[name = string("h_45_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [512, 1, 5]> main_blocks_1_convnext_0_1_dwconv__conv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16121408))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16124032))))[name = string("main_blocks_1_convnext_0_1_dwconv__conv_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_1_convnext_0_1_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_0_1_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16125120)))];
            tensor<fp16, [2, 512, ?]> h_45_cast_fp16 = conv(bias = main_blocks_1_convnext_0_1_dwconv__conv_bias_to_fp16, dilations = h_45_dilations_0, groups = h_45_groups_0, pad = h_45_pad_0, pad_type = h_45_pad_type_0, strides = h_45_strides_0, weight = main_blocks_1_convnext_0_1_dwconv__conv_weight_to_fp16_quantized, x = input_103_cast_fp16)[name = string("h_45_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_47_cast_fp16 = mul(x = h_45_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_47_cast_fp16")];
            tensor<int32, [3]> input_105_perm_0 = const()[name = string("input_105_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_695_axes_0 = const()[name = string("op_695_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_1_convnext_0_1_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_1_convnext_0_1_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16126208)))];
            tensor<fp16, [512]> main_blocks_1_convnext_0_1_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_0_1_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16127296)))];
            tensor<fp16, [2, ?, 512]> input_105_cast_fp16 = transpose(perm = input_105_perm_0, x = x_47_cast_fp16)[name = string("transpose_86")];
            tensor<fp16, [2, ?, 512]> var_695_cast_fp16 = layer_norm(axes = var_695_axes_0, beta = main_blocks_1_convnext_0_1_norm_norm_bias_to_fp16, epsilon = var_612_to_fp16, gamma = main_blocks_1_convnext_0_1_norm_norm_weight_to_fp16, x = input_105_cast_fp16)[name = string("op_695_cast_fp16")];
            tensor<int32, [3]> input_107_perm_0 = const()[name = string("input_107_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_47_pad_type_0 = const()[name = string("h_47_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_47_strides_0 = const()[name = string("h_47_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_47_pad_0 = const()[name = string("h_47_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_47_dilations_0 = const()[name = string("h_47_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_47_groups_0 = const()[name = string("h_47_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_1_convnext_0_1_pwconv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16128384))), scale = tensor<fp16, [2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17177024))))[name = string("main_blocks_1_convnext_0_1_pwconv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> main_blocks_1_convnext_0_1_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_0_1_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17181184)))];
            tensor<fp16, [2, 512, ?]> input_107_cast_fp16 = transpose(perm = input_107_perm_0, x = var_695_cast_fp16)[name = string("transpose_85")];
            tensor<fp16, [2, 2048, ?]> h_47_cast_fp16 = conv(bias = main_blocks_1_convnext_0_1_pwconv1_bias_to_fp16, dilations = h_47_dilations_0, groups = h_47_groups_0, pad = h_47_pad_0, pad_type = h_47_pad_type_0, strides = h_47_strides_0, weight = main_blocks_1_convnext_0_1_pwconv1_weight_to_fp16_quantized, x = input_107_cast_fp16)[name = string("h_47_cast_fp16")];
            string input_109_mode_0 = const()[name = string("input_109_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_109_cast_fp16 = gelu(mode = input_109_mode_0, x = h_47_cast_fp16)[name = string("input_109_cast_fp16")];
            string h_49_pad_type_0 = const()[name = string("h_49_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_49_strides_0 = const()[name = string("h_49_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_49_pad_0 = const()[name = string("h_49_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_49_dilations_0 = const()[name = string("h_49_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_49_groups_0 = const()[name = string("h_49_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> op_712_weight_0_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17185344))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18233984))))[name = string("op_712_weight_0_to_fp16_quantized")];
            tensor<fp16, [512]> var_712_bias_0_to_fp16 = const()[name = string("op_712_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18235072)))];
            tensor<fp16, [2, 512, ?]> var_712_cast_fp16 = conv(bias = var_712_bias_0_to_fp16, dilations = h_49_dilations_0, groups = h_49_groups_0, pad = h_49_pad_0, pad_type = h_49_pad_type_0, strides = h_49_strides_0, weight = op_712_weight_0_to_fp16_quantized, x = input_109_cast_fp16)[name = string("op_712_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_19_cast_fp16 = add(x = input_101_cast_fp16, y = var_712_cast_fp16)[name = string("out_19_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_49_cast_fp16 = mul(x = out_19_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_49_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_111_cast_fp16 = mul(x = x_49_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("input_111_cast_fp16")];
            tensor<int32, [6]> input_113_pad_0 = const()[name = string("input_113_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 8, 8])];
            string input_113_mode_0 = const()[name = string("input_113_mode_0"), val = string("replicate")];
            fp16 const_9_to_fp16 = const()[name = string("const_9_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_113_cast_fp16 = pad(constant_val = const_9_to_fp16, mode = input_113_mode_0, pad = input_113_pad_0, x = input_111_cast_fp16)[name = string("input_113_cast_fp16")];
            string h_51_pad_type_0 = const()[name = string("h_51_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_51_dilations_0 = const()[name = string("h_51_dilations_0"), val = tensor<int32, [1]>([4])];
            int32 h_51_groups_0 = const()[name = string("h_51_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_51_strides_0 = const()[name = string("h_51_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_51_pad_0 = const()[name = string("h_51_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [512, 1, 5]> main_blocks_1_convnext_0_2_dwconv__conv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18236160))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18238784))))[name = string("main_blocks_1_convnext_0_2_dwconv__conv_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_1_convnext_0_2_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_0_2_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18239872)))];
            tensor<fp16, [2, 512, ?]> h_51_cast_fp16 = conv(bias = main_blocks_1_convnext_0_2_dwconv__conv_bias_to_fp16, dilations = h_51_dilations_0, groups = h_51_groups_0, pad = h_51_pad_0, pad_type = h_51_pad_type_0, strides = h_51_strides_0, weight = main_blocks_1_convnext_0_2_dwconv__conv_weight_to_fp16_quantized, x = input_113_cast_fp16)[name = string("h_51_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_51_cast_fp16 = mul(x = h_51_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_51_cast_fp16")];
            tensor<int32, [3]> input_115_perm_0 = const()[name = string("input_115_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_737_axes_0 = const()[name = string("op_737_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_1_convnext_0_2_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_1_convnext_0_2_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18240960)))];
            tensor<fp16, [512]> main_blocks_1_convnext_0_2_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_0_2_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18242048)))];
            tensor<fp16, [2, ?, 512]> input_115_cast_fp16 = transpose(perm = input_115_perm_0, x = x_51_cast_fp16)[name = string("transpose_84")];
            tensor<fp16, [2, ?, 512]> var_737_cast_fp16 = layer_norm(axes = var_737_axes_0, beta = main_blocks_1_convnext_0_2_norm_norm_bias_to_fp16, epsilon = var_612_to_fp16, gamma = main_blocks_1_convnext_0_2_norm_norm_weight_to_fp16, x = input_115_cast_fp16)[name = string("op_737_cast_fp16")];
            tensor<int32, [3]> input_117_perm_0 = const()[name = string("input_117_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_53_pad_type_0 = const()[name = string("h_53_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_53_strides_0 = const()[name = string("h_53_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_53_pad_0 = const()[name = string("h_53_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_53_dilations_0 = const()[name = string("h_53_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_53_groups_0 = const()[name = string("h_53_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_1_convnext_0_2_pwconv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18243136))), scale = tensor<fp16, [2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19291776))))[name = string("main_blocks_1_convnext_0_2_pwconv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> main_blocks_1_convnext_0_2_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_0_2_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19295936)))];
            tensor<fp16, [2, 512, ?]> input_117_cast_fp16 = transpose(perm = input_117_perm_0, x = var_737_cast_fp16)[name = string("transpose_83")];
            tensor<fp16, [2, 2048, ?]> h_53_cast_fp16 = conv(bias = main_blocks_1_convnext_0_2_pwconv1_bias_to_fp16, dilations = h_53_dilations_0, groups = h_53_groups_0, pad = h_53_pad_0, pad_type = h_53_pad_type_0, strides = h_53_strides_0, weight = main_blocks_1_convnext_0_2_pwconv1_weight_to_fp16_quantized, x = input_117_cast_fp16)[name = string("h_53_cast_fp16")];
            string input_119_mode_0 = const()[name = string("input_119_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_119_cast_fp16 = gelu(mode = input_119_mode_0, x = h_53_cast_fp16)[name = string("input_119_cast_fp16")];
            string h_55_pad_type_0 = const()[name = string("h_55_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_55_strides_0 = const()[name = string("h_55_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_55_pad_0 = const()[name = string("h_55_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_55_dilations_0 = const()[name = string("h_55_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_55_groups_0 = const()[name = string("h_55_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> op_754_weight_0_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19300096))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20348736))))[name = string("op_754_weight_0_to_fp16_quantized")];
            tensor<fp16, [512]> var_754_bias_0_to_fp16 = const()[name = string("op_754_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20349824)))];
            tensor<fp16, [2, 512, ?]> var_754_cast_fp16 = conv(bias = var_754_bias_0_to_fp16, dilations = h_55_dilations_0, groups = h_55_groups_0, pad = h_55_pad_0, pad_type = h_55_pad_type_0, strides = h_55_strides_0, weight = op_754_weight_0_to_fp16_quantized, x = input_119_cast_fp16)[name = string("op_754_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_21_cast_fp16 = add(x = input_111_cast_fp16, y = var_754_cast_fp16)[name = string("out_21_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_53_cast_fp16 = mul(x = out_21_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_53_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_121_cast_fp16 = mul(x = x_53_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("input_121_cast_fp16")];
            tensor<int32, [6]> input_123_pad_0 = const()[name = string("input_123_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 16, 16])];
            string input_123_mode_0 = const()[name = string("input_123_mode_0"), val = string("replicate")];
            fp16 const_10_to_fp16 = const()[name = string("const_10_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_123_cast_fp16 = pad(constant_val = const_10_to_fp16, mode = input_123_mode_0, pad = input_123_pad_0, x = input_121_cast_fp16)[name = string("input_123_cast_fp16")];
            string h_57_pad_type_0 = const()[name = string("h_57_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_57_dilations_0 = const()[name = string("h_57_dilations_0"), val = tensor<int32, [1]>([8])];
            int32 h_57_groups_0 = const()[name = string("h_57_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_57_strides_0 = const()[name = string("h_57_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_57_pad_0 = const()[name = string("h_57_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [512, 1, 5]> main_blocks_1_convnext_0_3_dwconv__conv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20350912))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20353536))))[name = string("main_blocks_1_convnext_0_3_dwconv__conv_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_1_convnext_0_3_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_0_3_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20354624)))];
            tensor<fp16, [2, 512, ?]> h_57_cast_fp16 = conv(bias = main_blocks_1_convnext_0_3_dwconv__conv_bias_to_fp16, dilations = h_57_dilations_0, groups = h_57_groups_0, pad = h_57_pad_0, pad_type = h_57_pad_type_0, strides = h_57_strides_0, weight = main_blocks_1_convnext_0_3_dwconv__conv_weight_to_fp16_quantized, x = input_123_cast_fp16)[name = string("h_57_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_55_cast_fp16 = mul(x = h_57_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_55_cast_fp16")];
            tensor<int32, [3]> input_125_perm_0 = const()[name = string("input_125_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_779_axes_0 = const()[name = string("op_779_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_1_convnext_0_3_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_1_convnext_0_3_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20355712)))];
            tensor<fp16, [512]> main_blocks_1_convnext_0_3_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_0_3_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20356800)))];
            tensor<fp16, [2, ?, 512]> input_125_cast_fp16 = transpose(perm = input_125_perm_0, x = x_55_cast_fp16)[name = string("transpose_82")];
            tensor<fp16, [2, ?, 512]> var_779_cast_fp16 = layer_norm(axes = var_779_axes_0, beta = main_blocks_1_convnext_0_3_norm_norm_bias_to_fp16, epsilon = var_612_to_fp16, gamma = main_blocks_1_convnext_0_3_norm_norm_weight_to_fp16, x = input_125_cast_fp16)[name = string("op_779_cast_fp16")];
            tensor<int32, [3]> input_127_perm_0 = const()[name = string("input_127_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_59_pad_type_0 = const()[name = string("h_59_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_59_strides_0 = const()[name = string("h_59_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_59_pad_0 = const()[name = string("h_59_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_59_dilations_0 = const()[name = string("h_59_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_59_groups_0 = const()[name = string("h_59_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_1_convnext_0_3_pwconv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20357888))), scale = tensor<fp16, [2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21406528))))[name = string("main_blocks_1_convnext_0_3_pwconv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> main_blocks_1_convnext_0_3_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_0_3_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21410688)))];
            tensor<fp16, [2, 512, ?]> input_127_cast_fp16 = transpose(perm = input_127_perm_0, x = var_779_cast_fp16)[name = string("transpose_81")];
            tensor<fp16, [2, 2048, ?]> h_59_cast_fp16 = conv(bias = main_blocks_1_convnext_0_3_pwconv1_bias_to_fp16, dilations = h_59_dilations_0, groups = h_59_groups_0, pad = h_59_pad_0, pad_type = h_59_pad_type_0, strides = h_59_strides_0, weight = main_blocks_1_convnext_0_3_pwconv1_weight_to_fp16_quantized, x = input_127_cast_fp16)[name = string("h_59_cast_fp16")];
            string input_129_mode_0 = const()[name = string("input_129_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_129_cast_fp16 = gelu(mode = input_129_mode_0, x = h_59_cast_fp16)[name = string("input_129_cast_fp16")];
            string h_61_pad_type_0 = const()[name = string("h_61_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_61_strides_0 = const()[name = string("h_61_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_61_pad_0 = const()[name = string("h_61_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_61_dilations_0 = const()[name = string("h_61_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_61_groups_0 = const()[name = string("h_61_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> op_796_weight_0_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21414848))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22463488))))[name = string("op_796_weight_0_to_fp16_quantized")];
            tensor<fp16, [512]> var_796_bias_0_to_fp16 = const()[name = string("op_796_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22464576)))];
            tensor<fp16, [2, 512, ?]> var_796_cast_fp16 = conv(bias = var_796_bias_0_to_fp16, dilations = h_61_dilations_0, groups = h_61_groups_0, pad = h_61_pad_0, pad_type = h_61_pad_type_0, strides = h_61_strides_0, weight = op_796_weight_0_to_fp16_quantized, x = input_129_cast_fp16)[name = string("op_796_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_23_cast_fp16 = add(x = input_121_cast_fp16, y = var_796_cast_fp16)[name = string("out_23_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_57_cast_fp16 = mul(x = out_23_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_57_cast_fp16")];
            tensor<fp16, [512, 64]> main_blocks_1_time_cond_linear_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22465664))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22498496))))[name = string("main_blocks_1_time_cond_linear_linear_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_1_time_cond_linear_linear_bias_to_fp16 = const()[name = string("main_blocks_1_time_cond_linear_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22499584)))];
            tensor<fp16, [2, 512]> linear_11_cast_fp16 = linear(bias = main_blocks_1_time_cond_linear_linear_bias_to_fp16, weight = main_blocks_1_time_cond_linear_linear_weight_to_fp16_quantized, x = input_47_cast_fp16)[name = string("linear_11_cast_fp16")];
            tensor<int32, [1]> t_7_axes_0 = const()[name = string("t_7_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2, 512, 1]> t_7_cast_fp16 = expand_dims(axes = t_7_axes_0, x = linear_11_cast_fp16)[name = string("t_7_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_806_cast_fp16 = add(x = x_57_cast_fp16, y = t_7_cast_fp16)[name = string("op_806_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_59_cast_fp16 = mul(x = var_806_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_59_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_133_cast_fp16 = mul(x = x_59_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("input_133_cast_fp16")];
            tensor<int32, [6]> input_135_pad_0 = const()[name = string("input_135_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_135_mode_0 = const()[name = string("input_135_mode_0"), val = string("replicate")];
            fp16 const_11_to_fp16 = const()[name = string("const_11_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_135_cast_fp16 = pad(constant_val = const_11_to_fp16, mode = input_135_mode_0, pad = input_135_pad_0, x = input_133_cast_fp16)[name = string("input_135_cast_fp16")];
            string h_63_pad_type_0 = const()[name = string("h_63_pad_type_0"), val = string("valid")];
            int32 h_63_groups_0 = const()[name = string("h_63_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_63_strides_0 = const()[name = string("h_63_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_63_pad_0 = const()[name = string("h_63_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_63_dilations_0 = const()[name = string("h_63_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 5]> main_blocks_1_convnext_1_0_dwconv__conv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22500672))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22503296))))[name = string("main_blocks_1_convnext_1_0_dwconv__conv_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_1_convnext_1_0_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_1_0_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22504384)))];
            tensor<fp16, [2, 512, ?]> h_63_cast_fp16 = conv(bias = main_blocks_1_convnext_1_0_dwconv__conv_bias_to_fp16, dilations = h_63_dilations_0, groups = h_63_groups_0, pad = h_63_pad_0, pad_type = h_63_pad_type_0, strides = h_63_strides_0, weight = main_blocks_1_convnext_1_0_dwconv__conv_weight_to_fp16_quantized, x = input_135_cast_fp16)[name = string("h_63_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_61_cast_fp16 = mul(x = h_63_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_61_cast_fp16")];
            tensor<int32, [3]> input_137_perm_0 = const()[name = string("input_137_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_830_axes_0 = const()[name = string("op_830_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_1_convnext_1_0_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_1_convnext_1_0_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22505472)))];
            tensor<fp16, [512]> main_blocks_1_convnext_1_0_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_1_0_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22506560)))];
            tensor<fp16, [2, ?, 512]> input_137_cast_fp16 = transpose(perm = input_137_perm_0, x = x_61_cast_fp16)[name = string("transpose_80")];
            tensor<fp16, [2, ?, 512]> var_830_cast_fp16 = layer_norm(axes = var_830_axes_0, beta = main_blocks_1_convnext_1_0_norm_norm_bias_to_fp16, epsilon = var_612_to_fp16, gamma = main_blocks_1_convnext_1_0_norm_norm_weight_to_fp16, x = input_137_cast_fp16)[name = string("op_830_cast_fp16")];
            tensor<int32, [3]> input_139_perm_0 = const()[name = string("input_139_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_65_pad_type_0 = const()[name = string("h_65_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_65_strides_0 = const()[name = string("h_65_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_65_pad_0 = const()[name = string("h_65_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_65_dilations_0 = const()[name = string("h_65_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_65_groups_0 = const()[name = string("h_65_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_1_convnext_1_0_pwconv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22507648))), scale = tensor<fp16, [2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23556288))))[name = string("main_blocks_1_convnext_1_0_pwconv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> main_blocks_1_convnext_1_0_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_1_0_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23560448)))];
            tensor<fp16, [2, 512, ?]> input_139_cast_fp16 = transpose(perm = input_139_perm_0, x = var_830_cast_fp16)[name = string("transpose_79")];
            tensor<fp16, [2, 2048, ?]> h_65_cast_fp16 = conv(bias = main_blocks_1_convnext_1_0_pwconv1_bias_to_fp16, dilations = h_65_dilations_0, groups = h_65_groups_0, pad = h_65_pad_0, pad_type = h_65_pad_type_0, strides = h_65_strides_0, weight = main_blocks_1_convnext_1_0_pwconv1_weight_to_fp16_quantized, x = input_139_cast_fp16)[name = string("h_65_cast_fp16")];
            string input_141_mode_0 = const()[name = string("input_141_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_141_cast_fp16 = gelu(mode = input_141_mode_0, x = h_65_cast_fp16)[name = string("input_141_cast_fp16")];
            string h_67_pad_type_0 = const()[name = string("h_67_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_67_strides_0 = const()[name = string("h_67_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_67_pad_0 = const()[name = string("h_67_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_67_dilations_0 = const()[name = string("h_67_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_67_groups_0 = const()[name = string("h_67_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> op_847_weight_0_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23564608))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24613248))))[name = string("op_847_weight_0_to_fp16_quantized")];
            tensor<fp16, [512]> var_847_bias_0_to_fp16 = const()[name = string("op_847_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24614336)))];
            tensor<fp16, [2, 512, ?]> var_847_cast_fp16 = conv(bias = var_847_bias_0_to_fp16, dilations = h_67_dilations_0, groups = h_67_groups_0, pad = h_67_pad_0, pad_type = h_67_pad_type_0, strides = h_67_strides_0, weight = op_847_weight_0_to_fp16_quantized, x = input_141_cast_fp16)[name = string("op_847_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_25_cast_fp16 = add(x = input_133_cast_fp16, y = var_847_cast_fp16)[name = string("out_25_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_63_cast_fp16 = mul(x = out_25_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_63_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_m_5_cast_fp16 = mul(x = x_63_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_m_5_cast_fp16")];
            tensor<int32, [3]> var_857_shape_cast_fp16 = shape(x = x_63_cast_fp16)[name = string("op_857_shape_cast_fp16")];
            int32 gather_9_axis_0 = const()[name = string("gather_9_axis_0"), val = int32(0)];
            int32 gather_9_batch_dims_0 = const()[name = string("gather_9_batch_dims_0"), val = int32(0)];
            bool gather_9_validate_indices_0 = const()[name = string("gather_9_validate_indices_0"), val = bool(false)];
            string var_857_shape_cast_fp16_to_uint16_dtype_0 = const()[name = string("op_857_shape_cast_fp16_to_uint16_dtype_0"), val = string("uint16")];
            uint16 gather_9_indices_0_to_uint16 = const()[name = string("gather_9_indices_0_to_uint16"), val = uint16(2)];
            tensor<uint16, [3]> var_857_shape_cast_fp16_to_uint16 = cast(dtype = var_857_shape_cast_fp16_to_uint16_dtype_0, x = var_857_shape_cast_fp16)[name = string("cast_6")];
            uint16 gather_9_cast_uint16 = gather(axis = gather_9_axis_0, batch_dims = gather_9_batch_dims_0, indices = gather_9_indices_0_to_uint16, validate_indices = gather_9_validate_indices_0, x = var_857_shape_cast_fp16_to_uint16)[name = string("gather_9_cast_uint16")];
            string gather_9_cast_uint16_to_int32_dtype_0 = const()[name = string("gather_9_cast_uint16_to_int32_dtype_0"), val = string("int32")];
            tensor<int32, [3]> input_143_perm_0 = const()[name = string("input_143_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [512, 512]> main_blocks_1_text_attn_W_query_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24615424))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24877632))))[name = string("main_blocks_1_text_attn_W_query_linear_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_1_text_attn_W_query_linear_bias_to_fp16 = const()[name = string("main_blocks_1_text_attn_W_query_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24878720)))];
            tensor<fp16, [2, ?, 512]> input_143_cast_fp16 = transpose(perm = input_143_perm_0, x = x_63_cast_fp16)[name = string("transpose_78")];
            tensor<fp16, [2, ?, 512]> linear_12_cast_fp16 = linear(bias = main_blocks_1_text_attn_W_query_linear_bias_to_fp16, weight = main_blocks_1_text_attn_W_query_linear_weight_to_fp16_quantized, x = input_143_cast_fp16)[name = string("linear_12_cast_fp16")];
            tensor<int32, [4]> concat_14x = const()[name = string("concat_14x"), val = tensor<int32, [4]>([2, -1, 8, 64])];
            tensor<fp16, [2, ?, 8, 64]> var_866_cast_fp16 = reshape(shape = concat_14x, x = linear_12_cast_fp16)[name = string("op_866_cast_fp16")];
            tensor<int32, [4]> x_65_perm_0 = const()[name = string("x_65_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [512, 256]> main_blocks_1_text_attn_W_key_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24879808))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25010944))))[name = string("main_blocks_1_text_attn_W_key_linear_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_1_text_attn_W_key_linear_bias_to_fp16 = const()[name = string("main_blocks_1_text_attn_W_key_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25012032)))];
            tensor<fp16, [?, ?, 512]> linear_13_cast_fp16 = linear(bias = main_blocks_1_text_attn_W_key_linear_bias_to_fp16, weight = main_blocks_1_text_attn_W_key_linear_weight_to_fp16_quantized, x = input_61_cast_fp16)[name = string("linear_13_cast_fp16")];
            tensor<int32, [4]> concat_15x = const()[name = string("concat_15x"), val = tensor<int32, [4]>([2, -1, 8, 64])];
            tensor<fp16, [2, ?, 8, 64]> var_874_cast_fp16 = reshape(shape = concat_15x, x = linear_13_cast_fp16)[name = string("op_874_cast_fp16")];
            tensor<int32, [4]> x_67_perm_0 = const()[name = string("x_67_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [512, 256]> main_blocks_1_text_attn_W_value_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25013120))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25144256))))[name = string("main_blocks_1_text_attn_W_value_linear_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_1_text_attn_W_value_linear_bias_to_fp16 = const()[name = string("main_blocks_1_text_attn_W_value_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25145344)))];
            tensor<fp16, [?, ?, 512]> linear_14_cast_fp16 = linear(bias = main_blocks_1_text_attn_W_value_linear_bias_to_fp16, weight = main_blocks_1_text_attn_W_value_linear_weight_to_fp16_quantized, x = input_61_cast_fp16)[name = string("linear_14_cast_fp16")];
            tensor<int32, [4]> concat_16x = const()[name = string("concat_16x"), val = tensor<int32, [4]>([2, -1, 8, 64])];
            tensor<fp16, [2, ?, 8, 64]> var_882_cast_fp16 = reshape(shape = concat_16x, x = linear_14_cast_fp16)[name = string("op_882_cast_fp16")];
            tensor<int32, [4]> v_5_perm_0 = const()[name = string("v_5_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            int32 concat_17_values0_0 = const()[name = string("concat_17_values0_0"), val = int32(1)];
            int32 concat_17_values2_0 = const()[name = string("concat_17_values2_0"), val = int32(1)];
            int32 concat_17_axis_0 = const()[name = string("concat_17_axis_0"), val = int32(0)];
            bool concat_17_interleave_0 = const()[name = string("concat_17_interleave_0"), val = bool(false)];
            int32 gather_9_cast_uint16_to_int32 = cast(dtype = gather_9_cast_uint16_to_int32_dtype_0, x = gather_9_cast_uint16)[name = string("cast_5")];
            tensor<int32, [3]> concat_17 = concat(axis = concat_17_axis_0, interleave = concat_17_interleave_0, values = (concat_17_values0_0, gather_9_cast_uint16_to_int32, concat_17_values2_0))[name = string("concat_17")];
            tensor<int32, [3]> var_885_begin_0 = const()[name = string("op_885_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<bool, [3]> var_885_end_mask_0 = const()[name = string("op_885_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, ?, 1]> var_885_cast_fp16 = slice_by_index(begin = var_885_begin_0, end = concat_17, end_mask = var_885_end_mask_0, x = main_blocks_0_text_attn_increments_to_fp16)[name = string("op_885_cast_fp16")];
            tensor<int32, [3]> concat_18 = const()[name = string("concat_18"), val = tensor<int32, [3]>([2, -1, -1])];
            tensor<int32, [3]> shape_3_cast_fp16 = shape(x = var_885_cast_fp16)[name = string("shape_3_cast_fp16")];
            tensor<bool, [3]> equal_3 = const()[name = string("equal_3"), val = tensor<bool, [3]>([false, true, true])];
            tensor<int32, [3]> select_3 = select(a = shape_3_cast_fp16, b = concat_18, cond = equal_3)[name = string("select_3")];
            tensor<int32, [3]> real_div_3 = real_div(x = select_3, y = shape_3_cast_fp16)[name = string("real_div_3")];
            tensor<fp16, [?, ?, ?]> var_888_cast_fp16 = tile(reps = real_div_3, x = var_885_cast_fp16)[name = string("op_888_cast_fp16")];
            tensor<fp16, [2, ?, ?]> scaled_5_cast_fp16 = real_div(x = var_888_cast_fp16, y = var_427_cast_fp16)[name = string("scaled_5_cast_fp16")];
            tensor<fp16, [1, 1, 32]> main_blocks_1_text_attn_theta_to_fp16 = const()[name = string("main_blocks_1_text_attn_theta_to_fp16"), val = tensor<fp16, [1, 1, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25146432)))];
            tensor<fp16, [2, ?, 32]> angles_5_cast_fp16 = mul(x = scaled_5_cast_fp16, y = main_blocks_1_text_attn_theta_to_fp16)[name = string("angles_5_cast_fp16")];
            tensor<fp16, [2, ?, 32]> var_904_cast_fp16 = cos(x = angles_5_cast_fp16)[name = string("op_904_cast_fp16")];
            tensor<int32, [1]> cos_5_axes_0 = const()[name = string("cos_5_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, ?, 32]> cos_5_cast_fp16 = expand_dims(axes = cos_5_axes_0, x = var_904_cast_fp16)[name = string("cos_5_cast_fp16")];
            tensor<fp16, [2, ?, 32]> var_906_cast_fp16 = sin(x = angles_5_cast_fp16)[name = string("op_906_cast_fp16")];
            tensor<int32, [1]> sin_5_axes_0 = const()[name = string("sin_5_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, ?, 32]> sin_5_cast_fp16 = expand_dims(axes = sin_5_axes_0, x = var_906_cast_fp16)[name = string("sin_5_cast_fp16")];
            tensor<int32, [4]> x_a_5_begin_0 = const()[name = string("x_a_5_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x_a_5_end_0 = const()[name = string("x_a_5_end_0"), val = tensor<int32, [4]>([2, 8, 0, 32])];
            tensor<bool, [4]> x_a_5_end_mask_0 = const()[name = string("x_a_5_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 8, ?, 64]> x_65_cast_fp16 = transpose(perm = x_65_perm_0, x = var_866_cast_fp16)[name = string("transpose_77")];
            tensor<fp16, [2, 8, ?, 32]> x_a_5_cast_fp16 = slice_by_index(begin = x_a_5_begin_0, end = x_a_5_end_0, end_mask = x_a_5_end_mask_0, x = x_65_cast_fp16)[name = string("x_a_5_cast_fp16")];
            tensor<int32, [4]> x_b_5_begin_0 = const()[name = string("x_b_5_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x_b_5_end_0 = const()[name = string("x_b_5_end_0"), val = tensor<int32, [4]>([2, 8, 0, 64])];
            tensor<bool, [4]> x_b_5_end_mask_0 = const()[name = string("x_b_5_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [2, 8, ?, 32]> x_b_5_cast_fp16 = slice_by_index(begin = x_b_5_begin_0, end = x_b_5_end_0, end_mask = x_b_5_end_mask_0, x = x_65_cast_fp16)[name = string("x_b_5_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_910_cast_fp16 = mul(x = x_a_5_cast_fp16, y = cos_5_cast_fp16)[name = string("op_910_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_911_cast_fp16 = mul(x = x_b_5_cast_fp16, y = sin_5_cast_fp16)[name = string("op_911_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_a_5_cast_fp16 = sub(x = var_910_cast_fp16, y = var_911_cast_fp16)[name = string("rot_a_5_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_913_cast_fp16 = mul(x = x_a_5_cast_fp16, y = sin_5_cast_fp16)[name = string("op_913_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_914_cast_fp16 = mul(x = x_b_5_cast_fp16, y = cos_5_cast_fp16)[name = string("op_914_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_b_5_cast_fp16 = add(x = var_913_cast_fp16, y = var_914_cast_fp16)[name = string("rot_b_5_cast_fp16")];
            bool q_5_interleave_0 = const()[name = string("q_5_interleave_0"), val = bool(false)];
            tensor<fp16, [2, 8, ?, 64]> q_5_cast_fp16 = concat(axis = var_600, interleave = q_5_interleave_0, values = (rot_a_5_cast_fp16, rot_b_5_cast_fp16))[name = string("q_5_cast_fp16")];
            tensor<int32, [4]> x_a_7_begin_0 = const()[name = string("x_a_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x_a_7_end_0 = const()[name = string("x_a_7_end_0"), val = tensor<int32, [4]>([2, 8, 0, 32])];
            tensor<bool, [4]> x_a_7_end_mask_0 = const()[name = string("x_a_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 8, ?, 64]> x_67_cast_fp16 = transpose(perm = x_67_perm_0, x = var_874_cast_fp16)[name = string("transpose_76")];
            tensor<fp16, [2, 8, ?, 32]> x_a_7_cast_fp16 = slice_by_index(begin = x_a_7_begin_0, end = x_a_7_end_0, end_mask = x_a_7_end_mask_0, x = x_67_cast_fp16)[name = string("x_a_7_cast_fp16")];
            tensor<int32, [4]> x_b_7_begin_0 = const()[name = string("x_b_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x_b_7_end_0 = const()[name = string("x_b_7_end_0"), val = tensor<int32, [4]>([2, 8, 0, 64])];
            tensor<bool, [4]> x_b_7_end_mask_0 = const()[name = string("x_b_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [2, 8, ?, 32]> x_b_7_cast_fp16 = slice_by_index(begin = x_b_7_begin_0, end = x_b_7_end_0, end_mask = x_b_7_end_mask_0, x = x_67_cast_fp16)[name = string("x_b_7_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_928_cast_fp16 = mul(x = x_a_7_cast_fp16, y = cos_3_cast_fp16)[name = string("op_928_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_929_cast_fp16 = mul(x = x_b_7_cast_fp16, y = sin_3_cast_fp16)[name = string("op_929_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_a_7_cast_fp16 = sub(x = var_928_cast_fp16, y = var_929_cast_fp16)[name = string("rot_a_7_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_931_cast_fp16 = mul(x = x_a_7_cast_fp16, y = sin_3_cast_fp16)[name = string("op_931_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_932_cast_fp16 = mul(x = x_b_7_cast_fp16, y = cos_3_cast_fp16)[name = string("op_932_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_b_7_cast_fp16 = add(x = var_931_cast_fp16, y = var_932_cast_fp16)[name = string("rot_b_7_cast_fp16")];
            bool k_5_interleave_0 = const()[name = string("k_5_interleave_0"), val = bool(false)];
            tensor<fp16, [2, 8, ?, 64]> k_5_cast_fp16 = concat(axis = var_600, interleave = k_5_interleave_0, values = (rot_a_7_cast_fp16, rot_b_7_cast_fp16))[name = string("k_5_cast_fp16")];
            bool var_937_transpose_x_1 = const()[name = string("op_937_transpose_x_1"), val = bool(false)];
            bool var_937_transpose_y_1 = const()[name = string("op_937_transpose_y_1"), val = bool(true)];
            tensor<fp16, [2, 8, ?, ?]> var_937_cast_fp16 = matmul(transpose_x = var_937_transpose_x_1, transpose_y = var_937_transpose_y_1, x = q_5_cast_fp16, y = k_5_cast_fp16)[name = string("op_937_cast_fp16")];
            fp16 _inversed_scores_3_y_0_to_fp16 = const()[name = string("_inversed_scores_3_y_0_to_fp16"), val = fp16(0x1p-4)];
            tensor<fp16, [2, 8, ?, ?]> _inversed_scores_3_cast_fp16 = mul(x = var_937_cast_fp16, y = _inversed_scores_3_y_0_to_fp16)[name = string("_inversed_scores_3_cast_fp16")];
            tensor<fp16, [2, 8, ?, ?]> input_149_cast_fp16 = sub(x = _inversed_scores_3_cast_fp16, y = var_469_cast_fp16)[name = string("input_149_cast_fp16")];
            tensor<fp16, [2, 8, ?, ?]> attn_7_cast_fp16 = softmax(axis = var_600, x = input_149_cast_fp16)[name = string("attn_7_cast_fp16")];
            bool var_946_transpose_x_0 = const()[name = string("op_946_transpose_x_0"), val = bool(false)];
            bool var_946_transpose_y_0 = const()[name = string("op_946_transpose_y_0"), val = bool(false)];
            tensor<fp16, [2, 8, ?, 64]> v_5_cast_fp16 = transpose(perm = v_5_perm_0, x = var_882_cast_fp16)[name = string("transpose_75")];
            tensor<fp16, [2, 8, ?, 64]> var_946_cast_fp16 = matmul(transpose_x = var_946_transpose_x_0, transpose_y = var_946_transpose_y_0, x = attn_7_cast_fp16, y = v_5_cast_fp16)[name = string("op_946_cast_fp16")];
            tensor<int32, [4]> var_947_perm_0 = const()[name = string("op_947_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_21x = const()[name = string("concat_21x"), val = tensor<int32, [3]>([2, -1, 512])];
            tensor<fp16, [2, ?, 8, 64]> var_947_cast_fp16 = transpose(perm = var_947_perm_0, x = var_946_cast_fp16)[name = string("transpose_74")];
            tensor<fp16, [2, ?, 512]> input_151_cast_fp16 = reshape(shape = concat_21x, x = var_947_cast_fp16)[name = string("input_151_cast_fp16")];
            tensor<fp16, [512, 512]> main_blocks_1_text_attn_out_fc_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25146560))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25408768))))[name = string("main_blocks_1_text_attn_out_fc_linear_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_1_text_attn_out_fc_linear_bias_to_fp16 = const()[name = string("main_blocks_1_text_attn_out_fc_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25409856)))];
            tensor<fp16, [2, ?, 512]> linear_15_cast_fp16 = linear(bias = main_blocks_1_text_attn_out_fc_linear_bias_to_fp16, weight = main_blocks_1_text_attn_out_fc_linear_weight_to_fp16_quantized, x = input_151_cast_fp16)[name = string("linear_15_cast_fp16")];
            tensor<int32, [3]> out_27_perm_0 = const()[name = string("out_27_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> out_27_cast_fp16 = transpose(perm = out_27_perm_0, x = linear_15_cast_fp16)[name = string("transpose_73")];
            tensor<fp16, [2, 512, ?]> var_955_cast_fp16 = mul(x = out_27_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_955_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_69_cast_fp16 = add(x = x_m_5_cast_fp16, y = var_955_cast_fp16)[name = string("x_69_cast_fp16")];
            tensor<int32, [3]> input_153_perm_0 = const()[name = string("input_153_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_962_axes_0 = const()[name = string("op_962_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_1_text_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_1_text_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25410944)))];
            tensor<fp16, [512]> main_blocks_1_text_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_1_text_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25412032)))];
            tensor<fp16, [2, ?, 512]> input_153_cast_fp16 = transpose(perm = input_153_perm_0, x = x_69_cast_fp16)[name = string("transpose_72")];
            tensor<fp16, [2, ?, 512]> var_962_cast_fp16 = layer_norm(axes = var_962_axes_0, beta = main_blocks_1_text_norm_norm_bias_to_fp16, epsilon = var_612_to_fp16, gamma = main_blocks_1_text_norm_norm_weight_to_fp16, x = input_153_cast_fp16)[name = string("op_962_cast_fp16")];
            tensor<int32, [3]> var_963_perm_0 = const()[name = string("op_963_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> var_963_cast_fp16 = transpose(perm = var_963_perm_0, x = var_962_cast_fp16)[name = string("transpose_71")];
            tensor<fp16, [2, 512, ?]> x_71_cast_fp16 = mul(x = var_963_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_71_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_155_cast_fp16 = mul(x = x_71_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("input_155_cast_fp16")];
            tensor<int32, [6]> input_157_pad_0 = const()[name = string("input_157_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_157_mode_0 = const()[name = string("input_157_mode_0"), val = string("replicate")];
            fp16 const_12_to_fp16 = const()[name = string("const_12_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_157_cast_fp16 = pad(constant_val = const_12_to_fp16, mode = input_157_mode_0, pad = input_157_pad_0, x = input_155_cast_fp16)[name = string("input_157_cast_fp16")];
            string h_69_pad_type_0 = const()[name = string("h_69_pad_type_0"), val = string("valid")];
            int32 h_69_groups_0 = const()[name = string("h_69_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_69_strides_0 = const()[name = string("h_69_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_69_pad_0 = const()[name = string("h_69_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_69_dilations_0 = const()[name = string("h_69_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 5]> main_blocks_1_convnext_2_0_dwconv__conv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25413120))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25415744))))[name = string("main_blocks_1_convnext_2_0_dwconv__conv_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_1_convnext_2_0_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_2_0_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25416832)))];
            tensor<fp16, [2, 512, ?]> h_69_cast_fp16 = conv(bias = main_blocks_1_convnext_2_0_dwconv__conv_bias_to_fp16, dilations = h_69_dilations_0, groups = h_69_groups_0, pad = h_69_pad_0, pad_type = h_69_pad_type_0, strides = h_69_strides_0, weight = main_blocks_1_convnext_2_0_dwconv__conv_weight_to_fp16_quantized, x = input_157_cast_fp16)[name = string("h_69_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_73_cast_fp16 = mul(x = h_69_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_73_cast_fp16")];
            tensor<int32, [3]> input_159_perm_0 = const()[name = string("input_159_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_987_axes_0 = const()[name = string("op_987_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_1_convnext_2_0_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_1_convnext_2_0_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25417920)))];
            tensor<fp16, [512]> main_blocks_1_convnext_2_0_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_2_0_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25419008)))];
            tensor<fp16, [2, ?, 512]> input_159_cast_fp16 = transpose(perm = input_159_perm_0, x = x_73_cast_fp16)[name = string("transpose_70")];
            tensor<fp16, [2, ?, 512]> var_987_cast_fp16 = layer_norm(axes = var_987_axes_0, beta = main_blocks_1_convnext_2_0_norm_norm_bias_to_fp16, epsilon = var_612_to_fp16, gamma = main_blocks_1_convnext_2_0_norm_norm_weight_to_fp16, x = input_159_cast_fp16)[name = string("op_987_cast_fp16")];
            tensor<int32, [3]> input_161_perm_0 = const()[name = string("input_161_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_71_pad_type_0 = const()[name = string("h_71_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_71_strides_0 = const()[name = string("h_71_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_71_pad_0 = const()[name = string("h_71_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_71_dilations_0 = const()[name = string("h_71_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_71_groups_0 = const()[name = string("h_71_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_1_convnext_2_0_pwconv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25420096))), scale = tensor<fp16, [2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26468736))))[name = string("main_blocks_1_convnext_2_0_pwconv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> main_blocks_1_convnext_2_0_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_1_convnext_2_0_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26472896)))];
            tensor<fp16, [2, 512, ?]> input_161_cast_fp16 = transpose(perm = input_161_perm_0, x = var_987_cast_fp16)[name = string("transpose_69")];
            tensor<fp16, [2, 2048, ?]> h_71_cast_fp16 = conv(bias = main_blocks_1_convnext_2_0_pwconv1_bias_to_fp16, dilations = h_71_dilations_0, groups = h_71_groups_0, pad = h_71_pad_0, pad_type = h_71_pad_type_0, strides = h_71_strides_0, weight = main_blocks_1_convnext_2_0_pwconv1_weight_to_fp16_quantized, x = input_161_cast_fp16)[name = string("h_71_cast_fp16")];
            string input_163_mode_0 = const()[name = string("input_163_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_163_cast_fp16 = gelu(mode = input_163_mode_0, x = h_71_cast_fp16)[name = string("input_163_cast_fp16")];
            string h_73_pad_type_0 = const()[name = string("h_73_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_73_strides_0 = const()[name = string("h_73_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_73_pad_0 = const()[name = string("h_73_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_73_dilations_0 = const()[name = string("h_73_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_73_groups_0 = const()[name = string("h_73_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> op_1004_weight_0_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26477056))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27525696))))[name = string("op_1004_weight_0_to_fp16_quantized")];
            tensor<fp16, [512]> var_1004_bias_0_to_fp16 = const()[name = string("op_1004_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27526784)))];
            tensor<fp16, [2, 512, ?]> var_1004_cast_fp16 = conv(bias = var_1004_bias_0_to_fp16, dilations = h_73_dilations_0, groups = h_73_groups_0, pad = h_73_pad_0, pad_type = h_73_pad_type_0, strides = h_73_strides_0, weight = op_1004_weight_0_to_fp16_quantized, x = input_163_cast_fp16)[name = string("op_1004_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_29_cast_fp16 = add(x = input_155_cast_fp16, y = var_1004_cast_fp16)[name = string("out_29_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_75_cast_fp16 = mul(x = out_29_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_75_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_m_7_cast_fp16 = mul(x = x_75_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_m_7_cast_fp16")];
            tensor<int32, [3]> input_165_perm_0 = const()[name = string("input_165_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [256, 512]> main_blocks_1_style_attn_W_query_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [256, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27527872))), scale = tensor<fp16, [256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27659008))))[name = string("main_blocks_1_style_attn_W_query_linear_weight_to_fp16_quantized")];
            tensor<fp16, [256]> main_blocks_1_style_attn_W_query_linear_bias_to_fp16 = const()[name = string("main_blocks_1_style_attn_W_query_linear_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27659584)))];
            tensor<fp16, [2, ?, 512]> input_165_cast_fp16 = transpose(perm = input_165_perm_0, x = x_75_cast_fp16)[name = string("transpose_68")];
            tensor<fp16, [2, ?, 256]> linear_16_cast_fp16 = linear(bias = main_blocks_1_style_attn_W_query_linear_bias_to_fp16, weight = main_blocks_1_style_attn_W_query_linear_weight_to_fp16_quantized, x = input_165_cast_fp16)[name = string("linear_16_cast_fp16")];
            tensor<int32, [4]> concat_22x = const()[name = string("concat_22x"), val = tensor<int32, [4]>([2, -1, 2, 128])];
            tensor<fp16, [2, ?, 2, 128]> var_1021_cast_fp16 = reshape(shape = concat_22x, x = linear_16_cast_fp16)[name = string("op_1021_cast_fp16")];
            tensor<int32, [4]> q_7_perm_0 = const()[name = string("q_7_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [256, 256]> main_blocks_1_style_attn_W_value_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [256, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27660160))), scale = tensor<fp16, [256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27725760))))[name = string("main_blocks_1_style_attn_W_value_linear_weight_to_fp16_quantized")];
            tensor<fp16, [256]> main_blocks_1_style_attn_W_value_linear_bias_to_fp16 = const()[name = string("main_blocks_1_style_attn_W_value_linear_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27726336)))];
            tensor<fp16, [2, 50, 256]> linear_18_cast_fp16 = linear(bias = main_blocks_1_style_attn_W_value_linear_bias_to_fp16, weight = main_blocks_1_style_attn_W_value_linear_weight_to_fp16_quantized, x = input_83_cast_fp16)[name = string("linear_18_cast_fp16")];
            tensor<int32, [4]> var_1034 = const()[name = string("op_1034"), val = tensor<int32, [4]>([2, 50, 2, 128])];
            tensor<fp16, [2, 50, 2, 128]> var_1035_cast_fp16 = reshape(shape = var_1034, x = linear_18_cast_fp16)[name = string("op_1035_cast_fp16")];
            tensor<int32, [4]> v_7_perm_0 = const()[name = string("v_7_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool var_1039_transpose_x_0 = const()[name = string("op_1039_transpose_x_0"), val = bool(false)];
            bool var_1039_transpose_y_0 = const()[name = string("op_1039_transpose_y_0"), val = bool(false)];
            tensor<fp16, [2, 2, 128, 50]> op_1038_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2, 2, 128, 50]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27726912))), scale = tensor<fp16, [1, 1, 1, 50]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27752576))))[name = string("op_1038_to_fp16_quantized")];
            tensor<fp16, [2, 2, ?, 128]> q_7_cast_fp16 = transpose(perm = q_7_perm_0, x = var_1021_cast_fp16)[name = string("transpose_67")];
            tensor<fp16, [2, 2, ?, 50]> var_1039_cast_fp16 = matmul(transpose_x = var_1039_transpose_x_0, transpose_y = var_1039_transpose_y_0, x = q_7_cast_fp16, y = op_1038_to_fp16_quantized)[name = string("op_1039_cast_fp16")];
            fp16 _inversed_input_167_y_0_to_fp16 = const()[name = string("_inversed_input_167_y_0_to_fp16"), val = fp16(0x1p-4)];
            tensor<fp16, [2, 2, ?, 50]> _inversed_input_167_cast_fp16 = mul(x = var_1039_cast_fp16, y = _inversed_input_167_y_0_to_fp16)[name = string("_inversed_input_167_cast_fp16")];
            tensor<fp16, [2, 2, ?, 50]> attn_9_cast_fp16 = softmax(axis = var_600, x = _inversed_input_167_cast_fp16)[name = string("attn_9_cast_fp16")];
            tensor<fp16, [2, 2, ?, 50]> attn_11_cast_fp16 = mul(x = attn_9_cast_fp16, y = mask_1_cast_fp16)[name = string("attn_11_cast_fp16")];
            bool var_1046_transpose_x_0 = const()[name = string("op_1046_transpose_x_0"), val = bool(false)];
            bool var_1046_transpose_y_0 = const()[name = string("op_1046_transpose_y_0"), val = bool(false)];
            tensor<fp16, [2, 2, 50, 128]> v_7_cast_fp16 = transpose(perm = v_7_perm_0, x = var_1035_cast_fp16)[name = string("transpose_66")];
            tensor<fp16, [2, 2, ?, 128]> var_1046_cast_fp16 = matmul(transpose_x = var_1046_transpose_x_0, transpose_y = var_1046_transpose_y_0, x = attn_11_cast_fp16, y = v_7_cast_fp16)[name = string("op_1046_cast_fp16")];
            tensor<int32, [4]> var_1047_perm_0 = const()[name = string("op_1047_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_23x = const()[name = string("concat_23x"), val = tensor<int32, [3]>([2, -1, 256])];
            tensor<fp16, [2, ?, 2, 128]> var_1047_cast_fp16 = transpose(perm = var_1047_perm_0, x = var_1046_cast_fp16)[name = string("transpose_65")];
            tensor<fp16, [2, ?, 256]> input_169_cast_fp16 = reshape(shape = concat_23x, x = var_1047_cast_fp16)[name = string("input_169_cast_fp16")];
            tensor<fp16, [512, 256]> main_blocks_1_style_attn_out_fc_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27752768))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27883904))))[name = string("main_blocks_1_style_attn_out_fc_linear_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_1_style_attn_out_fc_linear_bias_to_fp16 = const()[name = string("main_blocks_1_style_attn_out_fc_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27884992)))];
            tensor<fp16, [2, ?, 512]> linear_19_cast_fp16 = linear(bias = main_blocks_1_style_attn_out_fc_linear_bias_to_fp16, weight = main_blocks_1_style_attn_out_fc_linear_weight_to_fp16_quantized, x = input_169_cast_fp16)[name = string("linear_19_cast_fp16")];
            tensor<int32, [3]> out_31_perm_0 = const()[name = string("out_31_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> out_31_cast_fp16 = transpose(perm = out_31_perm_0, x = linear_19_cast_fp16)[name = string("transpose_64")];
            tensor<fp16, [2, 512, ?]> var_1055_cast_fp16 = mul(x = out_31_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_1055_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_77_cast_fp16 = add(x = x_m_7_cast_fp16, y = var_1055_cast_fp16)[name = string("x_77_cast_fp16")];
            tensor<int32, [3]> input_171_perm_0 = const()[name = string("input_171_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1062_axes_0 = const()[name = string("op_1062_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_1_style_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_1_style_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27886080)))];
            tensor<fp16, [512]> main_blocks_1_style_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_1_style_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27887168)))];
            tensor<fp16, [2, ?, 512]> input_171_cast_fp16 = transpose(perm = input_171_perm_0, x = x_77_cast_fp16)[name = string("transpose_63")];
            tensor<fp16, [2, ?, 512]> var_1062_cast_fp16 = layer_norm(axes = var_1062_axes_0, beta = main_blocks_1_style_norm_norm_bias_to_fp16, epsilon = var_612_to_fp16, gamma = main_blocks_1_style_norm_norm_weight_to_fp16, x = input_171_cast_fp16)[name = string("op_1062_cast_fp16")];
            tensor<int32, [3]> var_1063_perm_0 = const()[name = string("op_1063_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> var_1063_cast_fp16 = transpose(perm = var_1063_perm_0, x = var_1062_cast_fp16)[name = string("transpose_62")];
            tensor<fp16, [2, 512, ?]> x_79_cast_fp16 = mul(x = var_1063_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_79_cast_fp16")];
            int32 var_1074 = const()[name = string("op_1074"), val = int32(-1)];
            tensor<fp16, [2, 512, ?]> input_173_cast_fp16 = mul(x = x_79_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("input_173_cast_fp16")];
            tensor<int32, [6]> input_175_pad_0 = const()[name = string("input_175_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_175_mode_0 = const()[name = string("input_175_mode_0"), val = string("replicate")];
            fp16 const_14_to_fp16 = const()[name = string("const_14_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_175_cast_fp16 = pad(constant_val = const_14_to_fp16, mode = input_175_mode_0, pad = input_175_pad_0, x = input_173_cast_fp16)[name = string("input_175_cast_fp16")];
            string h_75_pad_type_0 = const()[name = string("h_75_pad_type_0"), val = string("valid")];
            int32 h_75_groups_0 = const()[name = string("h_75_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_75_strides_0 = const()[name = string("h_75_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_75_pad_0 = const()[name = string("h_75_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_75_dilations_0 = const()[name = string("h_75_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 5]> main_blocks_2_convnext_0_0_dwconv__conv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27888256))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27890880))))[name = string("main_blocks_2_convnext_0_0_dwconv__conv_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_2_convnext_0_0_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_0_0_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27891968)))];
            tensor<fp16, [2, 512, ?]> h_75_cast_fp16 = conv(bias = main_blocks_2_convnext_0_0_dwconv__conv_bias_to_fp16, dilations = h_75_dilations_0, groups = h_75_groups_0, pad = h_75_pad_0, pad_type = h_75_pad_type_0, strides = h_75_strides_0, weight = main_blocks_2_convnext_0_0_dwconv__conv_weight_to_fp16_quantized, x = input_175_cast_fp16)[name = string("h_75_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_81_cast_fp16 = mul(x = h_75_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_81_cast_fp16")];
            tensor<int32, [3]> input_177_perm_0 = const()[name = string("input_177_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1127_axes_0 = const()[name = string("op_1127_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_2_convnext_0_0_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_2_convnext_0_0_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27893056)))];
            tensor<fp16, [512]> main_blocks_2_convnext_0_0_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_0_0_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27894144)))];
            fp16 var_1086_to_fp16 = const()[name = string("op_1086_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [2, ?, 512]> input_177_cast_fp16 = transpose(perm = input_177_perm_0, x = x_81_cast_fp16)[name = string("transpose_61")];
            tensor<fp16, [2, ?, 512]> var_1127_cast_fp16 = layer_norm(axes = var_1127_axes_0, beta = main_blocks_2_convnext_0_0_norm_norm_bias_to_fp16, epsilon = var_1086_to_fp16, gamma = main_blocks_2_convnext_0_0_norm_norm_weight_to_fp16, x = input_177_cast_fp16)[name = string("op_1127_cast_fp16")];
            tensor<int32, [3]> input_179_perm_0 = const()[name = string("input_179_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_77_pad_type_0 = const()[name = string("h_77_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_77_strides_0 = const()[name = string("h_77_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_77_pad_0 = const()[name = string("h_77_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_77_dilations_0 = const()[name = string("h_77_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_77_groups_0 = const()[name = string("h_77_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_2_convnext_0_0_pwconv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27895232))), scale = tensor<fp16, [2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28943872))))[name = string("main_blocks_2_convnext_0_0_pwconv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> main_blocks_2_convnext_0_0_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_0_0_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28948032)))];
            tensor<fp16, [2, 512, ?]> input_179_cast_fp16 = transpose(perm = input_179_perm_0, x = var_1127_cast_fp16)[name = string("transpose_60")];
            tensor<fp16, [2, 2048, ?]> h_77_cast_fp16 = conv(bias = main_blocks_2_convnext_0_0_pwconv1_bias_to_fp16, dilations = h_77_dilations_0, groups = h_77_groups_0, pad = h_77_pad_0, pad_type = h_77_pad_type_0, strides = h_77_strides_0, weight = main_blocks_2_convnext_0_0_pwconv1_weight_to_fp16_quantized, x = input_179_cast_fp16)[name = string("h_77_cast_fp16")];
            string input_181_mode_0 = const()[name = string("input_181_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_181_cast_fp16 = gelu(mode = input_181_mode_0, x = h_77_cast_fp16)[name = string("input_181_cast_fp16")];
            string h_79_pad_type_0 = const()[name = string("h_79_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_79_strides_0 = const()[name = string("h_79_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_79_pad_0 = const()[name = string("h_79_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_79_dilations_0 = const()[name = string("h_79_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_79_groups_0 = const()[name = string("h_79_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> op_1144_weight_0_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28952192))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30000832))))[name = string("op_1144_weight_0_to_fp16_quantized")];
            tensor<fp16, [512]> var_1144_bias_0_to_fp16 = const()[name = string("op_1144_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30001920)))];
            tensor<fp16, [2, 512, ?]> var_1144_cast_fp16 = conv(bias = var_1144_bias_0_to_fp16, dilations = h_79_dilations_0, groups = h_79_groups_0, pad = h_79_pad_0, pad_type = h_79_pad_type_0, strides = h_79_strides_0, weight = op_1144_weight_0_to_fp16_quantized, x = input_181_cast_fp16)[name = string("op_1144_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_33_cast_fp16 = add(x = input_173_cast_fp16, y = var_1144_cast_fp16)[name = string("out_33_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_83_cast_fp16 = mul(x = out_33_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_83_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_183_cast_fp16 = mul(x = x_83_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("input_183_cast_fp16")];
            tensor<int32, [6]> input_185_pad_0 = const()[name = string("input_185_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_185_mode_0 = const()[name = string("input_185_mode_0"), val = string("replicate")];
            fp16 const_15_to_fp16 = const()[name = string("const_15_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_185_cast_fp16 = pad(constant_val = const_15_to_fp16, mode = input_185_mode_0, pad = input_185_pad_0, x = input_183_cast_fp16)[name = string("input_185_cast_fp16")];
            string h_81_pad_type_0 = const()[name = string("h_81_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_81_dilations_0 = const()[name = string("h_81_dilations_0"), val = tensor<int32, [1]>([2])];
            int32 h_81_groups_0 = const()[name = string("h_81_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_81_strides_0 = const()[name = string("h_81_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_81_pad_0 = const()[name = string("h_81_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [512, 1, 5]> main_blocks_2_convnext_0_1_dwconv__conv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30003008))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30005632))))[name = string("main_blocks_2_convnext_0_1_dwconv__conv_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_2_convnext_0_1_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_0_1_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30006720)))];
            tensor<fp16, [2, 512, ?]> h_81_cast_fp16 = conv(bias = main_blocks_2_convnext_0_1_dwconv__conv_bias_to_fp16, dilations = h_81_dilations_0, groups = h_81_groups_0, pad = h_81_pad_0, pad_type = h_81_pad_type_0, strides = h_81_strides_0, weight = main_blocks_2_convnext_0_1_dwconv__conv_weight_to_fp16_quantized, x = input_185_cast_fp16)[name = string("h_81_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_85_cast_fp16 = mul(x = h_81_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_85_cast_fp16")];
            tensor<int32, [3]> input_187_perm_0 = const()[name = string("input_187_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1169_axes_0 = const()[name = string("op_1169_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_2_convnext_0_1_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_2_convnext_0_1_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30007808)))];
            tensor<fp16, [512]> main_blocks_2_convnext_0_1_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_0_1_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30008896)))];
            tensor<fp16, [2, ?, 512]> input_187_cast_fp16 = transpose(perm = input_187_perm_0, x = x_85_cast_fp16)[name = string("transpose_59")];
            tensor<fp16, [2, ?, 512]> var_1169_cast_fp16 = layer_norm(axes = var_1169_axes_0, beta = main_blocks_2_convnext_0_1_norm_norm_bias_to_fp16, epsilon = var_1086_to_fp16, gamma = main_blocks_2_convnext_0_1_norm_norm_weight_to_fp16, x = input_187_cast_fp16)[name = string("op_1169_cast_fp16")];
            tensor<int32, [3]> input_189_perm_0 = const()[name = string("input_189_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_83_pad_type_0 = const()[name = string("h_83_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_83_strides_0 = const()[name = string("h_83_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_83_pad_0 = const()[name = string("h_83_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_83_dilations_0 = const()[name = string("h_83_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_83_groups_0 = const()[name = string("h_83_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_2_convnext_0_1_pwconv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30009984))), scale = tensor<fp16, [2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31058624))))[name = string("main_blocks_2_convnext_0_1_pwconv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> main_blocks_2_convnext_0_1_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_0_1_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31062784)))];
            tensor<fp16, [2, 512, ?]> input_189_cast_fp16 = transpose(perm = input_189_perm_0, x = var_1169_cast_fp16)[name = string("transpose_58")];
            tensor<fp16, [2, 2048, ?]> h_83_cast_fp16 = conv(bias = main_blocks_2_convnext_0_1_pwconv1_bias_to_fp16, dilations = h_83_dilations_0, groups = h_83_groups_0, pad = h_83_pad_0, pad_type = h_83_pad_type_0, strides = h_83_strides_0, weight = main_blocks_2_convnext_0_1_pwconv1_weight_to_fp16_quantized, x = input_189_cast_fp16)[name = string("h_83_cast_fp16")];
            string input_191_mode_0 = const()[name = string("input_191_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_191_cast_fp16 = gelu(mode = input_191_mode_0, x = h_83_cast_fp16)[name = string("input_191_cast_fp16")];
            string h_85_pad_type_0 = const()[name = string("h_85_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_85_strides_0 = const()[name = string("h_85_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_85_pad_0 = const()[name = string("h_85_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_85_dilations_0 = const()[name = string("h_85_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_85_groups_0 = const()[name = string("h_85_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> op_1186_weight_0_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31066944))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32115584))))[name = string("op_1186_weight_0_to_fp16_quantized")];
            tensor<fp16, [512]> var_1186_bias_0_to_fp16 = const()[name = string("op_1186_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32116672)))];
            tensor<fp16, [2, 512, ?]> var_1186_cast_fp16 = conv(bias = var_1186_bias_0_to_fp16, dilations = h_85_dilations_0, groups = h_85_groups_0, pad = h_85_pad_0, pad_type = h_85_pad_type_0, strides = h_85_strides_0, weight = op_1186_weight_0_to_fp16_quantized, x = input_191_cast_fp16)[name = string("op_1186_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_35_cast_fp16 = add(x = input_183_cast_fp16, y = var_1186_cast_fp16)[name = string("out_35_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_87_cast_fp16 = mul(x = out_35_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_87_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_193_cast_fp16 = mul(x = x_87_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("input_193_cast_fp16")];
            tensor<int32, [6]> input_195_pad_0 = const()[name = string("input_195_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 8, 8])];
            string input_195_mode_0 = const()[name = string("input_195_mode_0"), val = string("replicate")];
            fp16 const_16_to_fp16 = const()[name = string("const_16_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_195_cast_fp16 = pad(constant_val = const_16_to_fp16, mode = input_195_mode_0, pad = input_195_pad_0, x = input_193_cast_fp16)[name = string("input_195_cast_fp16")];
            string h_87_pad_type_0 = const()[name = string("h_87_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_87_dilations_0 = const()[name = string("h_87_dilations_0"), val = tensor<int32, [1]>([4])];
            int32 h_87_groups_0 = const()[name = string("h_87_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_87_strides_0 = const()[name = string("h_87_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_87_pad_0 = const()[name = string("h_87_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [512, 1, 5]> main_blocks_2_convnext_0_2_dwconv__conv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32117760))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32120384))))[name = string("main_blocks_2_convnext_0_2_dwconv__conv_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_2_convnext_0_2_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_0_2_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32121472)))];
            tensor<fp16, [2, 512, ?]> h_87_cast_fp16 = conv(bias = main_blocks_2_convnext_0_2_dwconv__conv_bias_to_fp16, dilations = h_87_dilations_0, groups = h_87_groups_0, pad = h_87_pad_0, pad_type = h_87_pad_type_0, strides = h_87_strides_0, weight = main_blocks_2_convnext_0_2_dwconv__conv_weight_to_fp16_quantized, x = input_195_cast_fp16)[name = string("h_87_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_89_cast_fp16 = mul(x = h_87_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_89_cast_fp16")];
            tensor<int32, [3]> input_197_perm_0 = const()[name = string("input_197_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1211_axes_0 = const()[name = string("op_1211_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_2_convnext_0_2_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_2_convnext_0_2_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32122560)))];
            tensor<fp16, [512]> main_blocks_2_convnext_0_2_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_0_2_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32123648)))];
            tensor<fp16, [2, ?, 512]> input_197_cast_fp16 = transpose(perm = input_197_perm_0, x = x_89_cast_fp16)[name = string("transpose_57")];
            tensor<fp16, [2, ?, 512]> var_1211_cast_fp16 = layer_norm(axes = var_1211_axes_0, beta = main_blocks_2_convnext_0_2_norm_norm_bias_to_fp16, epsilon = var_1086_to_fp16, gamma = main_blocks_2_convnext_0_2_norm_norm_weight_to_fp16, x = input_197_cast_fp16)[name = string("op_1211_cast_fp16")];
            tensor<int32, [3]> input_199_perm_0 = const()[name = string("input_199_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_89_pad_type_0 = const()[name = string("h_89_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_89_strides_0 = const()[name = string("h_89_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_89_pad_0 = const()[name = string("h_89_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_89_dilations_0 = const()[name = string("h_89_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_89_groups_0 = const()[name = string("h_89_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_2_convnext_0_2_pwconv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32124736))), scale = tensor<fp16, [2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33173376))))[name = string("main_blocks_2_convnext_0_2_pwconv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> main_blocks_2_convnext_0_2_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_0_2_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33177536)))];
            tensor<fp16, [2, 512, ?]> input_199_cast_fp16 = transpose(perm = input_199_perm_0, x = var_1211_cast_fp16)[name = string("transpose_56")];
            tensor<fp16, [2, 2048, ?]> h_89_cast_fp16 = conv(bias = main_blocks_2_convnext_0_2_pwconv1_bias_to_fp16, dilations = h_89_dilations_0, groups = h_89_groups_0, pad = h_89_pad_0, pad_type = h_89_pad_type_0, strides = h_89_strides_0, weight = main_blocks_2_convnext_0_2_pwconv1_weight_to_fp16_quantized, x = input_199_cast_fp16)[name = string("h_89_cast_fp16")];
            string input_201_mode_0 = const()[name = string("input_201_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_201_cast_fp16 = gelu(mode = input_201_mode_0, x = h_89_cast_fp16)[name = string("input_201_cast_fp16")];
            string h_91_pad_type_0 = const()[name = string("h_91_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_91_strides_0 = const()[name = string("h_91_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_91_pad_0 = const()[name = string("h_91_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_91_dilations_0 = const()[name = string("h_91_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_91_groups_0 = const()[name = string("h_91_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> op_1228_weight_0_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33181696))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34230336))))[name = string("op_1228_weight_0_to_fp16_quantized")];
            tensor<fp16, [512]> var_1228_bias_0_to_fp16 = const()[name = string("op_1228_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34231424)))];
            tensor<fp16, [2, 512, ?]> var_1228_cast_fp16 = conv(bias = var_1228_bias_0_to_fp16, dilations = h_91_dilations_0, groups = h_91_groups_0, pad = h_91_pad_0, pad_type = h_91_pad_type_0, strides = h_91_strides_0, weight = op_1228_weight_0_to_fp16_quantized, x = input_201_cast_fp16)[name = string("op_1228_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_37_cast_fp16 = add(x = input_193_cast_fp16, y = var_1228_cast_fp16)[name = string("out_37_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_91_cast_fp16 = mul(x = out_37_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_91_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_203_cast_fp16 = mul(x = x_91_cast_fp16, y = latent_mask_b_cast_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, 16, 16])];
            string input_205_mode_0 = const()[name = string("input_205_mode_0"), val = string("replicate")];
            fp16 const_17_to_fp16 = const()[name = string("const_17_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_205_cast_fp16 = pad(constant_val = const_17_to_fp16, mode = input_205_mode_0, pad = input_205_pad_0, x = input_203_cast_fp16)[name = string("input_205_cast_fp16")];
            string h_93_pad_type_0 = const()[name = string("h_93_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_93_dilations_0 = const()[name = string("h_93_dilations_0"), val = tensor<int32, [1]>([8])];
            int32 h_93_groups_0 = const()[name = string("h_93_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_93_strides_0 = const()[name = string("h_93_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_93_pad_0 = const()[name = string("h_93_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [512, 1, 5]> main_blocks_2_convnext_0_3_dwconv__conv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34232512))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34235136))))[name = string("main_blocks_2_convnext_0_3_dwconv__conv_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_2_convnext_0_3_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_0_3_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34236224)))];
            tensor<fp16, [2, 512, ?]> h_93_cast_fp16 = conv(bias = main_blocks_2_convnext_0_3_dwconv__conv_bias_to_fp16, dilations = h_93_dilations_0, groups = h_93_groups_0, pad = h_93_pad_0, pad_type = h_93_pad_type_0, strides = h_93_strides_0, weight = main_blocks_2_convnext_0_3_dwconv__conv_weight_to_fp16_quantized, x = input_205_cast_fp16)[name = string("h_93_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_93_cast_fp16 = mul(x = h_93_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_93_cast_fp16")];
            tensor<int32, [3]> input_207_perm_0 = const()[name = string("input_207_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1253_axes_0 = const()[name = string("op_1253_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_2_convnext_0_3_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_2_convnext_0_3_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34237312)))];
            tensor<fp16, [512]> main_blocks_2_convnext_0_3_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_0_3_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34238400)))];
            tensor<fp16, [2, ?, 512]> input_207_cast_fp16 = transpose(perm = input_207_perm_0, x = x_93_cast_fp16)[name = string("transpose_55")];
            tensor<fp16, [2, ?, 512]> var_1253_cast_fp16 = layer_norm(axes = var_1253_axes_0, beta = main_blocks_2_convnext_0_3_norm_norm_bias_to_fp16, epsilon = var_1086_to_fp16, gamma = main_blocks_2_convnext_0_3_norm_norm_weight_to_fp16, x = input_207_cast_fp16)[name = string("op_1253_cast_fp16")];
            tensor<int32, [3]> input_209_perm_0 = const()[name = string("input_209_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_95_pad_type_0 = const()[name = string("h_95_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_95_strides_0 = const()[name = string("h_95_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_95_pad_0 = const()[name = string("h_95_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_95_dilations_0 = const()[name = string("h_95_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_95_groups_0 = const()[name = string("h_95_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_2_convnext_0_3_pwconv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34239488))), scale = tensor<fp16, [2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35288128))))[name = string("main_blocks_2_convnext_0_3_pwconv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> main_blocks_2_convnext_0_3_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_0_3_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35292288)))];
            tensor<fp16, [2, 512, ?]> input_209_cast_fp16 = transpose(perm = input_209_perm_0, x = var_1253_cast_fp16)[name = string("transpose_54")];
            tensor<fp16, [2, 2048, ?]> h_95_cast_fp16 = conv(bias = main_blocks_2_convnext_0_3_pwconv1_bias_to_fp16, dilations = h_95_dilations_0, groups = h_95_groups_0, pad = h_95_pad_0, pad_type = h_95_pad_type_0, strides = h_95_strides_0, weight = main_blocks_2_convnext_0_3_pwconv1_weight_to_fp16_quantized, x = input_209_cast_fp16)[name = string("h_95_cast_fp16")];
            string input_211_mode_0 = const()[name = string("input_211_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_211_cast_fp16 = gelu(mode = input_211_mode_0, x = h_95_cast_fp16)[name = string("input_211_cast_fp16")];
            string h_97_pad_type_0 = const()[name = string("h_97_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_97_strides_0 = const()[name = string("h_97_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_97_pad_0 = const()[name = string("h_97_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_97_dilations_0 = const()[name = string("h_97_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_97_groups_0 = const()[name = string("h_97_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> op_1270_weight_0_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35296448))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36345088))))[name = string("op_1270_weight_0_to_fp16_quantized")];
            tensor<fp16, [512]> var_1270_bias_0_to_fp16 = const()[name = string("op_1270_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36346176)))];
            tensor<fp16, [2, 512, ?]> var_1270_cast_fp16 = conv(bias = var_1270_bias_0_to_fp16, dilations = h_97_dilations_0, groups = h_97_groups_0, pad = h_97_pad_0, pad_type = h_97_pad_type_0, strides = h_97_strides_0, weight = op_1270_weight_0_to_fp16_quantized, x = input_211_cast_fp16)[name = string("op_1270_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_39_cast_fp16 = add(x = input_203_cast_fp16, y = var_1270_cast_fp16)[name = string("out_39_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_95_cast_fp16 = mul(x = out_39_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_95_cast_fp16")];
            tensor<fp16, [512, 64]> main_blocks_2_time_cond_linear_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36347264))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36380096))))[name = string("main_blocks_2_time_cond_linear_linear_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_2_time_cond_linear_linear_bias_to_fp16 = const()[name = string("main_blocks_2_time_cond_linear_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36381184)))];
            tensor<fp16, [2, 512]> linear_20_cast_fp16 = linear(bias = main_blocks_2_time_cond_linear_linear_bias_to_fp16, weight = main_blocks_2_time_cond_linear_linear_weight_to_fp16_quantized, x = input_47_cast_fp16)[name = string("linear_20_cast_fp16")];
            tensor<int32, [1]> t_9_axes_0 = const()[name = string("t_9_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2, 512, 1]> t_9_cast_fp16 = expand_dims(axes = t_9_axes_0, x = linear_20_cast_fp16)[name = string("t_9_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1280_cast_fp16 = add(x = x_95_cast_fp16, y = t_9_cast_fp16)[name = string("op_1280_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_97_cast_fp16 = mul(x = var_1280_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_97_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_215_cast_fp16 = mul(x = x_97_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("input_215_cast_fp16")];
            tensor<int32, [6]> input_217_pad_0 = const()[name = string("input_217_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_217_mode_0 = const()[name = string("input_217_mode_0"), val = string("replicate")];
            fp16 const_18_to_fp16 = const()[name = string("const_18_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_217_cast_fp16 = pad(constant_val = const_18_to_fp16, mode = input_217_mode_0, pad = input_217_pad_0, x = input_215_cast_fp16)[name = string("input_217_cast_fp16")];
            string h_99_pad_type_0 = const()[name = string("h_99_pad_type_0"), val = string("valid")];
            int32 h_99_groups_0 = const()[name = string("h_99_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_99_strides_0 = const()[name = string("h_99_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_99_pad_0 = const()[name = string("h_99_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_99_dilations_0 = const()[name = string("h_99_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 5]> main_blocks_2_convnext_1_0_dwconv__conv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36382272))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36384896))))[name = string("main_blocks_2_convnext_1_0_dwconv__conv_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_2_convnext_1_0_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_1_0_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36385984)))];
            tensor<fp16, [2, 512, ?]> h_99_cast_fp16 = conv(bias = main_blocks_2_convnext_1_0_dwconv__conv_bias_to_fp16, dilations = h_99_dilations_0, groups = h_99_groups_0, pad = h_99_pad_0, pad_type = h_99_pad_type_0, strides = h_99_strides_0, weight = main_blocks_2_convnext_1_0_dwconv__conv_weight_to_fp16_quantized, x = input_217_cast_fp16)[name = string("h_99_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_99_cast_fp16 = mul(x = h_99_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_99_cast_fp16")];
            tensor<int32, [3]> input_219_perm_0 = const()[name = string("input_219_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1304_axes_0 = const()[name = string("op_1304_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_2_convnext_1_0_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_2_convnext_1_0_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36387072)))];
            tensor<fp16, [512]> main_blocks_2_convnext_1_0_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_1_0_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36388160)))];
            tensor<fp16, [2, ?, 512]> input_219_cast_fp16 = transpose(perm = input_219_perm_0, x = x_99_cast_fp16)[name = string("transpose_53")];
            tensor<fp16, [2, ?, 512]> var_1304_cast_fp16 = layer_norm(axes = var_1304_axes_0, beta = main_blocks_2_convnext_1_0_norm_norm_bias_to_fp16, epsilon = var_1086_to_fp16, gamma = main_blocks_2_convnext_1_0_norm_norm_weight_to_fp16, x = input_219_cast_fp16)[name = string("op_1304_cast_fp16")];
            tensor<int32, [3]> input_221_perm_0 = const()[name = string("input_221_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_101_pad_type_0 = const()[name = string("h_101_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_101_strides_0 = const()[name = string("h_101_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_101_pad_0 = const()[name = string("h_101_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_101_dilations_0 = const()[name = string("h_101_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_101_groups_0 = const()[name = string("h_101_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_2_convnext_1_0_pwconv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36389248))), scale = tensor<fp16, [2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37437888))))[name = string("main_blocks_2_convnext_1_0_pwconv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> main_blocks_2_convnext_1_0_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_1_0_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37442048)))];
            tensor<fp16, [2, 512, ?]> input_221_cast_fp16 = transpose(perm = input_221_perm_0, x = var_1304_cast_fp16)[name = string("transpose_52")];
            tensor<fp16, [2, 2048, ?]> h_101_cast_fp16 = conv(bias = main_blocks_2_convnext_1_0_pwconv1_bias_to_fp16, dilations = h_101_dilations_0, groups = h_101_groups_0, pad = h_101_pad_0, pad_type = h_101_pad_type_0, strides = h_101_strides_0, weight = main_blocks_2_convnext_1_0_pwconv1_weight_to_fp16_quantized, x = input_221_cast_fp16)[name = string("h_101_cast_fp16")];
            string input_223_mode_0 = const()[name = string("input_223_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_223_cast_fp16 = gelu(mode = input_223_mode_0, x = h_101_cast_fp16)[name = string("input_223_cast_fp16")];
            string h_103_pad_type_0 = const()[name = string("h_103_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_103_strides_0 = const()[name = string("h_103_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_103_pad_0 = const()[name = string("h_103_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_103_dilations_0 = const()[name = string("h_103_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_103_groups_0 = const()[name = string("h_103_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> op_1321_weight_0_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37446208))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38494848))))[name = string("op_1321_weight_0_to_fp16_quantized")];
            tensor<fp16, [512]> var_1321_bias_0_to_fp16 = const()[name = string("op_1321_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38495936)))];
            tensor<fp16, [2, 512, ?]> var_1321_cast_fp16 = conv(bias = var_1321_bias_0_to_fp16, dilations = h_103_dilations_0, groups = h_103_groups_0, pad = h_103_pad_0, pad_type = h_103_pad_type_0, strides = h_103_strides_0, weight = op_1321_weight_0_to_fp16_quantized, x = input_223_cast_fp16)[name = string("op_1321_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_41_cast_fp16 = add(x = input_215_cast_fp16, y = var_1321_cast_fp16)[name = string("out_41_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_101_cast_fp16 = mul(x = out_41_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_101_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_m_9_cast_fp16 = mul(x = x_101_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_m_9_cast_fp16")];
            tensor<int32, [3]> var_1331_shape_cast_fp16 = shape(x = x_101_cast_fp16)[name = string("op_1331_shape_cast_fp16")];
            int32 gather_14_axis_0 = const()[name = string("gather_14_axis_0"), val = int32(0)];
            int32 gather_14_batch_dims_0 = const()[name = string("gather_14_batch_dims_0"), val = int32(0)];
            bool gather_14_validate_indices_0 = const()[name = string("gather_14_validate_indices_0"), val = bool(false)];
            string var_1331_shape_cast_fp16_to_uint16_dtype_0 = const()[name = string("op_1331_shape_cast_fp16_to_uint16_dtype_0"), val = string("uint16")];
            uint16 gather_14_indices_0_to_uint16 = const()[name = string("gather_14_indices_0_to_uint16"), val = uint16(2)];
            tensor<uint16, [3]> var_1331_shape_cast_fp16_to_uint16 = cast(dtype = var_1331_shape_cast_fp16_to_uint16_dtype_0, x = var_1331_shape_cast_fp16)[name = string("cast_4")];
            uint16 gather_14_cast_uint16 = gather(axis = gather_14_axis_0, batch_dims = gather_14_batch_dims_0, indices = gather_14_indices_0_to_uint16, validate_indices = gather_14_validate_indices_0, x = var_1331_shape_cast_fp16_to_uint16)[name = string("gather_14_cast_uint16")];
            string gather_14_cast_uint16_to_int32_dtype_0 = const()[name = string("gather_14_cast_uint16_to_int32_dtype_0"), val = string("int32")];
            tensor<int32, [3]> input_225_perm_0 = const()[name = string("input_225_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [512, 512]> main_blocks_2_text_attn_W_query_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38497024))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38759232))))[name = string("main_blocks_2_text_attn_W_query_linear_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_2_text_attn_W_query_linear_bias_to_fp16 = const()[name = string("main_blocks_2_text_attn_W_query_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38760320)))];
            tensor<fp16, [2, ?, 512]> input_225_cast_fp16 = transpose(perm = input_225_perm_0, x = x_101_cast_fp16)[name = string("transpose_51")];
            tensor<fp16, [2, ?, 512]> linear_21_cast_fp16 = linear(bias = main_blocks_2_text_attn_W_query_linear_bias_to_fp16, weight = main_blocks_2_text_attn_W_query_linear_weight_to_fp16_quantized, x = input_225_cast_fp16)[name = string("linear_21_cast_fp16")];
            tensor<int32, [4]> concat_24x = const()[name = string("concat_24x"), val = tensor<int32, [4]>([2, -1, 8, 64])];
            tensor<fp16, [2, ?, 8, 64]> var_1340_cast_fp16 = reshape(shape = concat_24x, x = linear_21_cast_fp16)[name = string("op_1340_cast_fp16")];
            tensor<int32, [4]> x_103_perm_0 = const()[name = string("x_103_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [512, 256]> main_blocks_2_text_attn_W_key_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38761408))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38892544))))[name = string("main_blocks_2_text_attn_W_key_linear_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_2_text_attn_W_key_linear_bias_to_fp16 = const()[name = string("main_blocks_2_text_attn_W_key_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38893632)))];
            tensor<fp16, [?, ?, 512]> linear_22_cast_fp16 = linear(bias = main_blocks_2_text_attn_W_key_linear_bias_to_fp16, weight = main_blocks_2_text_attn_W_key_linear_weight_to_fp16_quantized, x = input_61_cast_fp16)[name = string("linear_22_cast_fp16")];
            tensor<int32, [4]> concat_25x = const()[name = string("concat_25x"), val = tensor<int32, [4]>([2, -1, 8, 64])];
            tensor<fp16, [2, ?, 8, 64]> var_1348_cast_fp16 = reshape(shape = concat_25x, x = linear_22_cast_fp16)[name = string("op_1348_cast_fp16")];
            tensor<int32, [4]> x_105_perm_0 = const()[name = string("x_105_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [512, 256]> main_blocks_2_text_attn_W_value_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38894720))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39025856))))[name = string("main_blocks_2_text_attn_W_value_linear_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_2_text_attn_W_value_linear_bias_to_fp16 = const()[name = string("main_blocks_2_text_attn_W_value_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39026944)))];
            tensor<fp16, [?, ?, 512]> linear_23_cast_fp16 = linear(bias = main_blocks_2_text_attn_W_value_linear_bias_to_fp16, weight = main_blocks_2_text_attn_W_value_linear_weight_to_fp16_quantized, x = input_61_cast_fp16)[name = string("linear_23_cast_fp16")];
            tensor<int32, [4]> concat_26x = const()[name = string("concat_26x"), val = tensor<int32, [4]>([2, -1, 8, 64])];
            tensor<fp16, [2, ?, 8, 64]> var_1356_cast_fp16 = reshape(shape = concat_26x, x = linear_23_cast_fp16)[name = string("op_1356_cast_fp16")];
            tensor<int32, [4]> v_9_perm_0 = const()[name = string("v_9_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            int32 concat_27_values0_0 = const()[name = string("concat_27_values0_0"), val = int32(1)];
            int32 concat_27_values2_0 = const()[name = string("concat_27_values2_0"), val = int32(1)];
            int32 concat_27_axis_0 = const()[name = string("concat_27_axis_0"), val = int32(0)];
            bool concat_27_interleave_0 = const()[name = string("concat_27_interleave_0"), val = bool(false)];
            int32 gather_14_cast_uint16_to_int32 = cast(dtype = gather_14_cast_uint16_to_int32_dtype_0, x = gather_14_cast_uint16)[name = string("cast_3")];
            tensor<int32, [3]> concat_27 = concat(axis = concat_27_axis_0, interleave = concat_27_interleave_0, values = (concat_27_values0_0, gather_14_cast_uint16_to_int32, concat_27_values2_0))[name = string("concat_27")];
            tensor<int32, [3]> var_1359_begin_0 = const()[name = string("op_1359_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<bool, [3]> var_1359_end_mask_0 = const()[name = string("op_1359_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, ?, 1]> var_1359_cast_fp16 = slice_by_index(begin = var_1359_begin_0, end = concat_27, end_mask = var_1359_end_mask_0, x = main_blocks_0_text_attn_increments_to_fp16)[name = string("op_1359_cast_fp16")];
            tensor<int32, [3]> concat_28 = const()[name = string("concat_28"), val = tensor<int32, [3]>([2, -1, -1])];
            tensor<int32, [3]> shape_5_cast_fp16 = shape(x = var_1359_cast_fp16)[name = string("shape_5_cast_fp16")];
            tensor<bool, [3]> equal_5 = const()[name = string("equal_5"), val = tensor<bool, [3]>([false, true, true])];
            tensor<int32, [3]> select_5 = select(a = shape_5_cast_fp16, b = concat_28, cond = equal_5)[name = string("select_5")];
            tensor<int32, [3]> real_div_5 = real_div(x = select_5, y = shape_5_cast_fp16)[name = string("real_div_5")];
            tensor<fp16, [?, ?, ?]> var_1362_cast_fp16 = tile(reps = real_div_5, x = var_1359_cast_fp16)[name = string("op_1362_cast_fp16")];
            tensor<fp16, [2, ?, ?]> scaled_9_cast_fp16 = real_div(x = var_1362_cast_fp16, y = var_427_cast_fp16)[name = string("scaled_9_cast_fp16")];
            tensor<fp16, [1, 1, 32]> main_blocks_2_text_attn_theta_to_fp16 = const()[name = string("main_blocks_2_text_attn_theta_to_fp16"), val = tensor<fp16, [1, 1, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39028032)))];
            tensor<fp16, [2, ?, 32]> angles_9_cast_fp16 = mul(x = scaled_9_cast_fp16, y = main_blocks_2_text_attn_theta_to_fp16)[name = string("angles_9_cast_fp16")];
            tensor<fp16, [2, ?, 32]> var_1378_cast_fp16 = cos(x = angles_9_cast_fp16)[name = string("op_1378_cast_fp16")];
            tensor<int32, [1]> cos_9_axes_0 = const()[name = string("cos_9_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, ?, 32]> cos_9_cast_fp16 = expand_dims(axes = cos_9_axes_0, x = var_1378_cast_fp16)[name = string("cos_9_cast_fp16")];
            tensor<fp16, [2, ?, 32]> var_1380_cast_fp16 = sin(x = angles_9_cast_fp16)[name = string("op_1380_cast_fp16")];
            tensor<int32, [1]> sin_9_axes_0 = const()[name = string("sin_9_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, ?, 32]> sin_9_cast_fp16 = expand_dims(axes = sin_9_axes_0, x = var_1380_cast_fp16)[name = string("sin_9_cast_fp16")];
            tensor<int32, [4]> x_a_9_begin_0 = const()[name = string("x_a_9_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x_a_9_end_0 = const()[name = string("x_a_9_end_0"), val = tensor<int32, [4]>([2, 8, 0, 32])];
            tensor<bool, [4]> x_a_9_end_mask_0 = const()[name = string("x_a_9_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 8, ?, 64]> x_103_cast_fp16 = transpose(perm = x_103_perm_0, x = var_1340_cast_fp16)[name = string("transpose_50")];
            tensor<fp16, [2, 8, ?, 32]> x_a_9_cast_fp16 = slice_by_index(begin = x_a_9_begin_0, end = x_a_9_end_0, end_mask = x_a_9_end_mask_0, x = x_103_cast_fp16)[name = string("x_a_9_cast_fp16")];
            tensor<int32, [4]> x_b_9_begin_0 = const()[name = string("x_b_9_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x_b_9_end_0 = const()[name = string("x_b_9_end_0"), val = tensor<int32, [4]>([2, 8, 0, 64])];
            tensor<bool, [4]> x_b_9_end_mask_0 = const()[name = string("x_b_9_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [2, 8, ?, 32]> x_b_9_cast_fp16 = slice_by_index(begin = x_b_9_begin_0, end = x_b_9_end_0, end_mask = x_b_9_end_mask_0, x = x_103_cast_fp16)[name = string("x_b_9_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1384_cast_fp16 = mul(x = x_a_9_cast_fp16, y = cos_9_cast_fp16)[name = string("op_1384_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1385_cast_fp16 = mul(x = x_b_9_cast_fp16, y = sin_9_cast_fp16)[name = string("op_1385_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_a_9_cast_fp16 = sub(x = var_1384_cast_fp16, y = var_1385_cast_fp16)[name = string("rot_a_9_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1387_cast_fp16 = mul(x = x_a_9_cast_fp16, y = sin_9_cast_fp16)[name = string("op_1387_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1388_cast_fp16 = mul(x = x_b_9_cast_fp16, y = cos_9_cast_fp16)[name = string("op_1388_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_b_9_cast_fp16 = add(x = var_1387_cast_fp16, y = var_1388_cast_fp16)[name = string("rot_b_9_cast_fp16")];
            bool q_9_interleave_0 = const()[name = string("q_9_interleave_0"), val = bool(false)];
            tensor<fp16, [2, 8, ?, 64]> q_9_cast_fp16 = concat(axis = var_1074, interleave = q_9_interleave_0, values = (rot_a_9_cast_fp16, rot_b_9_cast_fp16))[name = string("q_9_cast_fp16")];
            tensor<int32, [4]> x_a_11_begin_0 = const()[name = string("x_a_11_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x_a_11_end_0 = const()[name = string("x_a_11_end_0"), val = tensor<int32, [4]>([2, 8, 0, 32])];
            tensor<bool, [4]> x_a_11_end_mask_0 = const()[name = string("x_a_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 8, ?, 64]> x_105_cast_fp16 = transpose(perm = x_105_perm_0, x = var_1348_cast_fp16)[name = string("transpose_49")];
            tensor<fp16, [2, 8, ?, 32]> x_a_11_cast_fp16 = slice_by_index(begin = x_a_11_begin_0, end = x_a_11_end_0, end_mask = x_a_11_end_mask_0, x = x_105_cast_fp16)[name = string("x_a_11_cast_fp16")];
            tensor<int32, [4]> x_b_11_begin_0 = const()[name = string("x_b_11_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x_b_11_end_0 = const()[name = string("x_b_11_end_0"), val = tensor<int32, [4]>([2, 8, 0, 64])];
            tensor<bool, [4]> x_b_11_end_mask_0 = const()[name = string("x_b_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [2, 8, ?, 32]> x_b_11_cast_fp16 = slice_by_index(begin = x_b_11_begin_0, end = x_b_11_end_0, end_mask = x_b_11_end_mask_0, x = x_105_cast_fp16)[name = string("x_b_11_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1402_cast_fp16 = mul(x = x_a_11_cast_fp16, y = cos_3_cast_fp16)[name = string("op_1402_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1403_cast_fp16 = mul(x = x_b_11_cast_fp16, y = sin_3_cast_fp16)[name = string("op_1403_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_a_11_cast_fp16 = sub(x = var_1402_cast_fp16, y = var_1403_cast_fp16)[name = string("rot_a_11_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1405_cast_fp16 = mul(x = x_a_11_cast_fp16, y = sin_3_cast_fp16)[name = string("op_1405_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1406_cast_fp16 = mul(x = x_b_11_cast_fp16, y = cos_3_cast_fp16)[name = string("op_1406_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_b_11_cast_fp16 = add(x = var_1405_cast_fp16, y = var_1406_cast_fp16)[name = string("rot_b_11_cast_fp16")];
            bool k_9_interleave_0 = const()[name = string("k_9_interleave_0"), val = bool(false)];
            tensor<fp16, [2, 8, ?, 64]> k_9_cast_fp16 = concat(axis = var_1074, interleave = k_9_interleave_0, values = (rot_a_11_cast_fp16, rot_b_11_cast_fp16))[name = string("k_9_cast_fp16")];
            bool var_1411_transpose_x_1 = const()[name = string("op_1411_transpose_x_1"), val = bool(false)];
            bool var_1411_transpose_y_1 = const()[name = string("op_1411_transpose_y_1"), val = bool(true)];
            tensor<fp16, [2, 8, ?, ?]> var_1411_cast_fp16 = matmul(transpose_x = var_1411_transpose_x_1, transpose_y = var_1411_transpose_y_1, x = q_9_cast_fp16, y = k_9_cast_fp16)[name = string("op_1411_cast_fp16")];
            fp16 _inversed_scores_5_y_0_to_fp16 = const()[name = string("_inversed_scores_5_y_0_to_fp16"), val = fp16(0x1p-4)];
            tensor<fp16, [2, 8, ?, ?]> _inversed_scores_5_cast_fp16 = mul(x = var_1411_cast_fp16, y = _inversed_scores_5_y_0_to_fp16)[name = string("_inversed_scores_5_cast_fp16")];
            tensor<fp16, [2, 8, ?, ?]> input_231_cast_fp16 = sub(x = _inversed_scores_5_cast_fp16, y = var_469_cast_fp16)[name = string("input_231_cast_fp16")];
            tensor<fp16, [2, 8, ?, ?]> attn_13_cast_fp16 = softmax(axis = var_1074, x = input_231_cast_fp16)[name = string("attn_13_cast_fp16")];
            bool var_1420_transpose_x_0 = const()[name = string("op_1420_transpose_x_0"), val = bool(false)];
            bool var_1420_transpose_y_0 = const()[name = string("op_1420_transpose_y_0"), val = bool(false)];
            tensor<fp16, [2, 8, ?, 64]> v_9_cast_fp16 = transpose(perm = v_9_perm_0, x = var_1356_cast_fp16)[name = string("transpose_48")];
            tensor<fp16, [2, 8, ?, 64]> var_1420_cast_fp16 = matmul(transpose_x = var_1420_transpose_x_0, transpose_y = var_1420_transpose_y_0, x = attn_13_cast_fp16, y = v_9_cast_fp16)[name = string("op_1420_cast_fp16")];
            tensor<int32, [4]> var_1421_perm_0 = const()[name = string("op_1421_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_31x = const()[name = string("concat_31x"), val = tensor<int32, [3]>([2, -1, 512])];
            tensor<fp16, [2, ?, 8, 64]> var_1421_cast_fp16 = transpose(perm = var_1421_perm_0, x = var_1420_cast_fp16)[name = string("transpose_47")];
            tensor<fp16, [2, ?, 512]> input_233_cast_fp16 = reshape(shape = concat_31x, x = var_1421_cast_fp16)[name = string("input_233_cast_fp16")];
            tensor<fp16, [512, 512]> main_blocks_2_text_attn_out_fc_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39028160))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39290368))))[name = string("main_blocks_2_text_attn_out_fc_linear_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_2_text_attn_out_fc_linear_bias_to_fp16 = const()[name = string("main_blocks_2_text_attn_out_fc_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39291456)))];
            tensor<fp16, [2, ?, 512]> linear_24_cast_fp16 = linear(bias = main_blocks_2_text_attn_out_fc_linear_bias_to_fp16, weight = main_blocks_2_text_attn_out_fc_linear_weight_to_fp16_quantized, x = input_233_cast_fp16)[name = string("linear_24_cast_fp16")];
            tensor<int32, [3]> out_43_perm_0 = const()[name = string("out_43_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> out_43_cast_fp16 = transpose(perm = out_43_perm_0, x = linear_24_cast_fp16)[name = string("transpose_46")];
            tensor<fp16, [2, 512, ?]> var_1429_cast_fp16 = mul(x = out_43_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_1429_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_107_cast_fp16 = add(x = x_m_9_cast_fp16, y = var_1429_cast_fp16)[name = string("x_107_cast_fp16")];
            tensor<int32, [3]> input_235_perm_0 = const()[name = string("input_235_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1436_axes_0 = const()[name = string("op_1436_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_2_text_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_2_text_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39292544)))];
            tensor<fp16, [512]> main_blocks_2_text_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_2_text_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39293632)))];
            tensor<fp16, [2, ?, 512]> input_235_cast_fp16 = transpose(perm = input_235_perm_0, x = x_107_cast_fp16)[name = string("transpose_45")];
            tensor<fp16, [2, ?, 512]> var_1436_cast_fp16 = layer_norm(axes = var_1436_axes_0, beta = main_blocks_2_text_norm_norm_bias_to_fp16, epsilon = var_1086_to_fp16, gamma = main_blocks_2_text_norm_norm_weight_to_fp16, x = input_235_cast_fp16)[name = string("op_1436_cast_fp16")];
            tensor<int32, [3]> var_1437_perm_0 = const()[name = string("op_1437_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> var_1437_cast_fp16 = transpose(perm = var_1437_perm_0, x = var_1436_cast_fp16)[name = string("transpose_44")];
            tensor<fp16, [2, 512, ?]> x_109_cast_fp16 = mul(x = var_1437_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_109_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_237_cast_fp16 = mul(x = x_109_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("input_237_cast_fp16")];
            tensor<int32, [6]> input_239_pad_0 = const()[name = string("input_239_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_239_mode_0 = const()[name = string("input_239_mode_0"), val = string("replicate")];
            fp16 const_19_to_fp16 = const()[name = string("const_19_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_239_cast_fp16 = pad(constant_val = const_19_to_fp16, mode = input_239_mode_0, pad = input_239_pad_0, x = input_237_cast_fp16)[name = string("input_239_cast_fp16")];
            string h_105_pad_type_0 = const()[name = string("h_105_pad_type_0"), val = string("valid")];
            int32 h_105_groups_0 = const()[name = string("h_105_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_105_strides_0 = const()[name = string("h_105_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_105_pad_0 = const()[name = string("h_105_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_105_dilations_0 = const()[name = string("h_105_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 5]> main_blocks_2_convnext_2_0_dwconv__conv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39294720))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39297344))))[name = string("main_blocks_2_convnext_2_0_dwconv__conv_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_2_convnext_2_0_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_2_0_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39298432)))];
            tensor<fp16, [2, 512, ?]> h_105_cast_fp16 = conv(bias = main_blocks_2_convnext_2_0_dwconv__conv_bias_to_fp16, dilations = h_105_dilations_0, groups = h_105_groups_0, pad = h_105_pad_0, pad_type = h_105_pad_type_0, strides = h_105_strides_0, weight = main_blocks_2_convnext_2_0_dwconv__conv_weight_to_fp16_quantized, x = input_239_cast_fp16)[name = string("h_105_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_111_cast_fp16 = mul(x = h_105_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_111_cast_fp16")];
            tensor<int32, [3]> input_241_perm_0 = const()[name = string("input_241_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1461_axes_0 = const()[name = string("op_1461_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_2_convnext_2_0_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_2_convnext_2_0_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39299520)))];
            tensor<fp16, [512]> main_blocks_2_convnext_2_0_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_2_0_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39300608)))];
            tensor<fp16, [2, ?, 512]> input_241_cast_fp16 = transpose(perm = input_241_perm_0, x = x_111_cast_fp16)[name = string("transpose_43")];
            tensor<fp16, [2, ?, 512]> var_1461_cast_fp16 = layer_norm(axes = var_1461_axes_0, beta = main_blocks_2_convnext_2_0_norm_norm_bias_to_fp16, epsilon = var_1086_to_fp16, gamma = main_blocks_2_convnext_2_0_norm_norm_weight_to_fp16, x = input_241_cast_fp16)[name = string("op_1461_cast_fp16")];
            tensor<int32, [3]> input_243_perm_0 = const()[name = string("input_243_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_107_pad_type_0 = const()[name = string("h_107_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_107_strides_0 = const()[name = string("h_107_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_107_pad_0 = const()[name = string("h_107_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_107_dilations_0 = const()[name = string("h_107_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_107_groups_0 = const()[name = string("h_107_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_2_convnext_2_0_pwconv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39301696))), scale = tensor<fp16, [2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40350336))))[name = string("main_blocks_2_convnext_2_0_pwconv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> main_blocks_2_convnext_2_0_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_2_convnext_2_0_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40354496)))];
            tensor<fp16, [2, 512, ?]> input_243_cast_fp16 = transpose(perm = input_243_perm_0, x = var_1461_cast_fp16)[name = string("transpose_42")];
            tensor<fp16, [2, 2048, ?]> h_107_cast_fp16 = conv(bias = main_blocks_2_convnext_2_0_pwconv1_bias_to_fp16, dilations = h_107_dilations_0, groups = h_107_groups_0, pad = h_107_pad_0, pad_type = h_107_pad_type_0, strides = h_107_strides_0, weight = main_blocks_2_convnext_2_0_pwconv1_weight_to_fp16_quantized, x = input_243_cast_fp16)[name = string("h_107_cast_fp16")];
            string input_245_mode_0 = const()[name = string("input_245_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_245_cast_fp16 = gelu(mode = input_245_mode_0, x = h_107_cast_fp16)[name = string("input_245_cast_fp16")];
            string h_109_pad_type_0 = const()[name = string("h_109_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_109_strides_0 = const()[name = string("h_109_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_109_pad_0 = const()[name = string("h_109_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_109_dilations_0 = const()[name = string("h_109_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_109_groups_0 = const()[name = string("h_109_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> op_1478_weight_0_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40358656))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41407296))))[name = string("op_1478_weight_0_to_fp16_quantized")];
            tensor<fp16, [512]> var_1478_bias_0_to_fp16 = const()[name = string("op_1478_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41408384)))];
            tensor<fp16, [2, 512, ?]> var_1478_cast_fp16 = conv(bias = var_1478_bias_0_to_fp16, dilations = h_109_dilations_0, groups = h_109_groups_0, pad = h_109_pad_0, pad_type = h_109_pad_type_0, strides = h_109_strides_0, weight = op_1478_weight_0_to_fp16_quantized, x = input_245_cast_fp16)[name = string("op_1478_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_45_cast_fp16 = add(x = input_237_cast_fp16, y = var_1478_cast_fp16)[name = string("out_45_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_113_cast_fp16 = mul(x = out_45_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_113_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_m_11_cast_fp16 = mul(x = x_113_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_m_11_cast_fp16")];
            tensor<int32, [3]> input_247_perm_0 = const()[name = string("input_247_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [256, 512]> main_blocks_2_style_attn_W_query_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [256, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41409472))), scale = tensor<fp16, [256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41540608))))[name = string("main_blocks_2_style_attn_W_query_linear_weight_to_fp16_quantized")];
            tensor<fp16, [256]> main_blocks_2_style_attn_W_query_linear_bias_to_fp16 = const()[name = string("main_blocks_2_style_attn_W_query_linear_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41541184)))];
            tensor<fp16, [2, ?, 512]> input_247_cast_fp16 = transpose(perm = input_247_perm_0, x = x_113_cast_fp16)[name = string("transpose_41")];
            tensor<fp16, [2, ?, 256]> linear_25_cast_fp16 = linear(bias = main_blocks_2_style_attn_W_query_linear_bias_to_fp16, weight = main_blocks_2_style_attn_W_query_linear_weight_to_fp16_quantized, x = input_247_cast_fp16)[name = string("linear_25_cast_fp16")];
            tensor<int32, [4]> concat_32x = const()[name = string("concat_32x"), val = tensor<int32, [4]>([2, -1, 2, 128])];
            tensor<fp16, [2, ?, 2, 128]> var_1495_cast_fp16 = reshape(shape = concat_32x, x = linear_25_cast_fp16)[name = string("op_1495_cast_fp16")];
            tensor<int32, [4]> q_11_perm_0 = const()[name = string("q_11_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [256, 256]> main_blocks_2_style_attn_W_value_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [256, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41541760))), scale = tensor<fp16, [256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41607360))))[name = string("main_blocks_2_style_attn_W_value_linear_weight_to_fp16_quantized")];
            tensor<fp16, [256]> main_blocks_2_style_attn_W_value_linear_bias_to_fp16 = const()[name = string("main_blocks_2_style_attn_W_value_linear_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41607936)))];
            tensor<fp16, [2, 50, 256]> linear_27_cast_fp16 = linear(bias = main_blocks_2_style_attn_W_value_linear_bias_to_fp16, weight = main_blocks_2_style_attn_W_value_linear_weight_to_fp16_quantized, x = input_83_cast_fp16)[name = string("linear_27_cast_fp16")];
            tensor<int32, [4]> var_1508 = const()[name = string("op_1508"), val = tensor<int32, [4]>([2, 50, 2, 128])];
            tensor<fp16, [2, 50, 2, 128]> var_1509_cast_fp16 = reshape(shape = var_1508, x = linear_27_cast_fp16)[name = string("op_1509_cast_fp16")];
            tensor<int32, [4]> v_11_perm_0 = const()[name = string("v_11_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool var_1513_transpose_x_0 = const()[name = string("op_1513_transpose_x_0"), val = bool(false)];
            bool var_1513_transpose_y_0 = const()[name = string("op_1513_transpose_y_0"), val = bool(false)];
            tensor<fp16, [2, 2, 128, 50]> op_1512_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2, 2, 128, 50]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41608512))), scale = tensor<fp16, [1, 1, 1, 50]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41634176))))[name = string("op_1512_to_fp16_quantized")];
            tensor<fp16, [2, 2, ?, 128]> q_11_cast_fp16 = transpose(perm = q_11_perm_0, x = var_1495_cast_fp16)[name = string("transpose_40")];
            tensor<fp16, [2, 2, ?, 50]> var_1513_cast_fp16 = matmul(transpose_x = var_1513_transpose_x_0, transpose_y = var_1513_transpose_y_0, x = q_11_cast_fp16, y = op_1512_to_fp16_quantized)[name = string("op_1513_cast_fp16")];
            fp16 _inversed_input_249_y_0_to_fp16 = const()[name = string("_inversed_input_249_y_0_to_fp16"), val = fp16(0x1p-4)];
            tensor<fp16, [2, 2, ?, 50]> _inversed_input_249_cast_fp16 = mul(x = var_1513_cast_fp16, y = _inversed_input_249_y_0_to_fp16)[name = string("_inversed_input_249_cast_fp16")];
            tensor<fp16, [2, 2, ?, 50]> attn_15_cast_fp16 = softmax(axis = var_1074, x = _inversed_input_249_cast_fp16)[name = string("attn_15_cast_fp16")];
            tensor<fp16, [2, 2, ?, 50]> attn_17_cast_fp16 = mul(x = attn_15_cast_fp16, y = mask_1_cast_fp16)[name = string("attn_17_cast_fp16")];
            bool var_1520_transpose_x_0 = const()[name = string("op_1520_transpose_x_0"), val = bool(false)];
            bool var_1520_transpose_y_0 = const()[name = string("op_1520_transpose_y_0"), val = bool(false)];
            tensor<fp16, [2, 2, 50, 128]> v_11_cast_fp16 = transpose(perm = v_11_perm_0, x = var_1509_cast_fp16)[name = string("transpose_39")];
            tensor<fp16, [2, 2, ?, 128]> var_1520_cast_fp16 = matmul(transpose_x = var_1520_transpose_x_0, transpose_y = var_1520_transpose_y_0, x = attn_17_cast_fp16, y = v_11_cast_fp16)[name = string("op_1520_cast_fp16")];
            tensor<int32, [4]> var_1521_perm_0 = const()[name = string("op_1521_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_33x = const()[name = string("concat_33x"), val = tensor<int32, [3]>([2, -1, 256])];
            tensor<fp16, [2, ?, 2, 128]> var_1521_cast_fp16 = transpose(perm = var_1521_perm_0, x = var_1520_cast_fp16)[name = string("transpose_38")];
            tensor<fp16, [2, ?, 256]> input_251_cast_fp16 = reshape(shape = concat_33x, x = var_1521_cast_fp16)[name = string("input_251_cast_fp16")];
            tensor<fp16, [512, 256]> main_blocks_2_style_attn_out_fc_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41634368))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41765504))))[name = string("main_blocks_2_style_attn_out_fc_linear_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_2_style_attn_out_fc_linear_bias_to_fp16 = const()[name = string("main_blocks_2_style_attn_out_fc_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41766592)))];
            tensor<fp16, [2, ?, 512]> linear_28_cast_fp16 = linear(bias = main_blocks_2_style_attn_out_fc_linear_bias_to_fp16, weight = main_blocks_2_style_attn_out_fc_linear_weight_to_fp16_quantized, x = input_251_cast_fp16)[name = string("linear_28_cast_fp16")];
            tensor<int32, [3]> out_47_perm_0 = const()[name = string("out_47_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> out_47_cast_fp16 = transpose(perm = out_47_perm_0, x = linear_28_cast_fp16)[name = string("transpose_37")];
            tensor<fp16, [2, 512, ?]> var_1529_cast_fp16 = mul(x = out_47_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_1529_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_115_cast_fp16 = add(x = x_m_11_cast_fp16, y = var_1529_cast_fp16)[name = string("x_115_cast_fp16")];
            tensor<int32, [3]> input_253_perm_0 = const()[name = string("input_253_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1536_axes_0 = const()[name = string("op_1536_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_2_style_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_2_style_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41767680)))];
            tensor<fp16, [512]> main_blocks_2_style_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_2_style_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41768768)))];
            tensor<fp16, [2, ?, 512]> input_253_cast_fp16 = transpose(perm = input_253_perm_0, x = x_115_cast_fp16)[name = string("transpose_36")];
            tensor<fp16, [2, ?, 512]> var_1536_cast_fp16 = layer_norm(axes = var_1536_axes_0, beta = main_blocks_2_style_norm_norm_bias_to_fp16, epsilon = var_1086_to_fp16, gamma = main_blocks_2_style_norm_norm_weight_to_fp16, x = input_253_cast_fp16)[name = string("op_1536_cast_fp16")];
            tensor<int32, [3]> var_1537_perm_0 = const()[name = string("op_1537_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> var_1537_cast_fp16 = transpose(perm = var_1537_perm_0, x = var_1536_cast_fp16)[name = string("transpose_35")];
            tensor<fp16, [2, 512, ?]> x_117_cast_fp16 = mul(x = var_1537_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_117_cast_fp16")];
            int32 var_1548 = const()[name = string("op_1548"), val = int32(-1)];
            tensor<fp16, [2, 512, ?]> input_255_cast_fp16 = mul(x = x_117_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("input_255_cast_fp16")];
            tensor<int32, [6]> input_257_pad_0 = const()[name = string("input_257_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_257_mode_0 = const()[name = string("input_257_mode_0"), val = string("replicate")];
            fp16 const_21_to_fp16 = const()[name = string("const_21_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_257_cast_fp16 = pad(constant_val = const_21_to_fp16, mode = input_257_mode_0, pad = input_257_pad_0, x = input_255_cast_fp16)[name = string("input_257_cast_fp16")];
            string h_111_pad_type_0 = const()[name = string("h_111_pad_type_0"), val = string("valid")];
            int32 h_111_groups_0 = const()[name = string("h_111_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_111_strides_0 = const()[name = string("h_111_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_111_pad_0 = const()[name = string("h_111_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_111_dilations_0 = const()[name = string("h_111_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 5]> main_blocks_3_convnext_0_0_dwconv__conv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41769856))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41772480))))[name = string("main_blocks_3_convnext_0_0_dwconv__conv_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_3_convnext_0_0_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_0_0_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41773568)))];
            tensor<fp16, [2, 512, ?]> h_111_cast_fp16 = conv(bias = main_blocks_3_convnext_0_0_dwconv__conv_bias_to_fp16, dilations = h_111_dilations_0, groups = h_111_groups_0, pad = h_111_pad_0, pad_type = h_111_pad_type_0, strides = h_111_strides_0, weight = main_blocks_3_convnext_0_0_dwconv__conv_weight_to_fp16_quantized, x = input_257_cast_fp16)[name = string("h_111_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_119_cast_fp16 = mul(x = h_111_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_119_cast_fp16")];
            tensor<int32, [3]> input_259_perm_0 = const()[name = string("input_259_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1601_axes_0 = const()[name = string("op_1601_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_3_convnext_0_0_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_3_convnext_0_0_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41774656)))];
            tensor<fp16, [512]> main_blocks_3_convnext_0_0_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_0_0_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41775744)))];
            fp16 var_1560_to_fp16 = const()[name = string("op_1560_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [2, ?, 512]> input_259_cast_fp16 = transpose(perm = input_259_perm_0, x = x_119_cast_fp16)[name = string("transpose_34")];
            tensor<fp16, [2, ?, 512]> var_1601_cast_fp16 = layer_norm(axes = var_1601_axes_0, beta = main_blocks_3_convnext_0_0_norm_norm_bias_to_fp16, epsilon = var_1560_to_fp16, gamma = main_blocks_3_convnext_0_0_norm_norm_weight_to_fp16, x = input_259_cast_fp16)[name = string("op_1601_cast_fp16")];
            tensor<int32, [3]> input_261_perm_0 = const()[name = string("input_261_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_113_pad_type_0 = const()[name = string("h_113_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_113_strides_0 = const()[name = string("h_113_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_113_pad_0 = const()[name = string("h_113_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_113_dilations_0 = const()[name = string("h_113_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_113_groups_0 = const()[name = string("h_113_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_3_convnext_0_0_pwconv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41776832))), scale = tensor<fp16, [2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42825472))))[name = string("main_blocks_3_convnext_0_0_pwconv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> main_blocks_3_convnext_0_0_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_0_0_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42829632)))];
            tensor<fp16, [2, 512, ?]> input_261_cast_fp16 = transpose(perm = input_261_perm_0, x = var_1601_cast_fp16)[name = string("transpose_33")];
            tensor<fp16, [2, 2048, ?]> h_113_cast_fp16 = conv(bias = main_blocks_3_convnext_0_0_pwconv1_bias_to_fp16, dilations = h_113_dilations_0, groups = h_113_groups_0, pad = h_113_pad_0, pad_type = h_113_pad_type_0, strides = h_113_strides_0, weight = main_blocks_3_convnext_0_0_pwconv1_weight_to_fp16_quantized, x = input_261_cast_fp16)[name = string("h_113_cast_fp16")];
            string input_263_mode_0 = const()[name = string("input_263_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_263_cast_fp16 = gelu(mode = input_263_mode_0, x = h_113_cast_fp16)[name = string("input_263_cast_fp16")];
            string h_115_pad_type_0 = const()[name = string("h_115_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_115_strides_0 = const()[name = string("h_115_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_115_pad_0 = const()[name = string("h_115_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_115_dilations_0 = const()[name = string("h_115_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_115_groups_0 = const()[name = string("h_115_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> op_1618_weight_0_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42833792))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43882432))))[name = string("op_1618_weight_0_to_fp16_quantized")];
            tensor<fp16, [512]> var_1618_bias_0_to_fp16 = const()[name = string("op_1618_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43883520)))];
            tensor<fp16, [2, 512, ?]> var_1618_cast_fp16 = conv(bias = var_1618_bias_0_to_fp16, dilations = h_115_dilations_0, groups = h_115_groups_0, pad = h_115_pad_0, pad_type = h_115_pad_type_0, strides = h_115_strides_0, weight = op_1618_weight_0_to_fp16_quantized, x = input_263_cast_fp16)[name = string("op_1618_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_49_cast_fp16 = add(x = input_255_cast_fp16, y = var_1618_cast_fp16)[name = string("out_49_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_121_cast_fp16 = mul(x = out_49_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_121_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_265_cast_fp16 = mul(x = x_121_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("input_265_cast_fp16")];
            tensor<int32, [6]> input_267_pad_0 = const()[name = string("input_267_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_267_mode_0 = const()[name = string("input_267_mode_0"), val = string("replicate")];
            fp16 const_22_to_fp16 = const()[name = string("const_22_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_267_cast_fp16 = pad(constant_val = const_22_to_fp16, mode = input_267_mode_0, pad = input_267_pad_0, x = input_265_cast_fp16)[name = string("input_267_cast_fp16")];
            string h_117_pad_type_0 = const()[name = string("h_117_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_117_dilations_0 = const()[name = string("h_117_dilations_0"), val = tensor<int32, [1]>([2])];
            int32 h_117_groups_0 = const()[name = string("h_117_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_117_strides_0 = const()[name = string("h_117_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_117_pad_0 = const()[name = string("h_117_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [512, 1, 5]> main_blocks_3_convnext_0_1_dwconv__conv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43884608))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43887232))))[name = string("main_blocks_3_convnext_0_1_dwconv__conv_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_3_convnext_0_1_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_0_1_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43888320)))];
            tensor<fp16, [2, 512, ?]> h_117_cast_fp16 = conv(bias = main_blocks_3_convnext_0_1_dwconv__conv_bias_to_fp16, dilations = h_117_dilations_0, groups = h_117_groups_0, pad = h_117_pad_0, pad_type = h_117_pad_type_0, strides = h_117_strides_0, weight = main_blocks_3_convnext_0_1_dwconv__conv_weight_to_fp16_quantized, x = input_267_cast_fp16)[name = string("h_117_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_123_cast_fp16 = mul(x = h_117_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_123_cast_fp16")];
            tensor<int32, [3]> input_269_perm_0 = const()[name = string("input_269_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1643_axes_0 = const()[name = string("op_1643_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_3_convnext_0_1_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_3_convnext_0_1_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43889408)))];
            tensor<fp16, [512]> main_blocks_3_convnext_0_1_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_0_1_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43890496)))];
            tensor<fp16, [2, ?, 512]> input_269_cast_fp16 = transpose(perm = input_269_perm_0, x = x_123_cast_fp16)[name = string("transpose_32")];
            tensor<fp16, [2, ?, 512]> var_1643_cast_fp16 = layer_norm(axes = var_1643_axes_0, beta = main_blocks_3_convnext_0_1_norm_norm_bias_to_fp16, epsilon = var_1560_to_fp16, gamma = main_blocks_3_convnext_0_1_norm_norm_weight_to_fp16, x = input_269_cast_fp16)[name = string("op_1643_cast_fp16")];
            tensor<int32, [3]> input_271_perm_0 = const()[name = string("input_271_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_119_pad_type_0 = const()[name = string("h_119_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_119_strides_0 = const()[name = string("h_119_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_119_pad_0 = const()[name = string("h_119_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_119_dilations_0 = const()[name = string("h_119_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_119_groups_0 = const()[name = string("h_119_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_3_convnext_0_1_pwconv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43891584))), scale = tensor<fp16, [2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44940224))))[name = string("main_blocks_3_convnext_0_1_pwconv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> main_blocks_3_convnext_0_1_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_0_1_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44944384)))];
            tensor<fp16, [2, 512, ?]> input_271_cast_fp16 = transpose(perm = input_271_perm_0, x = var_1643_cast_fp16)[name = string("transpose_31")];
            tensor<fp16, [2, 2048, ?]> h_119_cast_fp16 = conv(bias = main_blocks_3_convnext_0_1_pwconv1_bias_to_fp16, dilations = h_119_dilations_0, groups = h_119_groups_0, pad = h_119_pad_0, pad_type = h_119_pad_type_0, strides = h_119_strides_0, weight = main_blocks_3_convnext_0_1_pwconv1_weight_to_fp16_quantized, x = input_271_cast_fp16)[name = string("h_119_cast_fp16")];
            string input_273_mode_0 = const()[name = string("input_273_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_273_cast_fp16 = gelu(mode = input_273_mode_0, x = h_119_cast_fp16)[name = string("input_273_cast_fp16")];
            string h_121_pad_type_0 = const()[name = string("h_121_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_121_strides_0 = const()[name = string("h_121_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_121_pad_0 = const()[name = string("h_121_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_121_dilations_0 = const()[name = string("h_121_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_121_groups_0 = const()[name = string("h_121_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> op_1660_weight_0_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44948544))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45997184))))[name = string("op_1660_weight_0_to_fp16_quantized")];
            tensor<fp16, [512]> var_1660_bias_0_to_fp16 = const()[name = string("op_1660_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45998272)))];
            tensor<fp16, [2, 512, ?]> var_1660_cast_fp16 = conv(bias = var_1660_bias_0_to_fp16, dilations = h_121_dilations_0, groups = h_121_groups_0, pad = h_121_pad_0, pad_type = h_121_pad_type_0, strides = h_121_strides_0, weight = op_1660_weight_0_to_fp16_quantized, x = input_273_cast_fp16)[name = string("op_1660_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_51_cast_fp16 = add(x = input_265_cast_fp16, y = var_1660_cast_fp16)[name = string("out_51_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_125_cast_fp16 = mul(x = out_51_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_125_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_275_cast_fp16 = mul(x = x_125_cast_fp16, y = latent_mask_b_cast_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, 8, 8])];
            string input_277_mode_0 = const()[name = string("input_277_mode_0"), val = string("replicate")];
            fp16 const_23_to_fp16 = const()[name = string("const_23_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_277_cast_fp16 = pad(constant_val = const_23_to_fp16, mode = input_277_mode_0, pad = input_277_pad_0, x = input_275_cast_fp16)[name = string("input_277_cast_fp16")];
            string h_123_pad_type_0 = const()[name = string("h_123_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_123_dilations_0 = const()[name = string("h_123_dilations_0"), val = tensor<int32, [1]>([4])];
            int32 h_123_groups_0 = const()[name = string("h_123_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_123_strides_0 = const()[name = string("h_123_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_123_pad_0 = const()[name = string("h_123_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [512, 1, 5]> main_blocks_3_convnext_0_2_dwconv__conv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45999360))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46001984))))[name = string("main_blocks_3_convnext_0_2_dwconv__conv_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_3_convnext_0_2_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_0_2_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46003072)))];
            tensor<fp16, [2, 512, ?]> h_123_cast_fp16 = conv(bias = main_blocks_3_convnext_0_2_dwconv__conv_bias_to_fp16, dilations = h_123_dilations_0, groups = h_123_groups_0, pad = h_123_pad_0, pad_type = h_123_pad_type_0, strides = h_123_strides_0, weight = main_blocks_3_convnext_0_2_dwconv__conv_weight_to_fp16_quantized, x = input_277_cast_fp16)[name = string("h_123_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_127_cast_fp16 = mul(x = h_123_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_127_cast_fp16")];
            tensor<int32, [3]> input_279_perm_0 = const()[name = string("input_279_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1685_axes_0 = const()[name = string("op_1685_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_3_convnext_0_2_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_3_convnext_0_2_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46004160)))];
            tensor<fp16, [512]> main_blocks_3_convnext_0_2_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_0_2_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46005248)))];
            tensor<fp16, [2, ?, 512]> input_279_cast_fp16 = transpose(perm = input_279_perm_0, x = x_127_cast_fp16)[name = string("transpose_30")];
            tensor<fp16, [2, ?, 512]> var_1685_cast_fp16 = layer_norm(axes = var_1685_axes_0, beta = main_blocks_3_convnext_0_2_norm_norm_bias_to_fp16, epsilon = var_1560_to_fp16, gamma = main_blocks_3_convnext_0_2_norm_norm_weight_to_fp16, x = input_279_cast_fp16)[name = string("op_1685_cast_fp16")];
            tensor<int32, [3]> input_281_perm_0 = const()[name = string("input_281_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_125_pad_type_0 = const()[name = string("h_125_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_125_strides_0 = const()[name = string("h_125_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_125_pad_0 = const()[name = string("h_125_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_125_dilations_0 = const()[name = string("h_125_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_125_groups_0 = const()[name = string("h_125_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_3_convnext_0_2_pwconv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46006336))), scale = tensor<fp16, [2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47054976))))[name = string("main_blocks_3_convnext_0_2_pwconv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> main_blocks_3_convnext_0_2_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_0_2_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47059136)))];
            tensor<fp16, [2, 512, ?]> input_281_cast_fp16 = transpose(perm = input_281_perm_0, x = var_1685_cast_fp16)[name = string("transpose_29")];
            tensor<fp16, [2, 2048, ?]> h_125_cast_fp16 = conv(bias = main_blocks_3_convnext_0_2_pwconv1_bias_to_fp16, dilations = h_125_dilations_0, groups = h_125_groups_0, pad = h_125_pad_0, pad_type = h_125_pad_type_0, strides = h_125_strides_0, weight = main_blocks_3_convnext_0_2_pwconv1_weight_to_fp16_quantized, x = input_281_cast_fp16)[name = string("h_125_cast_fp16")];
            string input_283_mode_0 = const()[name = string("input_283_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_283_cast_fp16 = gelu(mode = input_283_mode_0, x = h_125_cast_fp16)[name = string("input_283_cast_fp16")];
            string h_127_pad_type_0 = const()[name = string("h_127_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_127_strides_0 = const()[name = string("h_127_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_127_pad_0 = const()[name = string("h_127_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_127_dilations_0 = const()[name = string("h_127_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_127_groups_0 = const()[name = string("h_127_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> op_1702_weight_0_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47063296))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48111936))))[name = string("op_1702_weight_0_to_fp16_quantized")];
            tensor<fp16, [512]> var_1702_bias_0_to_fp16 = const()[name = string("op_1702_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48113024)))];
            tensor<fp16, [2, 512, ?]> var_1702_cast_fp16 = conv(bias = var_1702_bias_0_to_fp16, dilations = h_127_dilations_0, groups = h_127_groups_0, pad = h_127_pad_0, pad_type = h_127_pad_type_0, strides = h_127_strides_0, weight = op_1702_weight_0_to_fp16_quantized, x = input_283_cast_fp16)[name = string("op_1702_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_53_cast_fp16 = add(x = input_275_cast_fp16, y = var_1702_cast_fp16)[name = string("out_53_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_129_cast_fp16 = mul(x = out_53_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_129_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_285_cast_fp16 = mul(x = x_129_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("input_285_cast_fp16")];
            tensor<int32, [6]> input_287_pad_0 = const()[name = string("input_287_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 16, 16])];
            string input_287_mode_0 = const()[name = string("input_287_mode_0"), val = string("replicate")];
            fp16 const_24_to_fp16 = const()[name = string("const_24_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_287_cast_fp16 = pad(constant_val = const_24_to_fp16, mode = input_287_mode_0, pad = input_287_pad_0, x = input_285_cast_fp16)[name = string("input_287_cast_fp16")];
            string h_129_pad_type_0 = const()[name = string("h_129_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_129_dilations_0 = const()[name = string("h_129_dilations_0"), val = tensor<int32, [1]>([8])];
            int32 h_129_groups_0 = const()[name = string("h_129_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_129_strides_0 = const()[name = string("h_129_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_129_pad_0 = const()[name = string("h_129_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<fp16, [512, 1, 5]> main_blocks_3_convnext_0_3_dwconv__conv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48114112))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48116736))))[name = string("main_blocks_3_convnext_0_3_dwconv__conv_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_3_convnext_0_3_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_0_3_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48117824)))];
            tensor<fp16, [2, 512, ?]> h_129_cast_fp16 = conv(bias = main_blocks_3_convnext_0_3_dwconv__conv_bias_to_fp16, dilations = h_129_dilations_0, groups = h_129_groups_0, pad = h_129_pad_0, pad_type = h_129_pad_type_0, strides = h_129_strides_0, weight = main_blocks_3_convnext_0_3_dwconv__conv_weight_to_fp16_quantized, x = input_287_cast_fp16)[name = string("h_129_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_131_cast_fp16 = mul(x = h_129_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_131_cast_fp16")];
            tensor<int32, [3]> input_289_perm_0 = const()[name = string("input_289_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1727_axes_0 = const()[name = string("op_1727_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_3_convnext_0_3_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_3_convnext_0_3_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48118912)))];
            tensor<fp16, [512]> main_blocks_3_convnext_0_3_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_0_3_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48120000)))];
            tensor<fp16, [2, ?, 512]> input_289_cast_fp16 = transpose(perm = input_289_perm_0, x = x_131_cast_fp16)[name = string("transpose_28")];
            tensor<fp16, [2, ?, 512]> var_1727_cast_fp16 = layer_norm(axes = var_1727_axes_0, beta = main_blocks_3_convnext_0_3_norm_norm_bias_to_fp16, epsilon = var_1560_to_fp16, gamma = main_blocks_3_convnext_0_3_norm_norm_weight_to_fp16, x = input_289_cast_fp16)[name = string("op_1727_cast_fp16")];
            tensor<int32, [3]> input_291_perm_0 = const()[name = string("input_291_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_131_pad_type_0 = const()[name = string("h_131_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_131_strides_0 = const()[name = string("h_131_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_131_pad_0 = const()[name = string("h_131_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_131_dilations_0 = const()[name = string("h_131_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_131_groups_0 = const()[name = string("h_131_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_3_convnext_0_3_pwconv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48121088))), scale = tensor<fp16, [2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49169728))))[name = string("main_blocks_3_convnext_0_3_pwconv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> main_blocks_3_convnext_0_3_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_0_3_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49173888)))];
            tensor<fp16, [2, 512, ?]> input_291_cast_fp16 = transpose(perm = input_291_perm_0, x = var_1727_cast_fp16)[name = string("transpose_27")];
            tensor<fp16, [2, 2048, ?]> h_131_cast_fp16 = conv(bias = main_blocks_3_convnext_0_3_pwconv1_bias_to_fp16, dilations = h_131_dilations_0, groups = h_131_groups_0, pad = h_131_pad_0, pad_type = h_131_pad_type_0, strides = h_131_strides_0, weight = main_blocks_3_convnext_0_3_pwconv1_weight_to_fp16_quantized, x = input_291_cast_fp16)[name = string("h_131_cast_fp16")];
            string input_293_mode_0 = const()[name = string("input_293_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_293_cast_fp16 = gelu(mode = input_293_mode_0, x = h_131_cast_fp16)[name = string("input_293_cast_fp16")];
            string h_133_pad_type_0 = const()[name = string("h_133_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_133_strides_0 = const()[name = string("h_133_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_133_pad_0 = const()[name = string("h_133_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_133_dilations_0 = const()[name = string("h_133_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_133_groups_0 = const()[name = string("h_133_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> op_1744_weight_0_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49178048))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50226688))))[name = string("op_1744_weight_0_to_fp16_quantized")];
            tensor<fp16, [512]> var_1744_bias_0_to_fp16 = const()[name = string("op_1744_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50227776)))];
            tensor<fp16, [2, 512, ?]> var_1744_cast_fp16 = conv(bias = var_1744_bias_0_to_fp16, dilations = h_133_dilations_0, groups = h_133_groups_0, pad = h_133_pad_0, pad_type = h_133_pad_type_0, strides = h_133_strides_0, weight = op_1744_weight_0_to_fp16_quantized, x = input_293_cast_fp16)[name = string("op_1744_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_55_cast_fp16 = add(x = input_285_cast_fp16, y = var_1744_cast_fp16)[name = string("out_55_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_133_cast_fp16 = mul(x = out_55_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_133_cast_fp16")];
            tensor<fp16, [512, 64]> main_blocks_3_time_cond_linear_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50228864))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50261696))))[name = string("main_blocks_3_time_cond_linear_linear_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_3_time_cond_linear_linear_bias_to_fp16 = const()[name = string("main_blocks_3_time_cond_linear_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50262784)))];
            tensor<fp16, [2, 512]> linear_29_cast_fp16 = linear(bias = main_blocks_3_time_cond_linear_linear_bias_to_fp16, weight = main_blocks_3_time_cond_linear_linear_weight_to_fp16_quantized, x = input_47_cast_fp16)[name = string("linear_29_cast_fp16")];
            tensor<int32, [1]> t_axes_0 = const()[name = string("t_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2, 512, 1]> t_cast_fp16 = expand_dims(axes = t_axes_0, x = linear_29_cast_fp16)[name = string("t_cast_fp16")];
            tensor<fp16, [2, 512, ?]> var_1754_cast_fp16 = add(x = x_133_cast_fp16, y = t_cast_fp16)[name = string("op_1754_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_135_cast_fp16 = mul(x = var_1754_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_135_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_297_cast_fp16 = mul(x = x_135_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("input_297_cast_fp16")];
            tensor<int32, [6]> input_299_pad_0 = const()[name = string("input_299_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_299_mode_0 = const()[name = string("input_299_mode_0"), val = string("replicate")];
            fp16 const_25_to_fp16 = const()[name = string("const_25_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_299_cast_fp16 = pad(constant_val = const_25_to_fp16, mode = input_299_mode_0, pad = input_299_pad_0, x = input_297_cast_fp16)[name = string("input_299_cast_fp16")];
            string h_135_pad_type_0 = const()[name = string("h_135_pad_type_0"), val = string("valid")];
            int32 h_135_groups_0 = const()[name = string("h_135_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_135_strides_0 = const()[name = string("h_135_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_135_pad_0 = const()[name = string("h_135_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_135_dilations_0 = const()[name = string("h_135_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 5]> main_blocks_3_convnext_1_0_dwconv__conv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50263872))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50266496))))[name = string("main_blocks_3_convnext_1_0_dwconv__conv_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_3_convnext_1_0_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_1_0_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50267584)))];
            tensor<fp16, [2, 512, ?]> h_135_cast_fp16 = conv(bias = main_blocks_3_convnext_1_0_dwconv__conv_bias_to_fp16, dilations = h_135_dilations_0, groups = h_135_groups_0, pad = h_135_pad_0, pad_type = h_135_pad_type_0, strides = h_135_strides_0, weight = main_blocks_3_convnext_1_0_dwconv__conv_weight_to_fp16_quantized, x = input_299_cast_fp16)[name = string("h_135_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_137_cast_fp16 = mul(x = h_135_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_137_cast_fp16")];
            tensor<int32, [3]> input_301_perm_0 = const()[name = string("input_301_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1778_axes_0 = const()[name = string("op_1778_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_3_convnext_1_0_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_3_convnext_1_0_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50268672)))];
            tensor<fp16, [512]> main_blocks_3_convnext_1_0_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_1_0_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50269760)))];
            tensor<fp16, [2, ?, 512]> input_301_cast_fp16 = transpose(perm = input_301_perm_0, x = x_137_cast_fp16)[name = string("transpose_26")];
            tensor<fp16, [2, ?, 512]> var_1778_cast_fp16 = layer_norm(axes = var_1778_axes_0, beta = main_blocks_3_convnext_1_0_norm_norm_bias_to_fp16, epsilon = var_1560_to_fp16, gamma = main_blocks_3_convnext_1_0_norm_norm_weight_to_fp16, x = input_301_cast_fp16)[name = string("op_1778_cast_fp16")];
            tensor<int32, [3]> input_303_perm_0 = const()[name = string("input_303_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_137_pad_type_0 = const()[name = string("h_137_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_137_strides_0 = const()[name = string("h_137_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_137_pad_0 = const()[name = string("h_137_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_137_dilations_0 = const()[name = string("h_137_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_137_groups_0 = const()[name = string("h_137_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_3_convnext_1_0_pwconv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50270848))), scale = tensor<fp16, [2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51319488))))[name = string("main_blocks_3_convnext_1_0_pwconv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> main_blocks_3_convnext_1_0_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_1_0_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51323648)))];
            tensor<fp16, [2, 512, ?]> input_303_cast_fp16 = transpose(perm = input_303_perm_0, x = var_1778_cast_fp16)[name = string("transpose_25")];
            tensor<fp16, [2, 2048, ?]> h_137_cast_fp16 = conv(bias = main_blocks_3_convnext_1_0_pwconv1_bias_to_fp16, dilations = h_137_dilations_0, groups = h_137_groups_0, pad = h_137_pad_0, pad_type = h_137_pad_type_0, strides = h_137_strides_0, weight = main_blocks_3_convnext_1_0_pwconv1_weight_to_fp16_quantized, x = input_303_cast_fp16)[name = string("h_137_cast_fp16")];
            string input_305_mode_0 = const()[name = string("input_305_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_305_cast_fp16 = gelu(mode = input_305_mode_0, x = h_137_cast_fp16)[name = string("input_305_cast_fp16")];
            string h_139_pad_type_0 = const()[name = string("h_139_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_139_strides_0 = const()[name = string("h_139_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_139_pad_0 = const()[name = string("h_139_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_139_dilations_0 = const()[name = string("h_139_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_139_groups_0 = const()[name = string("h_139_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> op_1795_weight_0_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51327808))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52376448))))[name = string("op_1795_weight_0_to_fp16_quantized")];
            tensor<fp16, [512]> var_1795_bias_0_to_fp16 = const()[name = string("op_1795_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52377536)))];
            tensor<fp16, [2, 512, ?]> var_1795_cast_fp16 = conv(bias = var_1795_bias_0_to_fp16, dilations = h_139_dilations_0, groups = h_139_groups_0, pad = h_139_pad_0, pad_type = h_139_pad_type_0, strides = h_139_strides_0, weight = op_1795_weight_0_to_fp16_quantized, x = input_305_cast_fp16)[name = string("op_1795_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_57_cast_fp16 = add(x = input_297_cast_fp16, y = var_1795_cast_fp16)[name = string("out_57_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_139_cast_fp16 = mul(x = out_57_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_139_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_m_13_cast_fp16 = mul(x = x_139_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_m_13_cast_fp16")];
            tensor<int32, [3]> var_1805_shape_cast_fp16 = shape(x = x_139_cast_fp16)[name = string("op_1805_shape_cast_fp16")];
            int32 gather_19_axis_0 = const()[name = string("gather_19_axis_0"), val = int32(0)];
            int32 gather_19_batch_dims_0 = const()[name = string("gather_19_batch_dims_0"), val = int32(0)];
            bool gather_19_validate_indices_0 = const()[name = string("gather_19_validate_indices_0"), val = bool(false)];
            string var_1805_shape_cast_fp16_to_uint16_dtype_0 = const()[name = string("op_1805_shape_cast_fp16_to_uint16_dtype_0"), val = string("uint16")];
            uint16 gather_19_indices_0_to_uint16 = const()[name = string("gather_19_indices_0_to_uint16"), val = uint16(2)];
            tensor<uint16, [3]> var_1805_shape_cast_fp16_to_uint16 = cast(dtype = var_1805_shape_cast_fp16_to_uint16_dtype_0, x = var_1805_shape_cast_fp16)[name = string("cast_2")];
            uint16 gather_19_cast_uint16 = gather(axis = gather_19_axis_0, batch_dims = gather_19_batch_dims_0, indices = gather_19_indices_0_to_uint16, validate_indices = gather_19_validate_indices_0, x = var_1805_shape_cast_fp16_to_uint16)[name = string("gather_19_cast_uint16")];
            string gather_19_cast_uint16_to_int32_dtype_0 = const()[name = string("gather_19_cast_uint16_to_int32_dtype_0"), val = string("int32")];
            tensor<int32, [3]> input_307_perm_0 = const()[name = string("input_307_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [512, 512]> main_blocks_3_text_attn_W_query_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52378624))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52640832))))[name = string("main_blocks_3_text_attn_W_query_linear_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_3_text_attn_W_query_linear_bias_to_fp16 = const()[name = string("main_blocks_3_text_attn_W_query_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52641920)))];
            tensor<fp16, [2, ?, 512]> input_307_cast_fp16 = transpose(perm = input_307_perm_0, x = x_139_cast_fp16)[name = string("transpose_24")];
            tensor<fp16, [2, ?, 512]> linear_30_cast_fp16 = linear(bias = main_blocks_3_text_attn_W_query_linear_bias_to_fp16, weight = main_blocks_3_text_attn_W_query_linear_weight_to_fp16_quantized, x = input_307_cast_fp16)[name = string("linear_30_cast_fp16")];
            tensor<int32, [4]> concat_34x = const()[name = string("concat_34x"), val = tensor<int32, [4]>([2, -1, 8, 64])];
            tensor<fp16, [2, ?, 8, 64]> var_1814_cast_fp16 = reshape(shape = concat_34x, x = linear_30_cast_fp16)[name = string("op_1814_cast_fp16")];
            tensor<int32, [4]> x_141_perm_0 = const()[name = string("x_141_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [512, 256]> main_blocks_3_text_attn_W_key_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52643008))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52774144))))[name = string("main_blocks_3_text_attn_W_key_linear_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_3_text_attn_W_key_linear_bias_to_fp16 = const()[name = string("main_blocks_3_text_attn_W_key_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52775232)))];
            tensor<fp16, [?, ?, 512]> linear_31_cast_fp16 = linear(bias = main_blocks_3_text_attn_W_key_linear_bias_to_fp16, weight = main_blocks_3_text_attn_W_key_linear_weight_to_fp16_quantized, x = input_61_cast_fp16)[name = string("linear_31_cast_fp16")];
            tensor<int32, [4]> concat_35x = const()[name = string("concat_35x"), val = tensor<int32, [4]>([2, -1, 8, 64])];
            tensor<fp16, [2, ?, 8, 64]> var_1822_cast_fp16 = reshape(shape = concat_35x, x = linear_31_cast_fp16)[name = string("op_1822_cast_fp16")];
            tensor<int32, [4]> x_143_perm_0 = const()[name = string("x_143_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [512, 256]> main_blocks_3_text_attn_W_value_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52776320))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52907456))))[name = string("main_blocks_3_text_attn_W_value_linear_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_3_text_attn_W_value_linear_bias_to_fp16 = const()[name = string("main_blocks_3_text_attn_W_value_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52908544)))];
            tensor<fp16, [?, ?, 512]> linear_32_cast_fp16 = linear(bias = main_blocks_3_text_attn_W_value_linear_bias_to_fp16, weight = main_blocks_3_text_attn_W_value_linear_weight_to_fp16_quantized, x = input_61_cast_fp16)[name = string("linear_32_cast_fp16")];
            tensor<int32, [4]> concat_36x = const()[name = string("concat_36x"), val = tensor<int32, [4]>([2, -1, 8, 64])];
            tensor<fp16, [2, ?, 8, 64]> var_1830_cast_fp16 = reshape(shape = concat_36x, x = linear_32_cast_fp16)[name = string("op_1830_cast_fp16")];
            tensor<int32, [4]> v_13_perm_0 = const()[name = string("v_13_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            int32 concat_37_values0_0 = const()[name = string("concat_37_values0_0"), val = int32(1)];
            int32 concat_37_values2_0 = const()[name = string("concat_37_values2_0"), val = int32(1)];
            int32 concat_37_axis_0 = const()[name = string("concat_37_axis_0"), val = int32(0)];
            bool concat_37_interleave_0 = const()[name = string("concat_37_interleave_0"), val = bool(false)];
            int32 gather_19_cast_uint16_to_int32 = cast(dtype = gather_19_cast_uint16_to_int32_dtype_0, x = gather_19_cast_uint16)[name = string("cast_1")];
            tensor<int32, [3]> concat_37 = concat(axis = concat_37_axis_0, interleave = concat_37_interleave_0, values = (concat_37_values0_0, gather_19_cast_uint16_to_int32, concat_37_values2_0))[name = string("concat_37")];
            tensor<int32, [3]> var_1833_begin_0 = const()[name = string("op_1833_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<bool, [3]> var_1833_end_mask_0 = const()[name = string("op_1833_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, ?, 1]> var_1833_cast_fp16 = slice_by_index(begin = var_1833_begin_0, end = concat_37, end_mask = var_1833_end_mask_0, x = main_blocks_0_text_attn_increments_to_fp16)[name = string("op_1833_cast_fp16")];
            tensor<int32, [3]> concat_38 = const()[name = string("concat_38"), val = tensor<int32, [3]>([2, -1, -1])];
            tensor<int32, [3]> shape_7_cast_fp16 = shape(x = var_1833_cast_fp16)[name = string("shape_7_cast_fp16")];
            tensor<bool, [3]> equal_7 = const()[name = string("equal_7"), val = tensor<bool, [3]>([false, true, true])];
            tensor<int32, [3]> select_7 = select(a = shape_7_cast_fp16, b = concat_38, cond = equal_7)[name = string("select_7")];
            tensor<int32, [3]> real_div_7 = real_div(x = select_7, y = shape_7_cast_fp16)[name = string("real_div_7")];
            tensor<fp16, [?, ?, ?]> var_1836_cast_fp16 = tile(reps = real_div_7, x = var_1833_cast_fp16)[name = string("op_1836_cast_fp16")];
            tensor<fp16, [2, ?, ?]> scaled_13_cast_fp16 = real_div(x = var_1836_cast_fp16, y = var_427_cast_fp16)[name = string("scaled_13_cast_fp16")];
            tensor<fp16, [1, 1, 32]> main_blocks_3_text_attn_theta_to_fp16 = const()[name = string("main_blocks_3_text_attn_theta_to_fp16"), val = tensor<fp16, [1, 1, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52909632)))];
            tensor<fp16, [2, ?, 32]> angles_13_cast_fp16 = mul(x = scaled_13_cast_fp16, y = main_blocks_3_text_attn_theta_to_fp16)[name = string("angles_13_cast_fp16")];
            tensor<fp16, [2, ?, 32]> var_1852_cast_fp16 = cos(x = angles_13_cast_fp16)[name = string("op_1852_cast_fp16")];
            tensor<int32, [1]> cos_13_axes_0 = const()[name = string("cos_13_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, ?, 32]> cos_13_cast_fp16 = expand_dims(axes = cos_13_axes_0, x = var_1852_cast_fp16)[name = string("cos_13_cast_fp16")];
            tensor<fp16, [2, ?, 32]> var_1854_cast_fp16 = sin(x = angles_13_cast_fp16)[name = string("op_1854_cast_fp16")];
            tensor<int32, [1]> sin_13_axes_0 = const()[name = string("sin_13_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, ?, 32]> sin_13_cast_fp16 = expand_dims(axes = sin_13_axes_0, x = var_1854_cast_fp16)[name = string("sin_13_cast_fp16")];
            tensor<int32, [4]> x_a_13_begin_0 = const()[name = string("x_a_13_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x_a_13_end_0 = const()[name = string("x_a_13_end_0"), val = tensor<int32, [4]>([2, 8, 0, 32])];
            tensor<bool, [4]> x_a_13_end_mask_0 = const()[name = string("x_a_13_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 8, ?, 64]> x_141_cast_fp16 = transpose(perm = x_141_perm_0, x = var_1814_cast_fp16)[name = string("transpose_23")];
            tensor<fp16, [2, 8, ?, 32]> x_a_13_cast_fp16 = slice_by_index(begin = x_a_13_begin_0, end = x_a_13_end_0, end_mask = x_a_13_end_mask_0, x = x_141_cast_fp16)[name = string("x_a_13_cast_fp16")];
            tensor<int32, [4]> x_b_13_begin_0 = const()[name = string("x_b_13_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x_b_13_end_0 = const()[name = string("x_b_13_end_0"), val = tensor<int32, [4]>([2, 8, 0, 64])];
            tensor<bool, [4]> x_b_13_end_mask_0 = const()[name = string("x_b_13_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [2, 8, ?, 32]> x_b_13_cast_fp16 = slice_by_index(begin = x_b_13_begin_0, end = x_b_13_end_0, end_mask = x_b_13_end_mask_0, x = x_141_cast_fp16)[name = string("x_b_13_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1858_cast_fp16 = mul(x = x_a_13_cast_fp16, y = cos_13_cast_fp16)[name = string("op_1858_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1859_cast_fp16 = mul(x = x_b_13_cast_fp16, y = sin_13_cast_fp16)[name = string("op_1859_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_a_13_cast_fp16 = sub(x = var_1858_cast_fp16, y = var_1859_cast_fp16)[name = string("rot_a_13_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1861_cast_fp16 = mul(x = x_a_13_cast_fp16, y = sin_13_cast_fp16)[name = string("op_1861_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1862_cast_fp16 = mul(x = x_b_13_cast_fp16, y = cos_13_cast_fp16)[name = string("op_1862_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_b_13_cast_fp16 = add(x = var_1861_cast_fp16, y = var_1862_cast_fp16)[name = string("rot_b_13_cast_fp16")];
            bool q_13_interleave_0 = const()[name = string("q_13_interleave_0"), val = bool(false)];
            tensor<fp16, [2, 8, ?, 64]> q_13_cast_fp16 = concat(axis = var_1548, interleave = q_13_interleave_0, values = (rot_a_13_cast_fp16, rot_b_13_cast_fp16))[name = string("q_13_cast_fp16")];
            tensor<int32, [4]> x_a_begin_0 = const()[name = string("x_a_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x_a_end_0 = const()[name = string("x_a_end_0"), val = tensor<int32, [4]>([2, 8, 0, 32])];
            tensor<bool, [4]> x_a_end_mask_0 = const()[name = string("x_a_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 8, ?, 64]> x_143_cast_fp16 = transpose(perm = x_143_perm_0, x = var_1822_cast_fp16)[name = string("transpose_22")];
            tensor<fp16, [2, 8, ?, 32]> x_a_cast_fp16 = slice_by_index(begin = x_a_begin_0, end = x_a_end_0, end_mask = x_a_end_mask_0, x = x_143_cast_fp16)[name = string("x_a_cast_fp16")];
            tensor<int32, [4]> x_b_begin_0 = const()[name = string("x_b_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x_b_end_0 = const()[name = string("x_b_end_0"), val = tensor<int32, [4]>([2, 8, 0, 64])];
            tensor<bool, [4]> x_b_end_mask_0 = const()[name = string("x_b_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [2, 8, ?, 32]> x_b_cast_fp16 = slice_by_index(begin = x_b_begin_0, end = x_b_end_0, end_mask = x_b_end_mask_0, x = x_143_cast_fp16)[name = string("x_b_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1876_cast_fp16 = mul(x = x_a_cast_fp16, y = cos_3_cast_fp16)[name = string("op_1876_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1877_cast_fp16 = mul(x = x_b_cast_fp16, y = sin_3_cast_fp16)[name = string("op_1877_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_a_cast_fp16 = sub(x = var_1876_cast_fp16, y = var_1877_cast_fp16)[name = string("rot_a_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1879_cast_fp16 = mul(x = x_a_cast_fp16, y = sin_3_cast_fp16)[name = string("op_1879_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> var_1880_cast_fp16 = mul(x = x_b_cast_fp16, y = cos_3_cast_fp16)[name = string("op_1880_cast_fp16")];
            tensor<fp16, [2, 8, ?, 32]> rot_b_cast_fp16 = add(x = var_1879_cast_fp16, y = var_1880_cast_fp16)[name = string("rot_b_cast_fp16")];
            bool k_13_interleave_0 = const()[name = string("k_13_interleave_0"), val = bool(false)];
            tensor<fp16, [2, 8, ?, 64]> k_13_cast_fp16 = concat(axis = var_1548, interleave = k_13_interleave_0, values = (rot_a_cast_fp16, rot_b_cast_fp16))[name = string("k_13_cast_fp16")];
            bool var_1885_transpose_x_1 = const()[name = string("op_1885_transpose_x_1"), val = bool(false)];
            bool var_1885_transpose_y_1 = const()[name = string("op_1885_transpose_y_1"), val = bool(true)];
            tensor<fp16, [2, 8, ?, ?]> var_1885_cast_fp16 = matmul(transpose_x = var_1885_transpose_x_1, transpose_y = var_1885_transpose_y_1, x = q_13_cast_fp16, y = k_13_cast_fp16)[name = string("op_1885_cast_fp16")];
            fp16 _inversed_scores_y_0_to_fp16 = const()[name = string("_inversed_scores_y_0_to_fp16"), val = fp16(0x1p-4)];
            tensor<fp16, [2, 8, ?, ?]> _inversed_scores_cast_fp16 = mul(x = var_1885_cast_fp16, y = _inversed_scores_y_0_to_fp16)[name = string("_inversed_scores_cast_fp16")];
            tensor<fp16, [2, 8, ?, ?]> input_313_cast_fp16 = sub(x = _inversed_scores_cast_fp16, y = var_469_cast_fp16)[name = string("input_313_cast_fp16")];
            tensor<fp16, [2, 8, ?, ?]> attn_19_cast_fp16 = softmax(axis = var_1548, x = input_313_cast_fp16)[name = string("attn_19_cast_fp16")];
            bool var_1894_transpose_x_0 = const()[name = string("op_1894_transpose_x_0"), val = bool(false)];
            bool var_1894_transpose_y_0 = const()[name = string("op_1894_transpose_y_0"), val = bool(false)];
            tensor<fp16, [2, 8, ?, 64]> v_13_cast_fp16 = transpose(perm = v_13_perm_0, x = var_1830_cast_fp16)[name = string("transpose_21")];
            tensor<fp16, [2, 8, ?, 64]> var_1894_cast_fp16 = matmul(transpose_x = var_1894_transpose_x_0, transpose_y = var_1894_transpose_y_0, x = attn_19_cast_fp16, y = v_13_cast_fp16)[name = string("op_1894_cast_fp16")];
            tensor<int32, [4]> var_1895_perm_0 = const()[name = string("op_1895_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_41x = const()[name = string("concat_41x"), val = tensor<int32, [3]>([2, -1, 512])];
            tensor<fp16, [2, ?, 8, 64]> var_1895_cast_fp16 = transpose(perm = var_1895_perm_0, x = var_1894_cast_fp16)[name = string("transpose_20")];
            tensor<fp16, [2, ?, 512]> input_315_cast_fp16 = reshape(shape = concat_41x, x = var_1895_cast_fp16)[name = string("input_315_cast_fp16")];
            tensor<fp16, [512, 512]> main_blocks_3_text_attn_out_fc_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52909760))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53171968))))[name = string("main_blocks_3_text_attn_out_fc_linear_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_3_text_attn_out_fc_linear_bias_to_fp16 = const()[name = string("main_blocks_3_text_attn_out_fc_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53173056)))];
            tensor<fp16, [2, ?, 512]> linear_33_cast_fp16 = linear(bias = main_blocks_3_text_attn_out_fc_linear_bias_to_fp16, weight = main_blocks_3_text_attn_out_fc_linear_weight_to_fp16_quantized, x = input_315_cast_fp16)[name = string("linear_33_cast_fp16")];
            tensor<int32, [3]> out_59_perm_0 = const()[name = string("out_59_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> out_59_cast_fp16 = transpose(perm = out_59_perm_0, x = linear_33_cast_fp16)[name = string("transpose_19")];
            tensor<fp16, [2, 512, ?]> var_1903_cast_fp16 = mul(x = out_59_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_1903_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_145_cast_fp16 = add(x = x_m_13_cast_fp16, y = var_1903_cast_fp16)[name = string("x_145_cast_fp16")];
            tensor<int32, [3]> input_317_perm_0 = const()[name = string("input_317_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1910_axes_0 = const()[name = string("op_1910_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_3_text_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_3_text_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53174144)))];
            tensor<fp16, [512]> main_blocks_3_text_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_3_text_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53175232)))];
            tensor<fp16, [2, ?, 512]> input_317_cast_fp16 = transpose(perm = input_317_perm_0, x = x_145_cast_fp16)[name = string("transpose_18")];
            tensor<fp16, [2, ?, 512]> var_1910_cast_fp16 = layer_norm(axes = var_1910_axes_0, beta = main_blocks_3_text_norm_norm_bias_to_fp16, epsilon = var_1560_to_fp16, gamma = main_blocks_3_text_norm_norm_weight_to_fp16, x = input_317_cast_fp16)[name = string("op_1910_cast_fp16")];
            tensor<int32, [3]> var_1911_perm_0 = const()[name = string("op_1911_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> var_1911_cast_fp16 = transpose(perm = var_1911_perm_0, x = var_1910_cast_fp16)[name = string("transpose_17")];
            tensor<fp16, [2, 512, ?]> x_147_cast_fp16 = mul(x = var_1911_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_147_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_319_cast_fp16 = mul(x = x_147_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("input_319_cast_fp16")];
            tensor<int32, [6]> input_321_pad_0 = const()[name = string("input_321_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_321_mode_0 = const()[name = string("input_321_mode_0"), val = string("replicate")];
            fp16 const_26_to_fp16 = const()[name = string("const_26_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_321_cast_fp16 = pad(constant_val = const_26_to_fp16, mode = input_321_mode_0, pad = input_321_pad_0, x = input_319_cast_fp16)[name = string("input_321_cast_fp16")];
            string h_141_pad_type_0 = const()[name = string("h_141_pad_type_0"), val = string("valid")];
            int32 h_141_groups_0 = const()[name = string("h_141_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_141_strides_0 = const()[name = string("h_141_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_141_pad_0 = const()[name = string("h_141_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_141_dilations_0 = const()[name = string("h_141_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 5]> main_blocks_3_convnext_2_0_dwconv__conv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53176320))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53178944))))[name = string("main_blocks_3_convnext_2_0_dwconv__conv_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_3_convnext_2_0_dwconv__conv_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_2_0_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53180032)))];
            tensor<fp16, [2, 512, ?]> h_141_cast_fp16 = conv(bias = main_blocks_3_convnext_2_0_dwconv__conv_bias_to_fp16, dilations = h_141_dilations_0, groups = h_141_groups_0, pad = h_141_pad_0, pad_type = h_141_pad_type_0, strides = h_141_strides_0, weight = main_blocks_3_convnext_2_0_dwconv__conv_weight_to_fp16_quantized, x = input_321_cast_fp16)[name = string("h_141_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_149_cast_fp16 = mul(x = h_141_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_149_cast_fp16")];
            tensor<int32, [3]> input_323_perm_0 = const()[name = string("input_323_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1935_axes_0 = const()[name = string("op_1935_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_3_convnext_2_0_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_3_convnext_2_0_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53181120)))];
            tensor<fp16, [512]> main_blocks_3_convnext_2_0_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_2_0_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53182208)))];
            tensor<fp16, [2, ?, 512]> input_323_cast_fp16 = transpose(perm = input_323_perm_0, x = x_149_cast_fp16)[name = string("transpose_16")];
            tensor<fp16, [2, ?, 512]> var_1935_cast_fp16 = layer_norm(axes = var_1935_axes_0, beta = main_blocks_3_convnext_2_0_norm_norm_bias_to_fp16, epsilon = var_1560_to_fp16, gamma = main_blocks_3_convnext_2_0_norm_norm_weight_to_fp16, x = input_323_cast_fp16)[name = string("op_1935_cast_fp16")];
            tensor<int32, [3]> input_325_perm_0 = const()[name = string("input_325_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_143_pad_type_0 = const()[name = string("h_143_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_143_strides_0 = const()[name = string("h_143_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_143_pad_0 = const()[name = string("h_143_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_143_dilations_0 = const()[name = string("h_143_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_143_groups_0 = const()[name = string("h_143_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> main_blocks_3_convnext_2_0_pwconv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53183296))), scale = tensor<fp16, [2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54231936))))[name = string("main_blocks_3_convnext_2_0_pwconv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> main_blocks_3_convnext_2_0_pwconv1_bias_to_fp16 = const()[name = string("main_blocks_3_convnext_2_0_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54236096)))];
            tensor<fp16, [2, 512, ?]> input_325_cast_fp16 = transpose(perm = input_325_perm_0, x = var_1935_cast_fp16)[name = string("transpose_15")];
            tensor<fp16, [2, 2048, ?]> h_143_cast_fp16 = conv(bias = main_blocks_3_convnext_2_0_pwconv1_bias_to_fp16, dilations = h_143_dilations_0, groups = h_143_groups_0, pad = h_143_pad_0, pad_type = h_143_pad_type_0, strides = h_143_strides_0, weight = main_blocks_3_convnext_2_0_pwconv1_weight_to_fp16_quantized, x = input_325_cast_fp16)[name = string("h_143_cast_fp16")];
            string input_327_mode_0 = const()[name = string("input_327_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_327_cast_fp16 = gelu(mode = input_327_mode_0, x = h_143_cast_fp16)[name = string("input_327_cast_fp16")];
            string h_145_pad_type_0 = const()[name = string("h_145_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_145_strides_0 = const()[name = string("h_145_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_145_pad_0 = const()[name = string("h_145_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_145_dilations_0 = const()[name = string("h_145_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_145_groups_0 = const()[name = string("h_145_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> op_1952_weight_0_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54240256))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55288896))))[name = string("op_1952_weight_0_to_fp16_quantized")];
            tensor<fp16, [512]> var_1952_bias_0_to_fp16 = const()[name = string("op_1952_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55289984)))];
            tensor<fp16, [2, 512, ?]> var_1952_cast_fp16 = conv(bias = var_1952_bias_0_to_fp16, dilations = h_145_dilations_0, groups = h_145_groups_0, pad = h_145_pad_0, pad_type = h_145_pad_type_0, strides = h_145_strides_0, weight = op_1952_weight_0_to_fp16_quantized, x = input_327_cast_fp16)[name = string("op_1952_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_61_cast_fp16 = add(x = input_319_cast_fp16, y = var_1952_cast_fp16)[name = string("out_61_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_151_cast_fp16 = mul(x = out_61_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_151_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_m_cast_fp16 = mul(x = x_151_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_m_cast_fp16")];
            tensor<int32, [3]> input_329_perm_0 = const()[name = string("input_329_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [256, 512]> main_blocks_3_style_attn_W_query_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [256, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55291072))), scale = tensor<fp16, [256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55422208))))[name = string("main_blocks_3_style_attn_W_query_linear_weight_to_fp16_quantized")];
            tensor<fp16, [256]> main_blocks_3_style_attn_W_query_linear_bias_to_fp16 = const()[name = string("main_blocks_3_style_attn_W_query_linear_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55422784)))];
            tensor<fp16, [2, ?, 512]> input_329_cast_fp16 = transpose(perm = input_329_perm_0, x = x_151_cast_fp16)[name = string("transpose_14")];
            tensor<fp16, [2, ?, 256]> linear_34_cast_fp16 = linear(bias = main_blocks_3_style_attn_W_query_linear_bias_to_fp16, weight = main_blocks_3_style_attn_W_query_linear_weight_to_fp16_quantized, x = input_329_cast_fp16)[name = string("linear_34_cast_fp16")];
            tensor<int32, [4]> concat_42x = const()[name = string("concat_42x"), val = tensor<int32, [4]>([2, -1, 2, 128])];
            tensor<fp16, [2, ?, 2, 128]> var_1969_cast_fp16 = reshape(shape = concat_42x, x = linear_34_cast_fp16)[name = string("op_1969_cast_fp16")];
            tensor<int32, [4]> q_perm_0 = const()[name = string("q_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [256, 256]> main_blocks_3_style_attn_W_value_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [256, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55423360))), scale = tensor<fp16, [256, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55488960))))[name = string("main_blocks_3_style_attn_W_value_linear_weight_to_fp16_quantized")];
            tensor<fp16, [256]> main_blocks_3_style_attn_W_value_linear_bias_to_fp16 = const()[name = string("main_blocks_3_style_attn_W_value_linear_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55489536)))];
            tensor<fp16, [2, 50, 256]> linear_36_cast_fp16 = linear(bias = main_blocks_3_style_attn_W_value_linear_bias_to_fp16, weight = main_blocks_3_style_attn_W_value_linear_weight_to_fp16_quantized, x = input_83_cast_fp16)[name = string("linear_36_cast_fp16")];
            tensor<int32, [4]> var_1982 = const()[name = string("op_1982"), val = tensor<int32, [4]>([2, 50, 2, 128])];
            tensor<fp16, [2, 50, 2, 128]> var_1983_cast_fp16 = reshape(shape = var_1982, x = linear_36_cast_fp16)[name = string("op_1983_cast_fp16")];
            tensor<int32, [4]> v_15_perm_0 = const()[name = string("v_15_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool var_1987_transpose_x_0 = const()[name = string("op_1987_transpose_x_0"), val = bool(false)];
            bool var_1987_transpose_y_0 = const()[name = string("op_1987_transpose_y_0"), val = bool(false)];
            tensor<fp16, [2, 2, 128, 50]> op_1986_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2, 2, 128, 50]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55490112))), scale = tensor<fp16, [1, 1, 1, 50]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55515776))))[name = string("op_1986_to_fp16_quantized")];
            tensor<fp16, [2, 2, ?, 128]> q_cast_fp16 = transpose(perm = q_perm_0, x = var_1969_cast_fp16)[name = string("transpose_13")];
            tensor<fp16, [2, 2, ?, 50]> var_1987_cast_fp16 = matmul(transpose_x = var_1987_transpose_x_0, transpose_y = var_1987_transpose_y_0, x = q_cast_fp16, y = op_1986_to_fp16_quantized)[name = string("op_1987_cast_fp16")];
            fp16 _inversed_input_331_y_0_to_fp16 = const()[name = string("_inversed_input_331_y_0_to_fp16"), val = fp16(0x1p-4)];
            tensor<fp16, [2, 2, ?, 50]> _inversed_input_331_cast_fp16 = mul(x = var_1987_cast_fp16, y = _inversed_input_331_y_0_to_fp16)[name = string("_inversed_input_331_cast_fp16")];
            tensor<fp16, [2, 2, ?, 50]> attn_21_cast_fp16 = softmax(axis = var_1548, x = _inversed_input_331_cast_fp16)[name = string("attn_21_cast_fp16")];
            tensor<fp16, [2, 2, ?, 50]> attn_cast_fp16 = mul(x = attn_21_cast_fp16, y = mask_1_cast_fp16)[name = string("attn_cast_fp16")];
            bool var_1994_transpose_x_0 = const()[name = string("op_1994_transpose_x_0"), val = bool(false)];
            bool var_1994_transpose_y_0 = const()[name = string("op_1994_transpose_y_0"), val = bool(false)];
            tensor<fp16, [2, 2, 50, 128]> v_15_cast_fp16 = transpose(perm = v_15_perm_0, x = var_1983_cast_fp16)[name = string("transpose_12")];
            tensor<fp16, [2, 2, ?, 128]> var_1994_cast_fp16 = matmul(transpose_x = var_1994_transpose_x_0, transpose_y = var_1994_transpose_y_0, x = attn_cast_fp16, y = v_15_cast_fp16)[name = string("op_1994_cast_fp16")];
            tensor<int32, [4]> var_1995_perm_0 = const()[name = string("op_1995_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_43x = const()[name = string("concat_43x"), val = tensor<int32, [3]>([2, -1, 256])];
            tensor<fp16, [2, ?, 2, 128]> var_1995_cast_fp16 = transpose(perm = var_1995_perm_0, x = var_1994_cast_fp16)[name = string("transpose_11")];
            tensor<fp16, [2, ?, 256]> input_333_cast_fp16 = reshape(shape = concat_43x, x = var_1995_cast_fp16)[name = string("input_333_cast_fp16")];
            tensor<fp16, [512, 256]> main_blocks_3_style_attn_out_fc_linear_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55515968))), scale = tensor<fp16, [512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55647104))))[name = string("main_blocks_3_style_attn_out_fc_linear_weight_to_fp16_quantized")];
            tensor<fp16, [512]> main_blocks_3_style_attn_out_fc_linear_bias_to_fp16 = const()[name = string("main_blocks_3_style_attn_out_fc_linear_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55648192)))];
            tensor<fp16, [2, ?, 512]> linear_37_cast_fp16 = linear(bias = main_blocks_3_style_attn_out_fc_linear_bias_to_fp16, weight = main_blocks_3_style_attn_out_fc_linear_weight_to_fp16_quantized, x = input_333_cast_fp16)[name = string("linear_37_cast_fp16")];
            tensor<int32, [3]> out_63_perm_0 = const()[name = string("out_63_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> out_63_cast_fp16 = transpose(perm = out_63_perm_0, x = linear_37_cast_fp16)[name = string("transpose_10")];
            tensor<fp16, [2, 512, ?]> var_2003_cast_fp16 = mul(x = out_63_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("op_2003_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_153_cast_fp16 = add(x = x_m_cast_fp16, y = var_2003_cast_fp16)[name = string("x_153_cast_fp16")];
            tensor<int32, [3]> input_335_perm_0 = const()[name = string("input_335_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_2010_axes_0 = const()[name = string("op_2010_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> main_blocks_3_style_norm_norm_weight_to_fp16 = const()[name = string("main_blocks_3_style_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55649280)))];
            tensor<fp16, [512]> main_blocks_3_style_norm_norm_bias_to_fp16 = const()[name = string("main_blocks_3_style_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55650368)))];
            tensor<fp16, [2, ?, 512]> input_335_cast_fp16 = transpose(perm = input_335_perm_0, x = x_153_cast_fp16)[name = string("transpose_9")];
            tensor<fp16, [2, ?, 512]> var_2010_cast_fp16 = layer_norm(axes = var_2010_axes_0, beta = main_blocks_3_style_norm_norm_bias_to_fp16, epsilon = var_1560_to_fp16, gamma = main_blocks_3_style_norm_norm_weight_to_fp16, x = input_335_cast_fp16)[name = string("op_2010_cast_fp16")];
            tensor<int32, [3]> var_2011_perm_0 = const()[name = string("op_2011_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [2, 512, ?]> var_2011_cast_fp16 = transpose(perm = var_2011_perm_0, x = var_2010_cast_fp16)[name = string("transpose_8")];
            tensor<fp16, [2, 512, ?]> x_155_cast_fp16 = mul(x = var_2011_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_155_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_337_cast_fp16 = mul(x = x_155_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("input_337_cast_fp16")];
            tensor<int32, [6]> input_339_pad_0 = const()[name = string("input_339_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_339_mode_0 = const()[name = string("input_339_mode_0"), val = string("replicate")];
            fp16 const_28_to_fp16 = const()[name = string("const_28_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_339_cast_fp16 = pad(constant_val = const_28_to_fp16, mode = input_339_mode_0, pad = input_339_pad_0, x = input_337_cast_fp16)[name = string("input_339_cast_fp16")];
            string h_147_pad_type_0 = const()[name = string("h_147_pad_type_0"), val = string("valid")];
            int32 h_147_groups_0 = const()[name = string("h_147_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_147_strides_0 = const()[name = string("h_147_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_147_pad_0 = const()[name = string("h_147_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_147_dilations_0 = const()[name = string("h_147_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 5]> last_convnext_0_dwconv__conv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55651456))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55654080))))[name = string("last_convnext_0_dwconv__conv_weight_to_fp16_quantized")];
            tensor<fp16, [512]> last_convnext_0_dwconv__conv_bias_to_fp16 = const()[name = string("last_convnext_0_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55655168)))];
            tensor<fp16, [2, 512, ?]> h_147_cast_fp16 = conv(bias = last_convnext_0_dwconv__conv_bias_to_fp16, dilations = h_147_dilations_0, groups = h_147_groups_0, pad = h_147_pad_0, pad_type = h_147_pad_type_0, strides = h_147_strides_0, weight = last_convnext_0_dwconv__conv_weight_to_fp16_quantized, x = input_339_cast_fp16)[name = string("h_147_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_157_cast_fp16 = mul(x = h_147_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_157_cast_fp16")];
            tensor<int32, [3]> input_341_perm_0 = const()[name = string("input_341_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_2045_axes_0 = const()[name = string("op_2045_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> last_convnext_0_norm_norm_weight_to_fp16 = const()[name = string("last_convnext_0_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55656256)))];
            tensor<fp16, [512]> last_convnext_0_norm_norm_bias_to_fp16 = const()[name = string("last_convnext_0_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55657344)))];
            fp16 var_2015_to_fp16 = const()[name = string("op_2015_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [2, ?, 512]> input_341_cast_fp16 = transpose(perm = input_341_perm_0, x = x_157_cast_fp16)[name = string("transpose_7")];
            tensor<fp16, [2, ?, 512]> var_2045_cast_fp16 = layer_norm(axes = var_2045_axes_0, beta = last_convnext_0_norm_norm_bias_to_fp16, epsilon = var_2015_to_fp16, gamma = last_convnext_0_norm_norm_weight_to_fp16, x = input_341_cast_fp16)[name = string("op_2045_cast_fp16")];
            tensor<int32, [3]> input_343_perm_0 = const()[name = string("input_343_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_149_pad_type_0 = const()[name = string("h_149_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_149_strides_0 = const()[name = string("h_149_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_149_pad_0 = const()[name = string("h_149_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_149_dilations_0 = const()[name = string("h_149_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_149_groups_0 = const()[name = string("h_149_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> last_convnext_0_pwconv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55658432))), scale = tensor<fp16, [2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56707072))))[name = string("last_convnext_0_pwconv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> last_convnext_0_pwconv1_bias_to_fp16 = const()[name = string("last_convnext_0_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56711232)))];
            tensor<fp16, [2, 512, ?]> input_343_cast_fp16 = transpose(perm = input_343_perm_0, x = var_2045_cast_fp16)[name = string("transpose_6")];
            tensor<fp16, [2, 2048, ?]> h_149_cast_fp16 = conv(bias = last_convnext_0_pwconv1_bias_to_fp16, dilations = h_149_dilations_0, groups = h_149_groups_0, pad = h_149_pad_0, pad_type = h_149_pad_type_0, strides = h_149_strides_0, weight = last_convnext_0_pwconv1_weight_to_fp16_quantized, x = input_343_cast_fp16)[name = string("h_149_cast_fp16")];
            string input_345_mode_0 = const()[name = string("input_345_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_345_cast_fp16 = gelu(mode = input_345_mode_0, x = h_149_cast_fp16)[name = string("input_345_cast_fp16")];
            string h_151_pad_type_0 = const()[name = string("h_151_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_151_strides_0 = const()[name = string("h_151_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_151_pad_0 = const()[name = string("h_151_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_151_dilations_0 = const()[name = string("h_151_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_151_groups_0 = const()[name = string("h_151_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> op_2062_weight_0_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56715392))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57764032))))[name = string("op_2062_weight_0_to_fp16_quantized")];
            tensor<fp16, [512]> var_2062_bias_0_to_fp16 = const()[name = string("op_2062_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57765120)))];
            tensor<fp16, [2, 512, ?]> var_2062_cast_fp16 = conv(bias = var_2062_bias_0_to_fp16, dilations = h_151_dilations_0, groups = h_151_groups_0, pad = h_151_pad_0, pad_type = h_151_pad_type_0, strides = h_151_strides_0, weight = op_2062_weight_0_to_fp16_quantized, x = input_345_cast_fp16)[name = string("op_2062_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_65_cast_fp16 = add(x = input_337_cast_fp16, y = var_2062_cast_fp16)[name = string("out_65_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_159_cast_fp16 = mul(x = out_65_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_159_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_347_cast_fp16 = mul(x = x_159_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("input_347_cast_fp16")];
            tensor<int32, [6]> input_349_pad_0 = const()[name = string("input_349_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_349_mode_0 = const()[name = string("input_349_mode_0"), val = string("replicate")];
            fp16 const_29_to_fp16 = const()[name = string("const_29_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_349_cast_fp16 = pad(constant_val = const_29_to_fp16, mode = input_349_mode_0, pad = input_349_pad_0, x = input_347_cast_fp16)[name = string("input_349_cast_fp16")];
            string h_153_pad_type_0 = const()[name = string("h_153_pad_type_0"), val = string("valid")];
            int32 h_153_groups_0 = const()[name = string("h_153_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_153_strides_0 = const()[name = string("h_153_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_153_pad_0 = const()[name = string("h_153_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_153_dilations_0 = const()[name = string("h_153_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 5]> last_convnext_1_dwconv__conv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57766208))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57768832))))[name = string("last_convnext_1_dwconv__conv_weight_to_fp16_quantized")];
            tensor<fp16, [512]> last_convnext_1_dwconv__conv_bias_to_fp16 = const()[name = string("last_convnext_1_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57769920)))];
            tensor<fp16, [2, 512, ?]> h_153_cast_fp16 = conv(bias = last_convnext_1_dwconv__conv_bias_to_fp16, dilations = h_153_dilations_0, groups = h_153_groups_0, pad = h_153_pad_0, pad_type = h_153_pad_type_0, strides = h_153_strides_0, weight = last_convnext_1_dwconv__conv_weight_to_fp16_quantized, x = input_349_cast_fp16)[name = string("h_153_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_161_cast_fp16 = mul(x = h_153_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_161_cast_fp16")];
            tensor<int32, [3]> input_351_perm_0 = const()[name = string("input_351_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_2097_axes_0 = const()[name = string("op_2097_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> last_convnext_1_norm_norm_weight_to_fp16 = const()[name = string("last_convnext_1_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57771008)))];
            tensor<fp16, [512]> last_convnext_1_norm_norm_bias_to_fp16 = const()[name = string("last_convnext_1_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57772096)))];
            fp16 var_2067_to_fp16 = const()[name = string("op_2067_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [2, ?, 512]> input_351_cast_fp16 = transpose(perm = input_351_perm_0, x = x_161_cast_fp16)[name = string("transpose_5")];
            tensor<fp16, [2, ?, 512]> var_2097_cast_fp16 = layer_norm(axes = var_2097_axes_0, beta = last_convnext_1_norm_norm_bias_to_fp16, epsilon = var_2067_to_fp16, gamma = last_convnext_1_norm_norm_weight_to_fp16, x = input_351_cast_fp16)[name = string("op_2097_cast_fp16")];
            tensor<int32, [3]> input_353_perm_0 = const()[name = string("input_353_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_155_pad_type_0 = const()[name = string("h_155_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_155_strides_0 = const()[name = string("h_155_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_155_pad_0 = const()[name = string("h_155_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_155_dilations_0 = const()[name = string("h_155_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_155_groups_0 = const()[name = string("h_155_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> last_convnext_1_pwconv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57773184))), scale = tensor<fp16, [2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58821824))))[name = string("last_convnext_1_pwconv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> last_convnext_1_pwconv1_bias_to_fp16 = const()[name = string("last_convnext_1_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58825984)))];
            tensor<fp16, [2, 512, ?]> input_353_cast_fp16 = transpose(perm = input_353_perm_0, x = var_2097_cast_fp16)[name = string("transpose_4")];
            tensor<fp16, [2, 2048, ?]> h_155_cast_fp16 = conv(bias = last_convnext_1_pwconv1_bias_to_fp16, dilations = h_155_dilations_0, groups = h_155_groups_0, pad = h_155_pad_0, pad_type = h_155_pad_type_0, strides = h_155_strides_0, weight = last_convnext_1_pwconv1_weight_to_fp16_quantized, x = input_353_cast_fp16)[name = string("h_155_cast_fp16")];
            string input_355_mode_0 = const()[name = string("input_355_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_355_cast_fp16 = gelu(mode = input_355_mode_0, x = h_155_cast_fp16)[name = string("input_355_cast_fp16")];
            string h_157_pad_type_0 = const()[name = string("h_157_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_157_strides_0 = const()[name = string("h_157_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_157_pad_0 = const()[name = string("h_157_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_157_dilations_0 = const()[name = string("h_157_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_157_groups_0 = const()[name = string("h_157_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> op_2114_weight_0_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58830144))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59878784))))[name = string("op_2114_weight_0_to_fp16_quantized")];
            tensor<fp16, [512]> var_2114_bias_0_to_fp16 = const()[name = string("op_2114_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59879872)))];
            tensor<fp16, [2, 512, ?]> var_2114_cast_fp16 = conv(bias = var_2114_bias_0_to_fp16, dilations = h_157_dilations_0, groups = h_157_groups_0, pad = h_157_pad_0, pad_type = h_157_pad_type_0, strides = h_157_strides_0, weight = op_2114_weight_0_to_fp16_quantized, x = input_355_cast_fp16)[name = string("op_2114_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_67_cast_fp16 = add(x = input_347_cast_fp16, y = var_2114_cast_fp16)[name = string("out_67_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_163_cast_fp16 = mul(x = out_67_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_163_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_357_cast_fp16 = mul(x = x_163_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("input_357_cast_fp16")];
            tensor<int32, [6]> input_359_pad_0 = const()[name = string("input_359_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_359_mode_0 = const()[name = string("input_359_mode_0"), val = string("replicate")];
            fp16 const_30_to_fp16 = const()[name = string("const_30_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_359_cast_fp16 = pad(constant_val = const_30_to_fp16, mode = input_359_mode_0, pad = input_359_pad_0, x = input_357_cast_fp16)[name = string("input_359_cast_fp16")];
            string h_159_pad_type_0 = const()[name = string("h_159_pad_type_0"), val = string("valid")];
            int32 h_159_groups_0 = const()[name = string("h_159_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_159_strides_0 = const()[name = string("h_159_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_159_pad_0 = const()[name = string("h_159_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_159_dilations_0 = const()[name = string("h_159_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 5]> last_convnext_2_dwconv__conv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59880960))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59883584))))[name = string("last_convnext_2_dwconv__conv_weight_to_fp16_quantized")];
            tensor<fp16, [512]> last_convnext_2_dwconv__conv_bias_to_fp16 = const()[name = string("last_convnext_2_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59884672)))];
            tensor<fp16, [2, 512, ?]> h_159_cast_fp16 = conv(bias = last_convnext_2_dwconv__conv_bias_to_fp16, dilations = h_159_dilations_0, groups = h_159_groups_0, pad = h_159_pad_0, pad_type = h_159_pad_type_0, strides = h_159_strides_0, weight = last_convnext_2_dwconv__conv_weight_to_fp16_quantized, x = input_359_cast_fp16)[name = string("h_159_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_165_cast_fp16 = mul(x = h_159_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_165_cast_fp16")];
            tensor<int32, [3]> input_361_perm_0 = const()[name = string("input_361_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_2149_axes_0 = const()[name = string("op_2149_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> last_convnext_2_norm_norm_weight_to_fp16 = const()[name = string("last_convnext_2_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59885760)))];
            tensor<fp16, [512]> last_convnext_2_norm_norm_bias_to_fp16 = const()[name = string("last_convnext_2_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59886848)))];
            fp16 var_2119_to_fp16 = const()[name = string("op_2119_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [2, ?, 512]> input_361_cast_fp16 = transpose(perm = input_361_perm_0, x = x_165_cast_fp16)[name = string("transpose_3")];
            tensor<fp16, [2, ?, 512]> var_2149_cast_fp16 = layer_norm(axes = var_2149_axes_0, beta = last_convnext_2_norm_norm_bias_to_fp16, epsilon = var_2119_to_fp16, gamma = last_convnext_2_norm_norm_weight_to_fp16, x = input_361_cast_fp16)[name = string("op_2149_cast_fp16")];
            tensor<int32, [3]> input_363_perm_0 = const()[name = string("input_363_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_161_pad_type_0 = const()[name = string("h_161_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_161_strides_0 = const()[name = string("h_161_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_161_pad_0 = const()[name = string("h_161_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_161_dilations_0 = const()[name = string("h_161_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_161_groups_0 = const()[name = string("h_161_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> last_convnext_2_pwconv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59887936))), scale = tensor<fp16, [2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60936576))))[name = string("last_convnext_2_pwconv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> last_convnext_2_pwconv1_bias_to_fp16 = const()[name = string("last_convnext_2_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60940736)))];
            tensor<fp16, [2, 512, ?]> input_363_cast_fp16 = transpose(perm = input_363_perm_0, x = var_2149_cast_fp16)[name = string("transpose_2")];
            tensor<fp16, [2, 2048, ?]> h_161_cast_fp16 = conv(bias = last_convnext_2_pwconv1_bias_to_fp16, dilations = h_161_dilations_0, groups = h_161_groups_0, pad = h_161_pad_0, pad_type = h_161_pad_type_0, strides = h_161_strides_0, weight = last_convnext_2_pwconv1_weight_to_fp16_quantized, x = input_363_cast_fp16)[name = string("h_161_cast_fp16")];
            string input_365_mode_0 = const()[name = string("input_365_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_365_cast_fp16 = gelu(mode = input_365_mode_0, x = h_161_cast_fp16)[name = string("input_365_cast_fp16")];
            string h_163_pad_type_0 = const()[name = string("h_163_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_163_strides_0 = const()[name = string("h_163_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_163_pad_0 = const()[name = string("h_163_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_163_dilations_0 = const()[name = string("h_163_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_163_groups_0 = const()[name = string("h_163_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> op_2166_weight_0_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60944896))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61993536))))[name = string("op_2166_weight_0_to_fp16_quantized")];
            tensor<fp16, [512]> var_2166_bias_0_to_fp16 = const()[name = string("op_2166_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61994624)))];
            tensor<fp16, [2, 512, ?]> var_2166_cast_fp16 = conv(bias = var_2166_bias_0_to_fp16, dilations = h_163_dilations_0, groups = h_163_groups_0, pad = h_163_pad_0, pad_type = h_163_pad_type_0, strides = h_163_strides_0, weight = op_2166_weight_0_to_fp16_quantized, x = input_365_cast_fp16)[name = string("op_2166_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_69_cast_fp16 = add(x = input_357_cast_fp16, y = var_2166_cast_fp16)[name = string("out_69_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_167_cast_fp16 = mul(x = out_69_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_167_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_367_cast_fp16 = mul(x = x_167_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("input_367_cast_fp16")];
            tensor<int32, [6]> input_369_pad_0 = const()[name = string("input_369_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 2])];
            string input_369_mode_0 = const()[name = string("input_369_mode_0"), val = string("replicate")];
            fp16 const_31_to_fp16 = const()[name = string("const_31_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [2, 512, ?]> input_369_cast_fp16 = pad(constant_val = const_31_to_fp16, mode = input_369_mode_0, pad = input_369_pad_0, x = input_367_cast_fp16)[name = string("input_369_cast_fp16")];
            string h_165_pad_type_0 = const()[name = string("h_165_pad_type_0"), val = string("valid")];
            int32 h_165_groups_0 = const()[name = string("h_165_groups_0"), val = int32(512)];
            tensor<int32, [1]> h_165_strides_0 = const()[name = string("h_165_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_165_pad_0 = const()[name = string("h_165_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_165_dilations_0 = const()[name = string("h_165_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 5]> last_convnext_3_dwconv__conv_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 1, 5]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61995712))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61998336))))[name = string("last_convnext_3_dwconv__conv_weight_to_fp16_quantized")];
            tensor<fp16, [512]> last_convnext_3_dwconv__conv_bias_to_fp16 = const()[name = string("last_convnext_3_dwconv__conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61999424)))];
            tensor<fp16, [2, 512, ?]> h_165_cast_fp16 = conv(bias = last_convnext_3_dwconv__conv_bias_to_fp16, dilations = h_165_dilations_0, groups = h_165_groups_0, pad = h_165_pad_0, pad_type = h_165_pad_type_0, strides = h_165_strides_0, weight = last_convnext_3_dwconv__conv_weight_to_fp16_quantized, x = input_369_cast_fp16)[name = string("h_165_cast_fp16")];
            tensor<fp16, [2, 512, ?]> x_cast_fp16 = mul(x = h_165_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("x_cast_fp16")];
            tensor<int32, [3]> input_371_perm_0 = const()[name = string("input_371_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_2201_axes_0 = const()[name = string("op_2201_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> last_convnext_3_norm_norm_weight_to_fp16 = const()[name = string("last_convnext_3_norm_norm_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62000512)))];
            tensor<fp16, [512]> last_convnext_3_norm_norm_bias_to_fp16 = const()[name = string("last_convnext_3_norm_norm_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62001600)))];
            fp16 var_2171_to_fp16 = const()[name = string("op_2171_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [2, ?, 512]> input_371_cast_fp16 = transpose(perm = input_371_perm_0, x = x_cast_fp16)[name = string("transpose_1")];
            tensor<fp16, [2, ?, 512]> var_2201_cast_fp16 = layer_norm(axes = var_2201_axes_0, beta = last_convnext_3_norm_norm_bias_to_fp16, epsilon = var_2171_to_fp16, gamma = last_convnext_3_norm_norm_weight_to_fp16, x = input_371_cast_fp16)[name = string("op_2201_cast_fp16")];
            tensor<int32, [3]> input_373_perm_0 = const()[name = string("input_373_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string h_167_pad_type_0 = const()[name = string("h_167_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_167_strides_0 = const()[name = string("h_167_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_167_pad_0 = const()[name = string("h_167_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_167_dilations_0 = const()[name = string("h_167_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_167_groups_0 = const()[name = string("h_167_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 512, 1]> last_convnext_3_pwconv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2048, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62002688))), scale = tensor<fp16, [2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63051328))))[name = string("last_convnext_3_pwconv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> last_convnext_3_pwconv1_bias_to_fp16 = const()[name = string("last_convnext_3_pwconv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63055488)))];
            tensor<fp16, [2, 512, ?]> input_373_cast_fp16 = transpose(perm = input_373_perm_0, x = var_2201_cast_fp16)[name = string("transpose_0")];
            tensor<fp16, [2, 2048, ?]> h_167_cast_fp16 = conv(bias = last_convnext_3_pwconv1_bias_to_fp16, dilations = h_167_dilations_0, groups = h_167_groups_0, pad = h_167_pad_0, pad_type = h_167_pad_type_0, strides = h_167_strides_0, weight = last_convnext_3_pwconv1_weight_to_fp16_quantized, x = input_373_cast_fp16)[name = string("h_167_cast_fp16")];
            string input_375_mode_0 = const()[name = string("input_375_mode_0"), val = string("EXACT")];
            tensor<fp16, [2, 2048, ?]> input_375_cast_fp16 = gelu(mode = input_375_mode_0, x = h_167_cast_fp16)[name = string("input_375_cast_fp16")];
            string h_pad_type_0 = const()[name = string("h_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> h_strides_0 = const()[name = string("h_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> h_pad_0 = const()[name = string("h_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> h_dilations_0 = const()[name = string("h_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 h_groups_0 = const()[name = string("h_groups_0"), val = int32(1)];
            tensor<fp16, [512, 2048, 1]> op_2218_weight_0_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [512, 2048, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63059648))), scale = tensor<fp16, [512, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64108288))))[name = string("op_2218_weight_0_to_fp16_quantized")];
            tensor<fp16, [512]> var_2218_bias_0_to_fp16 = const()[name = string("op_2218_bias_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64109376)))];
            tensor<fp16, [2, 512, ?]> var_2218_cast_fp16 = conv(bias = var_2218_bias_0_to_fp16, dilations = h_dilations_0, groups = h_groups_0, pad = h_pad_0, pad_type = h_pad_type_0, strides = h_strides_0, weight = op_2218_weight_0_to_fp16_quantized, x = input_375_cast_fp16)[name = string("op_2218_cast_fp16")];
            tensor<fp16, [2, 512, ?]> out_cast_fp16 = add(x = input_367_cast_fp16, y = var_2218_cast_fp16)[name = string("out_cast_fp16")];
            tensor<fp16, [2, 512, ?]> input_cast_fp16 = mul(x = out_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("input_cast_fp16")];
            string var_2231_pad_type_0 = const()[name = string("op_2231_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> var_2231_strides_0 = const()[name = string("op_2231_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_2231_pad_0 = const()[name = string("op_2231_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_2231_dilations_0 = const()[name = string("op_2231_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 var_2231_groups_0 = const()[name = string("op_2231_groups_0"), val = int32(1)];
            tensor<fp16, [144, 512, 1]> proj_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [144, 512, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64110464))), scale = tensor<fp16, [144, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64184256))))[name = string("proj_out_weight_to_fp16_quantized")];
            tensor<fp16, [2, 144, ?]> var_2231_cast_fp16 = conv(dilations = var_2231_dilations_0, groups = var_2231_groups_0, pad = var_2231_pad_0, pad_type = var_2231_pad_type_0, strides = var_2231_strides_0, weight = proj_out_weight_to_fp16_quantized, x = input_cast_fp16)[name = string("op_2231_cast_fp16")];
            tensor<fp16, [2, 144, ?]> v_cast_fp16 = mul(x = var_2231_cast_fp16, y = latent_mask_b_cast_fp16)[name = string("v_cast_fp16")];
            tensor<int32, [3]> cond_begin_0 = const()[name = string("cond_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> cond_end_0 = const()[name = string("cond_end_0"), val = tensor<int32, [3]>([1, 144, 0])];
            tensor<bool, [3]> cond_end_mask_0 = const()[name = string("cond_end_mask_0"), val = tensor<bool, [3]>([false, true, true])];
            tensor<fp16, [1, 144, ?]> cond_cast_fp16 = slice_by_index(begin = cond_begin_0, end = cond_end_0, end_mask = cond_end_mask_0, x = v_cast_fp16)[name = string("cond_cast_fp16")];
            tensor<int32, [3]> uncond_begin_0 = const()[name = string("uncond_begin_0"), val = tensor<int32, [3]>([1, 0, 0])];
            tensor<int32, [3]> uncond_end_0 = const()[name = string("uncond_end_0"), val = tensor<int32, [3]>([2, 144, 0])];
            tensor<bool, [3]> uncond_end_mask_0 = const()[name = string("uncond_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 144, ?]> uncond_cast_fp16 = slice_by_index(begin = uncond_begin_0, end = uncond_end_0, end_mask = uncond_end_mask_0, x = v_cast_fp16)[name = string("uncond_cast_fp16")];
            fp32 var_2241_epsilon_0 = const()[name = string("op_2241_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1]> var_2241_cast_fp16 = inverse(epsilon = var_2241_epsilon_0, x = total_step_to_fp16)[name = string("op_2241_cast_fp16")];
            tensor<int32, [3]> var_2247 = const()[name = string("op_2247"), val = tensor<int32, [3]>([-1, 1, 1])];
            tensor<fp16, [1, 1, 1]> step_cast_fp16 = reshape(shape = var_2247, x = var_2241_cast_fp16)[name = string("step_cast_fp16")];
            fp16 var_2249_to_fp16 = const()[name = string("op_2249_to_fp16"), val = fp16(0x1p+2)];
            tensor<fp16, [1, 144, ?]> var_2250_cast_fp16 = mul(x = cond_cast_fp16, y = var_2249_to_fp16)[name = string("op_2250_cast_fp16")];
            fp16 var_2251_to_fp16 = const()[name = string("op_2251_to_fp16"), val = fp16(0x1.8p+1)];
            tensor<fp16, [1, 144, ?]> var_2252_cast_fp16 = mul(x = uncond_cast_fp16, y = var_2251_to_fp16)[name = string("op_2252_cast_fp16")];
            tensor<fp16, [1, 144, ?]> var_2254_cast_fp16 = sub(x = var_2250_cast_fp16, y = var_2252_cast_fp16)[name = string("op_2254_cast_fp16")];
            tensor<fp16, [1, 144, ?]> var_2255_cast_fp16 = mul(x = step_cast_fp16, y = var_2254_cast_fp16)[name = string("op_2255_cast_fp16")];
            tensor<fp16, [1, 144, ?]> var_2257_cast_fp16 = add(x = noisy_latent_to_fp16, y = var_2255_cast_fp16)[name = string("op_2257_cast_fp16")];
            tensor<fp16, [1, 144, ?]> var_2258_cast_fp16 = mul(x = var_2257_cast_fp16, y = latent_mask_to_fp16)[name = string("op_2258_cast_fp16")];
            string var_2258_cast_fp16_to_fp32_dtype_0 = const()[name = string("op_2258_cast_fp16_to_fp32_dtype_0"), val = string("fp32")];
            tensor<fp32, [1, 144, ?]> denoised_latent = cast(dtype = var_2258_cast_fp16_to_fp32_dtype_0, x = var_2258_cast_fp16)[name = string("cast_0")];
        } -> (denoised_latent);
}