program(1.0)
[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "4.28.4"}, {"coremlc-version", "1429.0.0"}, {"coremltools-component-torch", "1.13.0"}, {"coremltools-version", "6.1"}})]
{
    func main<ios16>(tensor<fp16, [2, 1024, 1, 77]> encoder_hidden_states, tensor<fp16, [2, 4, 64, 64]> sample, tensor<fp16, [2]> timestep) {
            tensor<int32, []> var_25 = const()[name = tensor<string, []>("op_25"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_42_axes_0 = const()[name = tensor<string, []>("op_42_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1]> var_42_cast = expand_dims(axes = var_42_axes_0, x = timestep);
            tensor<fp16, [1, 160]> var_44_to_fp16 = const()[name = tensor<string, []>("op_44_to_fp16"), val = tensor<fp16, [1, 160]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
            tensor<fp16, [2, 160]> emb_3_cast = mul(x = var_42_cast, y = var_44_to_fp16);
            tensor<fp16, [2, 160]> var_49_cast = sin(x = emb_3_cast);
            tensor<fp16, [2, 160]> var_50_cast = cos(x = emb_3_cast);
            tensor<bool, []> emb_interleave_0 = const()[name = tensor<string, []>("emb_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 320]> emb_cast = concat(axis = var_25, interleave = emb_interleave_0, values = (var_49_cast, var_50_cast));
            tensor<int32, [2]> var_54_begin_0 = const()[name = tensor<string, []>("op_54_begin_0"), val = tensor<int32, [2]>([0, 160])];
            tensor<int32, [2]> var_54_end_0 = const()[name = tensor<string, []>("op_54_end_0"), val = tensor<int32, [2]>([2, 320])];
            tensor<bool, [2]> var_54_end_mask_0 = const()[name = tensor<string, []>("op_54_end_mask_0"), val = tensor<bool, [2]>([true, true])];
            tensor<fp16, [2, 160]> var_54_cast = slice_by_index(begin = var_54_begin_0, end = var_54_end_0, end_mask = var_54_end_mask_0, x = emb_cast);
            tensor<int32, [2]> var_56_begin_0 = const()[name = tensor<string, []>("op_56_begin_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [2]> var_56_end_0 = const()[name = tensor<string, []>("op_56_end_0"), val = tensor<int32, [2]>([2, 160])];
            tensor<bool, [2]> var_56_end_mask_0 = const()[name = tensor<string, []>("op_56_end_mask_0"), val = tensor<bool, [2]>([true, false])];
            tensor<fp16, [2, 160]> var_56_cast = slice_by_index(begin = var_56_begin_0, end = var_56_end_0, end_mask = var_56_end_mask_0, x = emb_cast);
            tensor<bool, []> sample_interleave_0 = const()[name = tensor<string, []>("sample_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 320]> sample_cast = concat(axis = var_25, interleave = sample_interleave_0, values = (var_54_cast, var_56_cast));
            tensor<int32, []> var_59 = const()[name = tensor<string, []>("op_59"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> var_66_axes_0 = const()[name = tensor<string, []>("op_66_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2, 320, 1]> var_66_cast = expand_dims(axes = var_66_axes_0, x = sample_cast);
            tensor<int32, [1]> input_1_axes_0 = const()[name = tensor<string, []>("input_1_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2, 320, 1, 1]> input_1_cast = expand_dims(axes = input_1_axes_0, x = var_66_cast);
            tensor<int32, [2]> var_70 = const()[name = tensor<string, []>("op_70"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_72 = const()[name = tensor<string, []>("op_72"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> input_3_pad_type_0 = const()[name = tensor<string, []>("input_3_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> input_3_pad_0 = const()[name = tensor<string, []>("input_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 320, 1, 1]> time_embedding_linear_1_weight_to_fp16 = const()[name = tensor<string, []>("time_embedding_linear_1_weight_to_fp16"), val = tensor<fp16, [1280, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(448)))];
            tensor<fp16, [1280]> time_embedding_linear_1_bias_to_fp16 = const()[name = tensor<string, []>("time_embedding_linear_1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(819712)))];
            tensor<fp16, [2, 1280, 1, 1]> input_3_cast = conv(bias = time_embedding_linear_1_bias_to_fp16, dilations = var_72, groups = var_59, pad = input_3_pad_0, pad_type = input_3_pad_type_0, strides = var_70, weight = time_embedding_linear_1_weight_to_fp16, x = input_1_cast);
            tensor<fp16, [2, 1280, 1, 1]> input_5_cast = silu(x = input_3_cast);
            tensor<int32, [2]> var_78 = const()[name = tensor<string, []>("op_78"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_80 = const()[name = tensor<string, []>("op_80"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> input_13_pad_type_0 = const()[name = tensor<string, []>("input_13_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> input_13_pad_0 = const()[name = tensor<string, []>("input_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> time_embedding_linear_2_weight_to_fp16 = const()[name = tensor<string, []>("time_embedding_linear_2_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(822336)))];
            tensor<fp16, [1280]> time_embedding_linear_2_bias_to_fp16 = const()[name = tensor<string, []>("time_embedding_linear_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4099200)))];
            tensor<fp16, [2, 1280, 1, 1]> input_13_cast = conv(bias = time_embedding_linear_2_bias_to_fp16, dilations = var_80, groups = var_59, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = var_78, weight = time_embedding_linear_2_weight_to_fp16, x = input_5_cast);
            tensor<int32, []> var_86 = const()[name = tensor<string, []>("op_86"), val = tensor<int32, []>(1)];
            tensor<int32, [2]> var_89 = const()[name = tensor<string, []>("op_89"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_91 = const()[name = tensor<string, []>("op_91"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> input_7_pad_type_0 = const()[name = tensor<string, []>("input_7_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> input_7_pad_0 = const()[name = tensor<string, []>("input_7_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [320, 4, 3, 3]> conv_in_weight_to_fp16 = const()[name = tensor<string, []>("conv_in_weight_to_fp16"), val = tensor<fp16, [320, 4, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4101824)))];
            tensor<fp16, [320]> conv_in_bias_to_fp16 = const()[name = tensor<string, []>("conv_in_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4124928)))];
            tensor<fp16, [2, 320, 64, 64]> input_7_cast = conv(bias = conv_in_bias_to_fp16, dilations = var_91, groups = var_86, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = var_89, weight = conv_in_weight_to_fp16, x = sample);
            tensor<bool, []> var_107 = const()[name = tensor<string, []>("op_107"), val = tensor<bool, []>(true)];
            tensor<int32, []> var_112 = const()[name = tensor<string, []>("op_112"), val = tensor<int32, []>(1)];
            tensor<int32, [5]> reshape_0_shape_0 = const()[name = tensor<string, []>("reshape_0_shape_0"), val = tensor<int32, [5]>([2, 32, 10, 64, 64])];
            tensor<fp16, [2, 32, 10, 64, 64]> reshape_0_cast = reshape(shape = reshape_0_shape_0, x = input_7_cast);
            tensor<int32, [3]> reduce_mean_0_axes_0 = const()[name = tensor<string, []>("reduce_mean_0_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_0_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_0_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_0_cast = reduce_mean(axes = reduce_mean_0_axes_0, keep_dims = reduce_mean_0_keep_dims_0, x = reshape_0_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> sub_0_cast = sub(x = reshape_0_cast, y = reduce_mean_0_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> square_0_cast = square(x = sub_0_cast);
            tensor<int32, [3]> reduce_mean_2_axes_0 = const()[name = tensor<string, []>("reduce_mean_2_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_2_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_2_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_2_cast = reduce_mean(axes = reduce_mean_2_axes_0, keep_dims = reduce_mean_2_keep_dims_0, x = square_0_cast);
            tensor<fp16, []> add_0_y_0_to_fp16 = const()[name = tensor<string, []>("add_0_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_0_cast = add(x = reduce_mean_2_cast, y = add_0_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_0_cast = sqrt(x = add_0_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> real_div_0_cast = real_div(x = sub_0_cast, y = sqrt_0_cast);
            tensor<int32, [4]> reshape_1_shape_0 = const()[name = tensor<string, []>("reshape_1_shape_0"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> reshape_1_cast = reshape(shape = reshape_1_shape_0, x = real_div_0_cast);
            tensor<fp16, [320]> add_1_mean_0_to_fp16 = const()[name = tensor<string, []>("add_1_mean_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4125632)))];
            tensor<fp16, [320]> add_1_variance_0_to_fp16 = const()[name = tensor<string, []>("add_1_variance_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4126336)))];
            tensor<fp16, [320]> add_1_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_1_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4127040)))];
            tensor<fp16, [320]> add_1_beta_0_to_fp16 = const()[name = tensor<string, []>("add_1_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4127744)))];
            tensor<fp16, []> add_1_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_1_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 320, 64, 64]> add_1_cast = batch_norm(beta = add_1_beta_0_to_fp16, epsilon = add_1_epsilon_0_to_fp16, gamma = add_1_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_1_cast);
            tensor<fp16, [2, 320, 64, 64]> input_11_cast = silu(x = add_1_cast);
            tensor<int32, [2]> var_134 = const()[name = tensor<string, []>("op_134"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_136 = const()[name = tensor<string, []>("op_136"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_1_pad_type_0 = const()[name = tensor<string, []>("hidden_states_1_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_1_pad_0 = const()[name = tensor<string, []>("hidden_states_1_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [320, 320, 3, 3]> down_blocks_0_resnets_0_conv1_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_resnets_0_conv1_weight_to_fp16"), val = tensor<fp16, [320, 320, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4128448)))];
            tensor<fp16, [320]> down_blocks_0_resnets_0_conv1_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_resnets_0_conv1_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5971712)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_1_cast = conv(bias = down_blocks_0_resnets_0_conv1_bias_to_fp16, dilations = var_136, groups = var_112, pad = hidden_states_1_pad_0, pad_type = hidden_states_1_pad_type_0, strides = var_134, weight = down_blocks_0_resnets_0_conv1_weight_to_fp16, x = input_11_cast);
            tensor<fp16, [2, 1280, 1, 1]> input_15_cast = silu(x = input_13_cast);
            tensor<int32, [2]> var_142 = const()[name = tensor<string, []>("op_142"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_144 = const()[name = tensor<string, []>("op_144"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_1_pad_type_0 = const()[name = tensor<string, []>("temb_1_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_1_pad_0 = const()[name = tensor<string, []>("temb_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1280, 1, 1]> down_blocks_0_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [320, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5972416)))];
            tensor<fp16, [320]> down_blocks_0_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6791680)))];
            tensor<fp16, [2, 320, 1, 1]> temb_1_cast = conv(bias = down_blocks_0_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_144, groups = var_112, pad = temb_1_pad_0, pad_type = temb_1_pad_type_0, strides = var_142, weight = down_blocks_0_resnets_0_time_emb_proj_weight_to_fp16, x = input_15_cast);
            tensor<fp16, [2, 320, 64, 64]> input_17_cast = add(x = hidden_states_1_cast, y = temb_1_cast);
            tensor<int32, [5]> reshape_4_shape_0 = const()[name = tensor<string, []>("reshape_4_shape_0"), val = tensor<int32, [5]>([2, 32, 10, 64, 64])];
            tensor<fp16, [2, 32, 10, 64, 64]> reshape_4_cast = reshape(shape = reshape_4_shape_0, x = input_17_cast);
            tensor<int32, [3]> reduce_mean_3_axes_0 = const()[name = tensor<string, []>("reduce_mean_3_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_3_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_3_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_3_cast = reduce_mean(axes = reduce_mean_3_axes_0, keep_dims = reduce_mean_3_keep_dims_0, x = reshape_4_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> sub_2_cast = sub(x = reshape_4_cast, y = reduce_mean_3_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> square_1_cast = square(x = sub_2_cast);
            tensor<int32, [3]> reduce_mean_5_axes_0 = const()[name = tensor<string, []>("reduce_mean_5_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_5_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_5_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_5_cast = reduce_mean(axes = reduce_mean_5_axes_0, keep_dims = reduce_mean_5_keep_dims_0, x = square_1_cast);
            tensor<fp16, []> add_2_y_0_to_fp16 = const()[name = tensor<string, []>("add_2_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_2_cast = add(x = reduce_mean_5_cast, y = add_2_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_1_cast = sqrt(x = add_2_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> real_div_1_cast = real_div(x = sub_2_cast, y = sqrt_1_cast);
            tensor<int32, [4]> reshape_5_shape_0 = const()[name = tensor<string, []>("reshape_5_shape_0"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> reshape_5_cast = reshape(shape = reshape_5_shape_0, x = real_div_1_cast);
            tensor<fp16, [320]> add_3_mean_0_to_fp16 = const()[name = tensor<string, []>("add_3_mean_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6792384)))];
            tensor<fp16, [320]> add_3_variance_0_to_fp16 = const()[name = tensor<string, []>("add_3_variance_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6793088)))];
            tensor<fp16, [320]> add_3_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_3_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6793792)))];
            tensor<fp16, [320]> add_3_beta_0_to_fp16 = const()[name = tensor<string, []>("add_3_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6794496)))];
            tensor<fp16, []> add_3_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_3_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 320, 64, 64]> add_3_cast = batch_norm(beta = add_3_beta_0_to_fp16, epsilon = add_3_epsilon_0_to_fp16, gamma = add_3_gamma_0_to_fp16, mean = add_3_mean_0_to_fp16, variance = add_3_variance_0_to_fp16, x = reshape_5_cast);
            tensor<fp16, [2, 320, 64, 64]> input_21_cast = silu(x = add_3_cast);
            tensor<int32, [2]> var_154 = const()[name = tensor<string, []>("op_154"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_156 = const()[name = tensor<string, []>("op_156"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_3_pad_type_0 = const()[name = tensor<string, []>("hidden_states_3_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_3_pad_0 = const()[name = tensor<string, []>("hidden_states_3_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [320, 320, 3, 3]> down_blocks_0_resnets_0_conv2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_resnets_0_conv2_weight_to_fp16"), val = tensor<fp16, [320, 320, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6795200)))];
            tensor<fp16, [320]> down_blocks_0_resnets_0_conv2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_resnets_0_conv2_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8638464)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_3_cast = conv(bias = down_blocks_0_resnets_0_conv2_bias_to_fp16, dilations = var_156, groups = var_112, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = var_154, weight = down_blocks_0_resnets_0_conv2_weight_to_fp16, x = input_21_cast);
            tensor<fp16, [2, 320, 64, 64]> hidden_states_5_cast = add(x = input_7_cast, y = hidden_states_3_cast);
            tensor<int32, [5]> reshape_8_shape_0 = const()[name = tensor<string, []>("reshape_8_shape_0"), val = tensor<int32, [5]>([2, 32, 10, 64, 64])];
            tensor<fp16, [2, 32, 10, 64, 64]> reshape_8_cast = reshape(shape = reshape_8_shape_0, x = hidden_states_5_cast);
            tensor<int32, [3]> reduce_mean_6_axes_0 = const()[name = tensor<string, []>("reduce_mean_6_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_6_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_6_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_6_cast = reduce_mean(axes = reduce_mean_6_axes_0, keep_dims = reduce_mean_6_keep_dims_0, x = reshape_8_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> sub_4_cast = sub(x = reshape_8_cast, y = reduce_mean_6_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> square_2_cast = square(x = sub_4_cast);
            tensor<int32, [3]> reduce_mean_8_axes_0 = const()[name = tensor<string, []>("reduce_mean_8_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_8_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_8_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_8_cast = reduce_mean(axes = reduce_mean_8_axes_0, keep_dims = reduce_mean_8_keep_dims_0, x = square_2_cast);
            tensor<fp16, []> add_4_y_0_to_fp16 = const()[name = tensor<string, []>("add_4_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_4_cast = add(x = reduce_mean_8_cast, y = add_4_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_2_cast = sqrt(x = add_4_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> real_div_2_cast = real_div(x = sub_4_cast, y = sqrt_2_cast);
            tensor<int32, [4]> reshape_9_shape_0 = const()[name = tensor<string, []>("reshape_9_shape_0"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> reshape_9_cast = reshape(shape = reshape_9_shape_0, x = real_div_2_cast);
            tensor<fp16, [320]> add_5_mean_0_to_fp16 = const()[name = tensor<string, []>("add_5_mean_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8639168)))];
            tensor<fp16, [320]> add_5_variance_0_to_fp16 = const()[name = tensor<string, []>("add_5_variance_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8639872)))];
            tensor<fp16, [320]> add_5_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_5_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8640576)))];
            tensor<fp16, [320]> add_5_beta_0_to_fp16 = const()[name = tensor<string, []>("add_5_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8641280)))];
            tensor<fp16, []> add_5_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_5_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 320, 64, 64]> add_5_cast = batch_norm(beta = add_5_beta_0_to_fp16, epsilon = add_5_epsilon_0_to_fp16, gamma = add_5_gamma_0_to_fp16, mean = add_5_mean_0_to_fp16, variance = add_5_variance_0_to_fp16, x = reshape_9_cast);
            tensor<int32, [2]> var_176 = const()[name = tensor<string, []>("op_176"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_178 = const()[name = tensor<string, []>("op_178"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_7_pad_type_0 = const()[name = tensor<string, []>("hidden_states_7_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_7_pad_0 = const()[name = tensor<string, []>("hidden_states_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> down_blocks_0_attentions_0_proj_in_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_0_proj_in_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8641984)))];
            tensor<fp16, [320]> down_blocks_0_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_0_proj_in_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8846848)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_7_cast = conv(bias = down_blocks_0_attentions_0_proj_in_bias_to_fp16, dilations = var_178, groups = var_112, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = var_176, weight = down_blocks_0_attentions_0_proj_in_weight_to_fp16, x = add_5_cast);
            tensor<int32, [4]> var_183 = const()[name = tensor<string, []>("op_183"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<fp16, [2, 320, 1, 4096]> inputs_1_cast = reshape(shape = var_183, x = hidden_states_7_cast);
            tensor<int32, [1]> var_193 = const()[name = tensor<string, []>("op_193"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_1_cast = reduce_mean(axes = var_193, keep_dims = var_107, x = inputs_1_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_1_cast = sub(x = inputs_1_cast, y = channels_mean_1_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_sq_1_cast = mul(x = zero_mean_1_cast, y = zero_mean_1_cast);
            tensor<int32, [1]> var_197 = const()[name = tensor<string, []>("op_197"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_198_cast = reduce_mean(axes = var_197, keep_dims = var_107, x = zero_mean_sq_1_cast);
            tensor<fp16, []> var_199_to_fp16 = const()[name = tensor<string, []>("op_199_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_200_cast = add(x = var_198_cast, y = var_199_to_fp16);
            tensor<fp16, []> denom_1_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_1_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_1_cast = rsqrt(epsilon = denom_1_epsilon_0_to_fp16, x = var_200_cast);
            tensor<fp16, [2, 320, 1, 4096]> out_1_cast = mul(x = zero_mean_1_cast, y = denom_1_cast);
            tensor<fp16, [1, 320, 1, 1]> var_204_to_fp16 = const()[name = tensor<string, []>("op_204_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8847552)))];
            tensor<fp16, [2, 320, 1, 4096]> var_205_cast = add(x = out_1_cast, y = var_204_to_fp16);
            tensor<fp16, [1, 320, 1, 1]> var_207_to_fp16 = const()[name = tensor<string, []>("op_207_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8848256)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_9_cast = mul(x = var_205_cast, y = var_207_to_fp16);
            tensor<int32, [2]> var_214 = const()[name = tensor<string, []>("op_214"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_216 = const()[name = tensor<string, []>("op_216"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_1_pad_type_0 = const()[name = tensor<string, []>("q_1_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_1_pad_0 = const()[name = tensor<string, []>("q_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8848960)))];
            tensor<fp16, [2, 320, 1, 4096]> q_1_cast = conv(dilations = var_216, groups = var_112, pad = q_1_pad_0, pad_type = q_1_pad_type_0, strides = var_214, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_9_cast);
            tensor<int32, [2]> var_220 = const()[name = tensor<string, []>("op_220"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_222 = const()[name = tensor<string, []>("op_222"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_1_pad_type_0 = const()[name = tensor<string, []>("k_1_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_1_pad_0 = const()[name = tensor<string, []>("k_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9053824)))];
            tensor<fp16, [2, 320, 1, 4096]> k_1_cast = conv(dilations = var_222, groups = var_112, pad = k_1_pad_0, pad_type = k_1_pad_type_0, strides = var_220, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_9_cast);
            tensor<int32, [2]> var_226 = const()[name = tensor<string, []>("op_226"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_228 = const()[name = tensor<string, []>("op_228"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_1_pad_type_0 = const()[name = tensor<string, []>("v_1_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_1_pad_0 = const()[name = tensor<string, []>("v_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9258688)))];
            tensor<fp16, [2, 320, 1, 4096]> v_1_cast = conv(dilations = var_228, groups = var_112, pad = v_1_pad_0, pad_type = v_1_pad_type_0, strides = var_226, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_9_cast);
            tensor<int32, [4]> var_232_begin_0 = const()[name = tensor<string, []>("op_232_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_232_end_0 = const()[name = tensor<string, []>("op_232_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_232_end_mask_0 = const()[name = tensor<string, []>("op_232_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_232_cast = slice_by_index(begin = var_232_begin_0, end = var_232_end_0, end_mask = var_232_end_mask_0, x = q_1_cast);
            tensor<int32, [4]> var_236_begin_0 = const()[name = tensor<string, []>("op_236_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_236_end_0 = const()[name = tensor<string, []>("op_236_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_236_end_mask_0 = const()[name = tensor<string, []>("op_236_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_236_cast = slice_by_index(begin = var_236_begin_0, end = var_236_end_0, end_mask = var_236_end_mask_0, x = q_1_cast);
            tensor<int32, [4]> var_240_begin_0 = const()[name = tensor<string, []>("op_240_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_240_end_0 = const()[name = tensor<string, []>("op_240_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_240_end_mask_0 = const()[name = tensor<string, []>("op_240_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_240_cast = slice_by_index(begin = var_240_begin_0, end = var_240_end_0, end_mask = var_240_end_mask_0, x = q_1_cast);
            tensor<int32, [4]> var_244_begin_0 = const()[name = tensor<string, []>("op_244_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_244_end_0 = const()[name = tensor<string, []>("op_244_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_244_end_mask_0 = const()[name = tensor<string, []>("op_244_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_244_cast = slice_by_index(begin = var_244_begin_0, end = var_244_end_0, end_mask = var_244_end_mask_0, x = q_1_cast);
            tensor<int32, [4]> var_248_begin_0 = const()[name = tensor<string, []>("op_248_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_248_end_0 = const()[name = tensor<string, []>("op_248_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_248_end_mask_0 = const()[name = tensor<string, []>("op_248_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_248_cast = slice_by_index(begin = var_248_begin_0, end = var_248_end_0, end_mask = var_248_end_mask_0, x = q_1_cast);
            tensor<int32, [4]> k_3_perm_0 = const()[name = tensor<string, []>("k_3_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_255_begin_0 = const()[name = tensor<string, []>("op_255_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_255_end_0 = const()[name = tensor<string, []>("op_255_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 64])];
            tensor<bool, [4]> var_255_end_mask_0 = const()[name = tensor<string, []>("op_255_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 320]> transpose_31 = transpose(perm = k_3_perm_0, x = k_1_cast);
            tensor<fp16, [2, 4096, 1, 64]> var_255_cast = slice_by_index(begin = var_255_begin_0, end = var_255_end_0, end_mask = var_255_end_mask_0, x = transpose_31);
            tensor<int32, [4]> var_259_begin_0 = const()[name = tensor<string, []>("op_259_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_259_end_0 = const()[name = tensor<string, []>("op_259_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 128])];
            tensor<bool, [4]> var_259_end_mask_0 = const()[name = tensor<string, []>("op_259_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_259_cast = slice_by_index(begin = var_259_begin_0, end = var_259_end_0, end_mask = var_259_end_mask_0, x = transpose_31);
            tensor<int32, [4]> var_263_begin_0 = const()[name = tensor<string, []>("op_263_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_263_end_0 = const()[name = tensor<string, []>("op_263_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 192])];
            tensor<bool, [4]> var_263_end_mask_0 = const()[name = tensor<string, []>("op_263_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_263_cast = slice_by_index(begin = var_263_begin_0, end = var_263_end_0, end_mask = var_263_end_mask_0, x = transpose_31);
            tensor<int32, [4]> var_267_begin_0 = const()[name = tensor<string, []>("op_267_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_267_end_0 = const()[name = tensor<string, []>("op_267_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 256])];
            tensor<bool, [4]> var_267_end_mask_0 = const()[name = tensor<string, []>("op_267_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_267_cast = slice_by_index(begin = var_267_begin_0, end = var_267_end_0, end_mask = var_267_end_mask_0, x = transpose_31);
            tensor<int32, [4]> var_271_begin_0 = const()[name = tensor<string, []>("op_271_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_271_end_0 = const()[name = tensor<string, []>("op_271_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_271_end_mask_0 = const()[name = tensor<string, []>("op_271_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_271_cast = slice_by_index(begin = var_271_begin_0, end = var_271_end_0, end_mask = var_271_end_mask_0, x = transpose_31);
            tensor<int32, [4]> var_273_begin_0 = const()[name = tensor<string, []>("op_273_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_273_end_0 = const()[name = tensor<string, []>("op_273_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_273_end_mask_0 = const()[name = tensor<string, []>("op_273_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_273_cast = slice_by_index(begin = var_273_begin_0, end = var_273_end_0, end_mask = var_273_end_mask_0, x = v_1_cast);
            tensor<int32, [4]> var_277_begin_0 = const()[name = tensor<string, []>("op_277_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_277_end_0 = const()[name = tensor<string, []>("op_277_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_277_end_mask_0 = const()[name = tensor<string, []>("op_277_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_277_cast = slice_by_index(begin = var_277_begin_0, end = var_277_end_0, end_mask = var_277_end_mask_0, x = v_1_cast);
            tensor<int32, [4]> var_281_begin_0 = const()[name = tensor<string, []>("op_281_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_281_end_0 = const()[name = tensor<string, []>("op_281_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_281_end_mask_0 = const()[name = tensor<string, []>("op_281_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_281_cast = slice_by_index(begin = var_281_begin_0, end = var_281_end_0, end_mask = var_281_end_mask_0, x = v_1_cast);
            tensor<int32, [4]> var_285_begin_0 = const()[name = tensor<string, []>("op_285_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_285_end_0 = const()[name = tensor<string, []>("op_285_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_285_end_mask_0 = const()[name = tensor<string, []>("op_285_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_285_cast = slice_by_index(begin = var_285_begin_0, end = var_285_end_0, end_mask = var_285_end_mask_0, x = v_1_cast);
            tensor<int32, [4]> var_289_begin_0 = const()[name = tensor<string, []>("op_289_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_289_end_0 = const()[name = tensor<string, []>("op_289_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_289_end_mask_0 = const()[name = tensor<string, []>("op_289_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_289_cast = slice_by_index(begin = var_289_begin_0, end = var_289_end_0, end_mask = var_289_end_mask_0, x = v_1_cast);
            tensor<string, []> var_293_equation_0 = const()[name = tensor<string, []>("op_293_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_293_cast = einsum(equation = var_293_equation_0, values = (var_255_cast, var_232_cast));
            tensor<fp16, []> var_294_to_fp16 = const()[name = tensor<string, []>("op_294_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_1_cast = mul(x = var_293_cast, y = var_294_to_fp16);
            tensor<string, []> var_297_equation_0 = const()[name = tensor<string, []>("op_297_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_297_cast = einsum(equation = var_297_equation_0, values = (var_259_cast, var_236_cast));
            tensor<fp16, []> var_298_to_fp16 = const()[name = tensor<string, []>("op_298_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_3_cast = mul(x = var_297_cast, y = var_298_to_fp16);
            tensor<string, []> var_301_equation_0 = const()[name = tensor<string, []>("op_301_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_301_cast = einsum(equation = var_301_equation_0, values = (var_263_cast, var_240_cast));
            tensor<fp16, []> var_302_to_fp16 = const()[name = tensor<string, []>("op_302_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_5_cast = mul(x = var_301_cast, y = var_302_to_fp16);
            tensor<string, []> var_305_equation_0 = const()[name = tensor<string, []>("op_305_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_305_cast = einsum(equation = var_305_equation_0, values = (var_267_cast, var_244_cast));
            tensor<fp16, []> var_306_to_fp16 = const()[name = tensor<string, []>("op_306_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_7_cast = mul(x = var_305_cast, y = var_306_to_fp16);
            tensor<string, []> var_309_equation_0 = const()[name = tensor<string, []>("op_309_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_309_cast = einsum(equation = var_309_equation_0, values = (var_271_cast, var_248_cast));
            tensor<fp16, []> var_310_to_fp16 = const()[name = tensor<string, []>("op_310_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_9_cast = mul(x = var_309_cast, y = var_310_to_fp16);
            tensor<fp16, [2, 4096, 1, 4096]> var_312_cast = softmax(axis = var_112, x = aw_1_cast);
            tensor<fp16, [2, 4096, 1, 4096]> var_313_cast = softmax(axis = var_112, x = aw_3_cast);
            tensor<fp16, [2, 4096, 1, 4096]> var_314_cast = softmax(axis = var_112, x = aw_5_cast);
            tensor<fp16, [2, 4096, 1, 4096]> var_315_cast = softmax(axis = var_112, x = aw_7_cast);
            tensor<fp16, [2, 4096, 1, 4096]> var_316_cast = softmax(axis = var_112, x = aw_9_cast);
            tensor<string, []> var_318_equation_0 = const()[name = tensor<string, []>("op_318_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_318_cast = einsum(equation = var_318_equation_0, values = (var_273_cast, var_312_cast));
            tensor<string, []> var_320_equation_0 = const()[name = tensor<string, []>("op_320_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_320_cast = einsum(equation = var_320_equation_0, values = (var_277_cast, var_313_cast));
            tensor<string, []> var_322_equation_0 = const()[name = tensor<string, []>("op_322_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_322_cast = einsum(equation = var_322_equation_0, values = (var_281_cast, var_314_cast));
            tensor<string, []> var_324_equation_0 = const()[name = tensor<string, []>("op_324_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_324_cast = einsum(equation = var_324_equation_0, values = (var_285_cast, var_315_cast));
            tensor<string, []> var_326_equation_0 = const()[name = tensor<string, []>("op_326_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_326_cast = einsum(equation = var_326_equation_0, values = (var_289_cast, var_316_cast));
            tensor<bool, []> input_25_interleave_0 = const()[name = tensor<string, []>("input_25_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 320, 1, 4096]> input_25_cast = concat(axis = var_112, interleave = input_25_interleave_0, values = (var_318_cast, var_320_cast, var_322_cast, var_324_cast, var_326_cast));
            tensor<int32, [2]> var_332 = const()[name = tensor<string, []>("op_332"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_334 = const()[name = tensor<string, []>("op_334"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_336_pad_type_0 = const()[name = tensor<string, []>("op_336_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_336_pad_0 = const()[name = tensor<string, []>("op_336_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9463552)))];
            tensor<fp16, [320]> down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9668416)))];
            tensor<fp16, [2, 320, 1, 4096]> var_336_cast = conv(bias = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_334, groups = var_112, pad = var_336_pad_0, pad_type = var_336_pad_type_0, strides = var_332, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_25_cast);
            tensor<fp16, [2, 320, 1, 4096]> inputs_3_cast = add(x = var_336_cast, y = inputs_1_cast);
            tensor<int32, [1]> var_340 = const()[name = tensor<string, []>("op_340"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_3_cast = reduce_mean(axes = var_340, keep_dims = var_107, x = inputs_3_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_3_cast = sub(x = inputs_3_cast, y = channels_mean_3_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_sq_3_cast = mul(x = zero_mean_3_cast, y = zero_mean_3_cast);
            tensor<int32, [1]> var_344 = const()[name = tensor<string, []>("op_344"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_345_cast = reduce_mean(axes = var_344, keep_dims = var_107, x = zero_mean_sq_3_cast);
            tensor<fp16, []> var_346_to_fp16 = const()[name = tensor<string, []>("op_346_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_347_cast = add(x = var_345_cast, y = var_346_to_fp16);
            tensor<fp16, []> denom_3_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_3_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_3_cast = rsqrt(epsilon = denom_3_epsilon_0_to_fp16, x = var_347_cast);
            tensor<fp16, [2, 320, 1, 4096]> out_3_cast = mul(x = zero_mean_3_cast, y = denom_3_cast);
            tensor<fp16, [1, 320, 1, 1]> var_351_to_fp16 = const()[name = tensor<string, []>("op_351_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9669120)))];
            tensor<fp16, [2, 320, 1, 4096]> var_352_cast = add(x = out_3_cast, y = var_351_to_fp16);
            tensor<fp16, [1, 320, 1, 1]> var_354_to_fp16 = const()[name = tensor<string, []>("op_354_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9669824)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_11_cast = mul(x = var_352_cast, y = var_354_to_fp16);
            tensor<int32, [2]> var_361 = const()[name = tensor<string, []>("op_361"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_363 = const()[name = tensor<string, []>("op_363"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_3_pad_type_0 = const()[name = tensor<string, []>("q_3_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_3_pad_0 = const()[name = tensor<string, []>("q_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9670528)))];
            tensor<fp16, [2, 320, 1, 4096]> q_3_cast = conv(dilations = var_363, groups = var_112, pad = q_3_pad_0, pad_type = q_3_pad_type_0, strides = var_361, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_11_cast);
            tensor<int32, [2]> var_367 = const()[name = tensor<string, []>("op_367"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_369 = const()[name = tensor<string, []>("op_369"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_5_pad_type_0 = const()[name = tensor<string, []>("k_5_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_5_pad_0 = const()[name = tensor<string, []>("k_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1024, 1, 1]> down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [320, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9875392)))];
            tensor<fp16, [2, 320, 1, 77]> k_5_cast = conv(dilations = var_369, groups = var_112, pad = k_5_pad_0, pad_type = k_5_pad_type_0, strides = var_367, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [2]> var_373 = const()[name = tensor<string, []>("op_373"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_375 = const()[name = tensor<string, []>("op_375"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_3_pad_type_0 = const()[name = tensor<string, []>("v_3_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_3_pad_0 = const()[name = tensor<string, []>("v_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1024, 1, 1]> down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [320, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10530816)))];
            tensor<fp16, [2, 320, 1, 77]> v_3_cast = conv(dilations = var_375, groups = var_112, pad = v_3_pad_0, pad_type = v_3_pad_type_0, strides = var_373, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [4]> var_379_begin_0 = const()[name = tensor<string, []>("op_379_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_379_end_0 = const()[name = tensor<string, []>("op_379_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_379_end_mask_0 = const()[name = tensor<string, []>("op_379_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_379_cast = slice_by_index(begin = var_379_begin_0, end = var_379_end_0, end_mask = var_379_end_mask_0, x = q_3_cast);
            tensor<int32, [4]> var_383_begin_0 = const()[name = tensor<string, []>("op_383_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_383_end_0 = const()[name = tensor<string, []>("op_383_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_383_end_mask_0 = const()[name = tensor<string, []>("op_383_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_383_cast = slice_by_index(begin = var_383_begin_0, end = var_383_end_0, end_mask = var_383_end_mask_0, x = q_3_cast);
            tensor<int32, [4]> var_387_begin_0 = const()[name = tensor<string, []>("op_387_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_387_end_0 = const()[name = tensor<string, []>("op_387_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_387_end_mask_0 = const()[name = tensor<string, []>("op_387_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_387_cast = slice_by_index(begin = var_387_begin_0, end = var_387_end_0, end_mask = var_387_end_mask_0, x = q_3_cast);
            tensor<int32, [4]> var_391_begin_0 = const()[name = tensor<string, []>("op_391_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_391_end_0 = const()[name = tensor<string, []>("op_391_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_391_end_mask_0 = const()[name = tensor<string, []>("op_391_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_391_cast = slice_by_index(begin = var_391_begin_0, end = var_391_end_0, end_mask = var_391_end_mask_0, x = q_3_cast);
            tensor<int32, [4]> var_395_begin_0 = const()[name = tensor<string, []>("op_395_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_395_end_0 = const()[name = tensor<string, []>("op_395_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_395_end_mask_0 = const()[name = tensor<string, []>("op_395_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_395_cast = slice_by_index(begin = var_395_begin_0, end = var_395_end_0, end_mask = var_395_end_mask_0, x = q_3_cast);
            tensor<int32, [4]> k_7_perm_0 = const()[name = tensor<string, []>("k_7_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_402_begin_0 = const()[name = tensor<string, []>("op_402_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_402_end_0 = const()[name = tensor<string, []>("op_402_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_402_end_mask_0 = const()[name = tensor<string, []>("op_402_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 320]> transpose_30 = transpose(perm = k_7_perm_0, x = k_5_cast);
            tensor<fp16, [2, 77, 1, 64]> var_402_cast = slice_by_index(begin = var_402_begin_0, end = var_402_end_0, end_mask = var_402_end_mask_0, x = transpose_30);
            tensor<int32, [4]> var_406_begin_0 = const()[name = tensor<string, []>("op_406_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_406_end_0 = const()[name = tensor<string, []>("op_406_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_406_end_mask_0 = const()[name = tensor<string, []>("op_406_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_406_cast = slice_by_index(begin = var_406_begin_0, end = var_406_end_0, end_mask = var_406_end_mask_0, x = transpose_30);
            tensor<int32, [4]> var_410_begin_0 = const()[name = tensor<string, []>("op_410_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_410_end_0 = const()[name = tensor<string, []>("op_410_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_410_end_mask_0 = const()[name = tensor<string, []>("op_410_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_410_cast = slice_by_index(begin = var_410_begin_0, end = var_410_end_0, end_mask = var_410_end_mask_0, x = transpose_30);
            tensor<int32, [4]> var_414_begin_0 = const()[name = tensor<string, []>("op_414_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_414_end_0 = const()[name = tensor<string, []>("op_414_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_414_end_mask_0 = const()[name = tensor<string, []>("op_414_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_414_cast = slice_by_index(begin = var_414_begin_0, end = var_414_end_0, end_mask = var_414_end_mask_0, x = transpose_30);
            tensor<int32, [4]> var_418_begin_0 = const()[name = tensor<string, []>("op_418_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_418_end_0 = const()[name = tensor<string, []>("op_418_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_418_end_mask_0 = const()[name = tensor<string, []>("op_418_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_418_cast = slice_by_index(begin = var_418_begin_0, end = var_418_end_0, end_mask = var_418_end_mask_0, x = transpose_30);
            tensor<int32, [4]> var_420_begin_0 = const()[name = tensor<string, []>("op_420_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_420_end_0 = const()[name = tensor<string, []>("op_420_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_420_end_mask_0 = const()[name = tensor<string, []>("op_420_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_420_cast = slice_by_index(begin = var_420_begin_0, end = var_420_end_0, end_mask = var_420_end_mask_0, x = v_3_cast);
            tensor<int32, [4]> var_424_begin_0 = const()[name = tensor<string, []>("op_424_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_424_end_0 = const()[name = tensor<string, []>("op_424_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_424_end_mask_0 = const()[name = tensor<string, []>("op_424_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_424_cast = slice_by_index(begin = var_424_begin_0, end = var_424_end_0, end_mask = var_424_end_mask_0, x = v_3_cast);
            tensor<int32, [4]> var_428_begin_0 = const()[name = tensor<string, []>("op_428_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_428_end_0 = const()[name = tensor<string, []>("op_428_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_428_end_mask_0 = const()[name = tensor<string, []>("op_428_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_428_cast = slice_by_index(begin = var_428_begin_0, end = var_428_end_0, end_mask = var_428_end_mask_0, x = v_3_cast);
            tensor<int32, [4]> var_432_begin_0 = const()[name = tensor<string, []>("op_432_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_432_end_0 = const()[name = tensor<string, []>("op_432_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_432_end_mask_0 = const()[name = tensor<string, []>("op_432_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_432_cast = slice_by_index(begin = var_432_begin_0, end = var_432_end_0, end_mask = var_432_end_mask_0, x = v_3_cast);
            tensor<int32, [4]> var_436_begin_0 = const()[name = tensor<string, []>("op_436_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_436_end_0 = const()[name = tensor<string, []>("op_436_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_436_end_mask_0 = const()[name = tensor<string, []>("op_436_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_436_cast = slice_by_index(begin = var_436_begin_0, end = var_436_end_0, end_mask = var_436_end_mask_0, x = v_3_cast);
            tensor<string, []> var_440_equation_0 = const()[name = tensor<string, []>("op_440_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_440_cast = einsum(equation = var_440_equation_0, values = (var_402_cast, var_379_cast));
            tensor<fp16, []> var_441_to_fp16 = const()[name = tensor<string, []>("op_441_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_11_cast = mul(x = var_440_cast, y = var_441_to_fp16);
            tensor<string, []> var_444_equation_0 = const()[name = tensor<string, []>("op_444_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_444_cast = einsum(equation = var_444_equation_0, values = (var_406_cast, var_383_cast));
            tensor<fp16, []> var_445_to_fp16 = const()[name = tensor<string, []>("op_445_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_13_cast = mul(x = var_444_cast, y = var_445_to_fp16);
            tensor<string, []> var_448_equation_0 = const()[name = tensor<string, []>("op_448_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_448_cast = einsum(equation = var_448_equation_0, values = (var_410_cast, var_387_cast));
            tensor<fp16, []> var_449_to_fp16 = const()[name = tensor<string, []>("op_449_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_15_cast = mul(x = var_448_cast, y = var_449_to_fp16);
            tensor<string, []> var_452_equation_0 = const()[name = tensor<string, []>("op_452_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_452_cast = einsum(equation = var_452_equation_0, values = (var_414_cast, var_391_cast));
            tensor<fp16, []> var_453_to_fp16 = const()[name = tensor<string, []>("op_453_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_17_cast = mul(x = var_452_cast, y = var_453_to_fp16);
            tensor<string, []> var_456_equation_0 = const()[name = tensor<string, []>("op_456_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_456_cast = einsum(equation = var_456_equation_0, values = (var_418_cast, var_395_cast));
            tensor<fp16, []> var_457_to_fp16 = const()[name = tensor<string, []>("op_457_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_19_cast = mul(x = var_456_cast, y = var_457_to_fp16);
            tensor<fp16, [2, 77, 1, 4096]> var_459_cast = softmax(axis = var_112, x = aw_11_cast);
            tensor<fp16, [2, 77, 1, 4096]> var_460_cast = softmax(axis = var_112, x = aw_13_cast);
            tensor<fp16, [2, 77, 1, 4096]> var_461_cast = softmax(axis = var_112, x = aw_15_cast);
            tensor<fp16, [2, 77, 1, 4096]> var_462_cast = softmax(axis = var_112, x = aw_17_cast);
            tensor<fp16, [2, 77, 1, 4096]> var_463_cast = softmax(axis = var_112, x = aw_19_cast);
            tensor<string, []> var_465_equation_0 = const()[name = tensor<string, []>("op_465_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_465_cast = einsum(equation = var_465_equation_0, values = (var_420_cast, var_459_cast));
            tensor<string, []> var_467_equation_0 = const()[name = tensor<string, []>("op_467_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_467_cast = einsum(equation = var_467_equation_0, values = (var_424_cast, var_460_cast));
            tensor<string, []> var_469_equation_0 = const()[name = tensor<string, []>("op_469_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_469_cast = einsum(equation = var_469_equation_0, values = (var_428_cast, var_461_cast));
            tensor<string, []> var_471_equation_0 = const()[name = tensor<string, []>("op_471_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_471_cast = einsum(equation = var_471_equation_0, values = (var_432_cast, var_462_cast));
            tensor<string, []> var_473_equation_0 = const()[name = tensor<string, []>("op_473_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_473_cast = einsum(equation = var_473_equation_0, values = (var_436_cast, var_463_cast));
            tensor<bool, []> input_27_interleave_0 = const()[name = tensor<string, []>("input_27_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 320, 1, 4096]> input_27_cast = concat(axis = var_112, interleave = input_27_interleave_0, values = (var_465_cast, var_467_cast, var_469_cast, var_471_cast, var_473_cast));
            tensor<int32, [2]> var_479 = const()[name = tensor<string, []>("op_479"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_481 = const()[name = tensor<string, []>("op_481"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_483_pad_type_0 = const()[name = tensor<string, []>("op_483_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_483_pad_0 = const()[name = tensor<string, []>("op_483_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11186240)))];
            tensor<fp16, [320]> down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11391104)))];
            tensor<fp16, [2, 320, 1, 4096]> var_483_cast = conv(bias = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_481, groups = var_112, pad = var_483_pad_0, pad_type = var_483_pad_type_0, strides = var_479, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_27_cast);
            tensor<fp16, [2, 320, 1, 4096]> inputs_5_cast = add(x = var_483_cast, y = inputs_3_cast);
            tensor<int32, [1]> var_487 = const()[name = tensor<string, []>("op_487"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_5_cast = reduce_mean(axes = var_487, keep_dims = var_107, x = inputs_5_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_5_cast = sub(x = inputs_5_cast, y = channels_mean_5_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_sq_5_cast = mul(x = zero_mean_5_cast, y = zero_mean_5_cast);
            tensor<int32, [1]> var_491 = const()[name = tensor<string, []>("op_491"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_492_cast = reduce_mean(axes = var_491, keep_dims = var_107, x = zero_mean_sq_5_cast);
            tensor<fp16, []> var_493_to_fp16 = const()[name = tensor<string, []>("op_493_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_494_cast = add(x = var_492_cast, y = var_493_to_fp16);
            tensor<fp16, []> denom_5_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_5_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_5_cast = rsqrt(epsilon = denom_5_epsilon_0_to_fp16, x = var_494_cast);
            tensor<fp16, [2, 320, 1, 4096]> out_5_cast = mul(x = zero_mean_5_cast, y = denom_5_cast);
            tensor<fp16, [1, 320, 1, 1]> var_498_to_fp16 = const()[name = tensor<string, []>("op_498_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11391808)))];
            tensor<fp16, [2, 320, 1, 4096]> var_499_cast = add(x = out_5_cast, y = var_498_to_fp16);
            tensor<fp16, [1, 320, 1, 1]> var_501_to_fp16 = const()[name = tensor<string, []>("op_501_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11392512)))];
            tensor<fp16, [2, 320, 1, 4096]> input_29_cast = mul(x = var_499_cast, y = var_501_to_fp16);
            tensor<int32, [2]> var_509 = const()[name = tensor<string, []>("op_509"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_511 = const()[name = tensor<string, []>("op_511"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_513_pad_type_0 = const()[name = tensor<string, []>("op_513_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_513_pad_0 = const()[name = tensor<string, []>("op_513_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [2560, 320, 1, 1]> down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [2560, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11393216)))];
            tensor<fp16, [2560]> down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13031680)))];
            tensor<fp16, [2, 2560, 1, 4096]> var_513_cast = conv(bias = down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_511, groups = var_112, pad = var_513_pad_0, pad_type = var_513_pad_type_0, strides = var_509, weight = down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_29_cast);
            tensor<int32, [2]> var_514_split_sizes_0 = const()[name = tensor<string, []>("op_514_split_sizes_0"), val = tensor<int32, [2]>([1280, 1280])];
            tensor<int32, []> var_514_axis_0 = const()[name = tensor<string, []>("op_514_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 1280, 1, 4096]> var_514_cast_0, tensor<fp16, [2, 1280, 1, 4096]> var_514_cast_1 = split(axis = var_514_axis_0, split_sizes = var_514_split_sizes_0, x = var_513_cast);
            tensor<string, []> var_516_mode_0 = const()[name = tensor<string, []>("op_516_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 1280, 1, 4096]> var_516_cast = gelu(mode = var_516_mode_0, x = var_514_cast_1);
            tensor<fp16, [2, 1280, 1, 4096]> input_31_cast = mul(x = var_514_cast_0, y = var_516_cast);
            tensor<int32, [2]> var_520 = const()[name = tensor<string, []>("op_520"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_522 = const()[name = tensor<string, []>("op_522"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_524_pad_type_0 = const()[name = tensor<string, []>("op_524_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_524_pad_0 = const()[name = tensor<string, []>("op_524_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1280, 1, 1]> down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [320, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13036864)))];
            tensor<fp16, [320]> down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13856128)))];
            tensor<fp16, [2, 320, 1, 4096]> var_524_cast = conv(bias = down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_522, groups = var_112, pad = var_524_pad_0, pad_type = var_524_pad_type_0, strides = var_520, weight = down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_31_cast);
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_15_cast = add(x = var_524_cast, y = inputs_5_cast);
            tensor<int32, [4]> var_526 = const()[name = tensor<string, []>("op_526"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> input_33_cast = reshape(shape = var_526, x = hidden_states_15_cast);
            tensor<int32, [2]> var_530 = const()[name = tensor<string, []>("op_530"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_532 = const()[name = tensor<string, []>("op_532"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_17_pad_type_0 = const()[name = tensor<string, []>("hidden_states_17_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_17_pad_0 = const()[name = tensor<string, []>("hidden_states_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> down_blocks_0_attentions_0_proj_out_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_0_proj_out_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13856832)))];
            tensor<fp16, [320]> down_blocks_0_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_0_proj_out_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14061696)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_17_cast = conv(bias = down_blocks_0_attentions_0_proj_out_bias_to_fp16, dilations = var_532, groups = var_112, pad = hidden_states_17_pad_0, pad_type = hidden_states_17_pad_type_0, strides = var_530, weight = down_blocks_0_attentions_0_proj_out_weight_to_fp16, x = input_33_cast);
            tensor<fp16, [2, 320, 64, 64]> input_35_cast = add(x = hidden_states_17_cast, y = hidden_states_5_cast);
            tensor<int32, [5]> reshape_12_shape_0 = const()[name = tensor<string, []>("reshape_12_shape_0"), val = tensor<int32, [5]>([2, 32, 10, 64, 64])];
            tensor<fp16, [2, 32, 10, 64, 64]> reshape_12_cast = reshape(shape = reshape_12_shape_0, x = input_35_cast);
            tensor<int32, [3]> reduce_mean_9_axes_0 = const()[name = tensor<string, []>("reduce_mean_9_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_9_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_9_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_9_cast = reduce_mean(axes = reduce_mean_9_axes_0, keep_dims = reduce_mean_9_keep_dims_0, x = reshape_12_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> sub_6_cast = sub(x = reshape_12_cast, y = reduce_mean_9_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> square_3_cast = square(x = sub_6_cast);
            tensor<int32, [3]> reduce_mean_11_axes_0 = const()[name = tensor<string, []>("reduce_mean_11_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_11_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_11_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_11_cast = reduce_mean(axes = reduce_mean_11_axes_0, keep_dims = reduce_mean_11_keep_dims_0, x = square_3_cast);
            tensor<fp16, []> add_6_y_0_to_fp16 = const()[name = tensor<string, []>("add_6_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_6_cast = add(x = reduce_mean_11_cast, y = add_6_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_3_cast = sqrt(x = add_6_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> real_div_3_cast = real_div(x = sub_6_cast, y = sqrt_3_cast);
            tensor<int32, [4]> reshape_13_shape_0 = const()[name = tensor<string, []>("reshape_13_shape_0"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> reshape_13_cast = reshape(shape = reshape_13_shape_0, x = real_div_3_cast);
            tensor<fp16, [320]> add_7_mean_0_to_fp16 = const()[name = tensor<string, []>("add_7_mean_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14062400)))];
            tensor<fp16, [320]> add_7_variance_0_to_fp16 = const()[name = tensor<string, []>("add_7_variance_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14063104)))];
            tensor<fp16, [320]> add_7_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_7_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14063808)))];
            tensor<fp16, [320]> add_7_beta_0_to_fp16 = const()[name = tensor<string, []>("add_7_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14064512)))];
            tensor<fp16, []> add_7_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_7_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 320, 64, 64]> add_7_cast = batch_norm(beta = add_7_beta_0_to_fp16, epsilon = add_7_epsilon_0_to_fp16, gamma = add_7_gamma_0_to_fp16, mean = add_7_mean_0_to_fp16, variance = add_7_variance_0_to_fp16, x = reshape_13_cast);
            tensor<fp16, [2, 320, 64, 64]> input_39_cast = silu(x = add_7_cast);
            tensor<int32, [2]> var_547 = const()[name = tensor<string, []>("op_547"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_549 = const()[name = tensor<string, []>("op_549"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_19_pad_type_0 = const()[name = tensor<string, []>("hidden_states_19_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_19_pad_0 = const()[name = tensor<string, []>("hidden_states_19_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [320, 320, 3, 3]> down_blocks_0_resnets_1_conv1_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_resnets_1_conv1_weight_to_fp16"), val = tensor<fp16, [320, 320, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14065216)))];
            tensor<fp16, [320]> down_blocks_0_resnets_1_conv1_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_resnets_1_conv1_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15908480)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_19_cast = conv(bias = down_blocks_0_resnets_1_conv1_bias_to_fp16, dilations = var_549, groups = var_112, pad = hidden_states_19_pad_0, pad_type = hidden_states_19_pad_type_0, strides = var_547, weight = down_blocks_0_resnets_1_conv1_weight_to_fp16, x = input_39_cast);
            tensor<int32, [2]> var_555 = const()[name = tensor<string, []>("op_555"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_557 = const()[name = tensor<string, []>("op_557"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_3_pad_type_0 = const()[name = tensor<string, []>("temb_3_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_3_pad_0 = const()[name = tensor<string, []>("temb_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1280, 1, 1]> down_blocks_0_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [320, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15909184)))];
            tensor<fp16, [320]> down_blocks_0_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16728448)))];
            tensor<fp16, [2, 320, 1, 1]> temb_3_cast = conv(bias = down_blocks_0_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_557, groups = var_112, pad = temb_3_pad_0, pad_type = temb_3_pad_type_0, strides = var_555, weight = down_blocks_0_resnets_1_time_emb_proj_weight_to_fp16, x = input_15_cast);
            tensor<fp16, [2, 320, 64, 64]> input_43_cast = add(x = hidden_states_19_cast, y = temb_3_cast);
            tensor<int32, [5]> reshape_16_shape_0 = const()[name = tensor<string, []>("reshape_16_shape_0"), val = tensor<int32, [5]>([2, 32, 10, 64, 64])];
            tensor<fp16, [2, 32, 10, 64, 64]> reshape_16_cast = reshape(shape = reshape_16_shape_0, x = input_43_cast);
            tensor<int32, [3]> reduce_mean_12_axes_0 = const()[name = tensor<string, []>("reduce_mean_12_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_12_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_12_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_12_cast = reduce_mean(axes = reduce_mean_12_axes_0, keep_dims = reduce_mean_12_keep_dims_0, x = reshape_16_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> sub_8_cast = sub(x = reshape_16_cast, y = reduce_mean_12_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> square_4_cast = square(x = sub_8_cast);
            tensor<int32, [3]> reduce_mean_14_axes_0 = const()[name = tensor<string, []>("reduce_mean_14_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_14_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_14_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_14_cast = reduce_mean(axes = reduce_mean_14_axes_0, keep_dims = reduce_mean_14_keep_dims_0, x = square_4_cast);
            tensor<fp16, []> add_8_y_0_to_fp16 = const()[name = tensor<string, []>("add_8_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_8_cast = add(x = reduce_mean_14_cast, y = add_8_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_4_cast = sqrt(x = add_8_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> real_div_4_cast = real_div(x = sub_8_cast, y = sqrt_4_cast);
            tensor<int32, [4]> reshape_17_shape_0 = const()[name = tensor<string, []>("reshape_17_shape_0"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> reshape_17_cast = reshape(shape = reshape_17_shape_0, x = real_div_4_cast);
            tensor<fp16, [320]> add_9_mean_0_to_fp16 = const()[name = tensor<string, []>("add_9_mean_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16729152)))];
            tensor<fp16, [320]> add_9_variance_0_to_fp16 = const()[name = tensor<string, []>("add_9_variance_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16729856)))];
            tensor<fp16, [320]> add_9_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_9_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16730560)))];
            tensor<fp16, [320]> add_9_beta_0_to_fp16 = const()[name = tensor<string, []>("add_9_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16731264)))];
            tensor<fp16, []> add_9_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_9_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 320, 64, 64]> add_9_cast = batch_norm(beta = add_9_beta_0_to_fp16, epsilon = add_9_epsilon_0_to_fp16, gamma = add_9_gamma_0_to_fp16, mean = add_9_mean_0_to_fp16, variance = add_9_variance_0_to_fp16, x = reshape_17_cast);
            tensor<fp16, [2, 320, 64, 64]> input_47_cast = silu(x = add_9_cast);
            tensor<int32, [2]> var_567 = const()[name = tensor<string, []>("op_567"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_569 = const()[name = tensor<string, []>("op_569"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_21_pad_type_0 = const()[name = tensor<string, []>("hidden_states_21_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_21_pad_0 = const()[name = tensor<string, []>("hidden_states_21_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [320, 320, 3, 3]> down_blocks_0_resnets_1_conv2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_resnets_1_conv2_weight_to_fp16"), val = tensor<fp16, [320, 320, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16731968)))];
            tensor<fp16, [320]> down_blocks_0_resnets_1_conv2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_resnets_1_conv2_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18575232)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_21_cast = conv(bias = down_blocks_0_resnets_1_conv2_bias_to_fp16, dilations = var_569, groups = var_112, pad = hidden_states_21_pad_0, pad_type = hidden_states_21_pad_type_0, strides = var_567, weight = down_blocks_0_resnets_1_conv2_weight_to_fp16, x = input_47_cast);
            tensor<fp16, [2, 320, 64, 64]> hidden_states_23_cast = add(x = input_35_cast, y = hidden_states_21_cast);
            tensor<int32, [5]> reshape_20_shape_0 = const()[name = tensor<string, []>("reshape_20_shape_0"), val = tensor<int32, [5]>([2, 32, 10, 64, 64])];
            tensor<fp16, [2, 32, 10, 64, 64]> reshape_20_cast = reshape(shape = reshape_20_shape_0, x = hidden_states_23_cast);
            tensor<int32, [3]> reduce_mean_15_axes_0 = const()[name = tensor<string, []>("reduce_mean_15_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_15_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_15_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_15_cast = reduce_mean(axes = reduce_mean_15_axes_0, keep_dims = reduce_mean_15_keep_dims_0, x = reshape_20_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> sub_10_cast = sub(x = reshape_20_cast, y = reduce_mean_15_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> square_5_cast = square(x = sub_10_cast);
            tensor<int32, [3]> reduce_mean_17_axes_0 = const()[name = tensor<string, []>("reduce_mean_17_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_17_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_17_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_17_cast = reduce_mean(axes = reduce_mean_17_axes_0, keep_dims = reduce_mean_17_keep_dims_0, x = square_5_cast);
            tensor<fp16, []> add_10_y_0_to_fp16 = const()[name = tensor<string, []>("add_10_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_10_cast = add(x = reduce_mean_17_cast, y = add_10_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_5_cast = sqrt(x = add_10_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> real_div_5_cast = real_div(x = sub_10_cast, y = sqrt_5_cast);
            tensor<int32, [4]> reshape_21_shape_0 = const()[name = tensor<string, []>("reshape_21_shape_0"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> reshape_21_cast = reshape(shape = reshape_21_shape_0, x = real_div_5_cast);
            tensor<fp16, [320]> add_11_mean_0_to_fp16 = const()[name = tensor<string, []>("add_11_mean_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18575936)))];
            tensor<fp16, [320]> add_11_variance_0_to_fp16 = const()[name = tensor<string, []>("add_11_variance_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18576640)))];
            tensor<fp16, [320]> add_11_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_11_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18577344)))];
            tensor<fp16, [320]> add_11_beta_0_to_fp16 = const()[name = tensor<string, []>("add_11_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18578048)))];
            tensor<fp16, []> add_11_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_11_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 320, 64, 64]> add_11_cast = batch_norm(beta = add_11_beta_0_to_fp16, epsilon = add_11_epsilon_0_to_fp16, gamma = add_11_gamma_0_to_fp16, mean = add_11_mean_0_to_fp16, variance = add_11_variance_0_to_fp16, x = reshape_21_cast);
            tensor<int32, [2]> var_589 = const()[name = tensor<string, []>("op_589"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_591 = const()[name = tensor<string, []>("op_591"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_25_pad_type_0 = const()[name = tensor<string, []>("hidden_states_25_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_25_pad_0 = const()[name = tensor<string, []>("hidden_states_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> down_blocks_0_attentions_1_proj_in_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_1_proj_in_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18578752)))];
            tensor<fp16, [320]> down_blocks_0_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_1_proj_in_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18783616)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_25_cast = conv(bias = down_blocks_0_attentions_1_proj_in_bias_to_fp16, dilations = var_591, groups = var_112, pad = hidden_states_25_pad_0, pad_type = hidden_states_25_pad_type_0, strides = var_589, weight = down_blocks_0_attentions_1_proj_in_weight_to_fp16, x = add_11_cast);
            tensor<int32, [4]> var_596 = const()[name = tensor<string, []>("op_596"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<fp16, [2, 320, 1, 4096]> inputs_7_cast = reshape(shape = var_596, x = hidden_states_25_cast);
            tensor<int32, [1]> var_606 = const()[name = tensor<string, []>("op_606"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_7_cast = reduce_mean(axes = var_606, keep_dims = var_107, x = inputs_7_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_7_cast = sub(x = inputs_7_cast, y = channels_mean_7_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_sq_7_cast = mul(x = zero_mean_7_cast, y = zero_mean_7_cast);
            tensor<int32, [1]> var_610 = const()[name = tensor<string, []>("op_610"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_611_cast = reduce_mean(axes = var_610, keep_dims = var_107, x = zero_mean_sq_7_cast);
            tensor<fp16, []> var_612_to_fp16 = const()[name = tensor<string, []>("op_612_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_613_cast = add(x = var_611_cast, y = var_612_to_fp16);
            tensor<fp16, []> denom_7_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_7_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_7_cast = rsqrt(epsilon = denom_7_epsilon_0_to_fp16, x = var_613_cast);
            tensor<fp16, [2, 320, 1, 4096]> out_7_cast = mul(x = zero_mean_7_cast, y = denom_7_cast);
            tensor<fp16, [1, 320, 1, 1]> var_617_to_fp16 = const()[name = tensor<string, []>("op_617_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18784320)))];
            tensor<fp16, [2, 320, 1, 4096]> var_618_cast = add(x = out_7_cast, y = var_617_to_fp16);
            tensor<fp16, [1, 320, 1, 1]> var_620_to_fp16 = const()[name = tensor<string, []>("op_620_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18785024)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_27_cast = mul(x = var_618_cast, y = var_620_to_fp16);
            tensor<int32, [2]> var_627 = const()[name = tensor<string, []>("op_627"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_629 = const()[name = tensor<string, []>("op_629"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_5_pad_type_0 = const()[name = tensor<string, []>("q_5_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_5_pad_0 = const()[name = tensor<string, []>("q_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18785728)))];
            tensor<fp16, [2, 320, 1, 4096]> q_5_cast = conv(dilations = var_629, groups = var_112, pad = q_5_pad_0, pad_type = q_5_pad_type_0, strides = var_627, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_27_cast);
            tensor<int32, [2]> var_633 = const()[name = tensor<string, []>("op_633"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_635 = const()[name = tensor<string, []>("op_635"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_9_pad_type_0 = const()[name = tensor<string, []>("k_9_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_9_pad_0 = const()[name = tensor<string, []>("k_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18990592)))];
            tensor<fp16, [2, 320, 1, 4096]> k_9_cast = conv(dilations = var_635, groups = var_112, pad = k_9_pad_0, pad_type = k_9_pad_type_0, strides = var_633, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_27_cast);
            tensor<int32, [2]> var_639 = const()[name = tensor<string, []>("op_639"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_641 = const()[name = tensor<string, []>("op_641"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_5_pad_type_0 = const()[name = tensor<string, []>("v_5_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_5_pad_0 = const()[name = tensor<string, []>("v_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19195456)))];
            tensor<fp16, [2, 320, 1, 4096]> v_5_cast = conv(dilations = var_641, groups = var_112, pad = v_5_pad_0, pad_type = v_5_pad_type_0, strides = var_639, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_27_cast);
            tensor<int32, [4]> var_645_begin_0 = const()[name = tensor<string, []>("op_645_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_645_end_0 = const()[name = tensor<string, []>("op_645_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_645_end_mask_0 = const()[name = tensor<string, []>("op_645_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_645_cast = slice_by_index(begin = var_645_begin_0, end = var_645_end_0, end_mask = var_645_end_mask_0, x = q_5_cast);
            tensor<int32, [4]> var_649_begin_0 = const()[name = tensor<string, []>("op_649_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_649_end_0 = const()[name = tensor<string, []>("op_649_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_649_end_mask_0 = const()[name = tensor<string, []>("op_649_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_649_cast = slice_by_index(begin = var_649_begin_0, end = var_649_end_0, end_mask = var_649_end_mask_0, x = q_5_cast);
            tensor<int32, [4]> var_653_begin_0 = const()[name = tensor<string, []>("op_653_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_653_end_0 = const()[name = tensor<string, []>("op_653_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_653_end_mask_0 = const()[name = tensor<string, []>("op_653_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_653_cast = slice_by_index(begin = var_653_begin_0, end = var_653_end_0, end_mask = var_653_end_mask_0, x = q_5_cast);
            tensor<int32, [4]> var_657_begin_0 = const()[name = tensor<string, []>("op_657_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_657_end_0 = const()[name = tensor<string, []>("op_657_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_657_end_mask_0 = const()[name = tensor<string, []>("op_657_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_657_cast = slice_by_index(begin = var_657_begin_0, end = var_657_end_0, end_mask = var_657_end_mask_0, x = q_5_cast);
            tensor<int32, [4]> var_661_begin_0 = const()[name = tensor<string, []>("op_661_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_661_end_0 = const()[name = tensor<string, []>("op_661_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_661_end_mask_0 = const()[name = tensor<string, []>("op_661_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_661_cast = slice_by_index(begin = var_661_begin_0, end = var_661_end_0, end_mask = var_661_end_mask_0, x = q_5_cast);
            tensor<int32, [4]> k_11_perm_0 = const()[name = tensor<string, []>("k_11_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_668_begin_0 = const()[name = tensor<string, []>("op_668_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_668_end_0 = const()[name = tensor<string, []>("op_668_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 64])];
            tensor<bool, [4]> var_668_end_mask_0 = const()[name = tensor<string, []>("op_668_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 320]> transpose_29 = transpose(perm = k_11_perm_0, x = k_9_cast);
            tensor<fp16, [2, 4096, 1, 64]> var_668_cast = slice_by_index(begin = var_668_begin_0, end = var_668_end_0, end_mask = var_668_end_mask_0, x = transpose_29);
            tensor<int32, [4]> var_672_begin_0 = const()[name = tensor<string, []>("op_672_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_672_end_0 = const()[name = tensor<string, []>("op_672_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 128])];
            tensor<bool, [4]> var_672_end_mask_0 = const()[name = tensor<string, []>("op_672_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_672_cast = slice_by_index(begin = var_672_begin_0, end = var_672_end_0, end_mask = var_672_end_mask_0, x = transpose_29);
            tensor<int32, [4]> var_676_begin_0 = const()[name = tensor<string, []>("op_676_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_676_end_0 = const()[name = tensor<string, []>("op_676_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 192])];
            tensor<bool, [4]> var_676_end_mask_0 = const()[name = tensor<string, []>("op_676_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_676_cast = slice_by_index(begin = var_676_begin_0, end = var_676_end_0, end_mask = var_676_end_mask_0, x = transpose_29);
            tensor<int32, [4]> var_680_begin_0 = const()[name = tensor<string, []>("op_680_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_680_end_0 = const()[name = tensor<string, []>("op_680_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 256])];
            tensor<bool, [4]> var_680_end_mask_0 = const()[name = tensor<string, []>("op_680_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_680_cast = slice_by_index(begin = var_680_begin_0, end = var_680_end_0, end_mask = var_680_end_mask_0, x = transpose_29);
            tensor<int32, [4]> var_684_begin_0 = const()[name = tensor<string, []>("op_684_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_684_end_0 = const()[name = tensor<string, []>("op_684_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_684_end_mask_0 = const()[name = tensor<string, []>("op_684_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_684_cast = slice_by_index(begin = var_684_begin_0, end = var_684_end_0, end_mask = var_684_end_mask_0, x = transpose_29);
            tensor<int32, [4]> var_686_begin_0 = const()[name = tensor<string, []>("op_686_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_686_end_0 = const()[name = tensor<string, []>("op_686_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_686_end_mask_0 = const()[name = tensor<string, []>("op_686_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_686_cast = slice_by_index(begin = var_686_begin_0, end = var_686_end_0, end_mask = var_686_end_mask_0, x = v_5_cast);
            tensor<int32, [4]> var_690_begin_0 = const()[name = tensor<string, []>("op_690_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_690_end_0 = const()[name = tensor<string, []>("op_690_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_690_end_mask_0 = const()[name = tensor<string, []>("op_690_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_690_cast = slice_by_index(begin = var_690_begin_0, end = var_690_end_0, end_mask = var_690_end_mask_0, x = v_5_cast);
            tensor<int32, [4]> var_694_begin_0 = const()[name = tensor<string, []>("op_694_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_694_end_0 = const()[name = tensor<string, []>("op_694_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_694_end_mask_0 = const()[name = tensor<string, []>("op_694_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_694_cast = slice_by_index(begin = var_694_begin_0, end = var_694_end_0, end_mask = var_694_end_mask_0, x = v_5_cast);
            tensor<int32, [4]> var_698_begin_0 = const()[name = tensor<string, []>("op_698_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_698_end_0 = const()[name = tensor<string, []>("op_698_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_698_end_mask_0 = const()[name = tensor<string, []>("op_698_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_698_cast = slice_by_index(begin = var_698_begin_0, end = var_698_end_0, end_mask = var_698_end_mask_0, x = v_5_cast);
            tensor<int32, [4]> var_702_begin_0 = const()[name = tensor<string, []>("op_702_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_702_end_0 = const()[name = tensor<string, []>("op_702_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_702_end_mask_0 = const()[name = tensor<string, []>("op_702_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_702_cast = slice_by_index(begin = var_702_begin_0, end = var_702_end_0, end_mask = var_702_end_mask_0, x = v_5_cast);
            tensor<string, []> var_706_equation_0 = const()[name = tensor<string, []>("op_706_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_706_cast = einsum(equation = var_706_equation_0, values = (var_668_cast, var_645_cast));
            tensor<fp16, []> var_707_to_fp16 = const()[name = tensor<string, []>("op_707_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_21_cast = mul(x = var_706_cast, y = var_707_to_fp16);
            tensor<string, []> var_710_equation_0 = const()[name = tensor<string, []>("op_710_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_710_cast = einsum(equation = var_710_equation_0, values = (var_672_cast, var_649_cast));
            tensor<fp16, []> var_711_to_fp16 = const()[name = tensor<string, []>("op_711_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_23_cast = mul(x = var_710_cast, y = var_711_to_fp16);
            tensor<string, []> var_714_equation_0 = const()[name = tensor<string, []>("op_714_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_714_cast = einsum(equation = var_714_equation_0, values = (var_676_cast, var_653_cast));
            tensor<fp16, []> var_715_to_fp16 = const()[name = tensor<string, []>("op_715_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_25_cast = mul(x = var_714_cast, y = var_715_to_fp16);
            tensor<string, []> var_718_equation_0 = const()[name = tensor<string, []>("op_718_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_718_cast = einsum(equation = var_718_equation_0, values = (var_680_cast, var_657_cast));
            tensor<fp16, []> var_719_to_fp16 = const()[name = tensor<string, []>("op_719_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_27_cast = mul(x = var_718_cast, y = var_719_to_fp16);
            tensor<string, []> var_722_equation_0 = const()[name = tensor<string, []>("op_722_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_722_cast = einsum(equation = var_722_equation_0, values = (var_684_cast, var_661_cast));
            tensor<fp16, []> var_723_to_fp16 = const()[name = tensor<string, []>("op_723_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_29_cast = mul(x = var_722_cast, y = var_723_to_fp16);
            tensor<fp16, [2, 4096, 1, 4096]> var_725_cast = softmax(axis = var_112, x = aw_21_cast);
            tensor<fp16, [2, 4096, 1, 4096]> var_726_cast = softmax(axis = var_112, x = aw_23_cast);
            tensor<fp16, [2, 4096, 1, 4096]> var_727_cast = softmax(axis = var_112, x = aw_25_cast);
            tensor<fp16, [2, 4096, 1, 4096]> var_728_cast = softmax(axis = var_112, x = aw_27_cast);
            tensor<fp16, [2, 4096, 1, 4096]> var_729_cast = softmax(axis = var_112, x = aw_29_cast);
            tensor<string, []> var_731_equation_0 = const()[name = tensor<string, []>("op_731_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_731_cast = einsum(equation = var_731_equation_0, values = (var_686_cast, var_725_cast));
            tensor<string, []> var_733_equation_0 = const()[name = tensor<string, []>("op_733_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_733_cast = einsum(equation = var_733_equation_0, values = (var_690_cast, var_726_cast));
            tensor<string, []> var_735_equation_0 = const()[name = tensor<string, []>("op_735_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_735_cast = einsum(equation = var_735_equation_0, values = (var_694_cast, var_727_cast));
            tensor<string, []> var_737_equation_0 = const()[name = tensor<string, []>("op_737_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_737_cast = einsum(equation = var_737_equation_0, values = (var_698_cast, var_728_cast));
            tensor<string, []> var_739_equation_0 = const()[name = tensor<string, []>("op_739_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_739_cast = einsum(equation = var_739_equation_0, values = (var_702_cast, var_729_cast));
            tensor<bool, []> input_51_interleave_0 = const()[name = tensor<string, []>("input_51_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 320, 1, 4096]> input_51_cast = concat(axis = var_112, interleave = input_51_interleave_0, values = (var_731_cast, var_733_cast, var_735_cast, var_737_cast, var_739_cast));
            tensor<int32, [2]> var_745 = const()[name = tensor<string, []>("op_745"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_747 = const()[name = tensor<string, []>("op_747"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_749_pad_type_0 = const()[name = tensor<string, []>("op_749_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_749_pad_0 = const()[name = tensor<string, []>("op_749_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19400320)))];
            tensor<fp16, [320]> down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19605184)))];
            tensor<fp16, [2, 320, 1, 4096]> var_749_cast = conv(bias = down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_747, groups = var_112, pad = var_749_pad_0, pad_type = var_749_pad_type_0, strides = var_745, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_51_cast);
            tensor<fp16, [2, 320, 1, 4096]> inputs_9_cast = add(x = var_749_cast, y = inputs_7_cast);
            tensor<int32, [1]> var_753 = const()[name = tensor<string, []>("op_753"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_9_cast = reduce_mean(axes = var_753, keep_dims = var_107, x = inputs_9_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_9_cast = sub(x = inputs_9_cast, y = channels_mean_9_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_sq_9_cast = mul(x = zero_mean_9_cast, y = zero_mean_9_cast);
            tensor<int32, [1]> var_757 = const()[name = tensor<string, []>("op_757"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_758_cast = reduce_mean(axes = var_757, keep_dims = var_107, x = zero_mean_sq_9_cast);
            tensor<fp16, []> var_759_to_fp16 = const()[name = tensor<string, []>("op_759_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_760_cast = add(x = var_758_cast, y = var_759_to_fp16);
            tensor<fp16, []> denom_9_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_9_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_9_cast = rsqrt(epsilon = denom_9_epsilon_0_to_fp16, x = var_760_cast);
            tensor<fp16, [2, 320, 1, 4096]> out_9_cast = mul(x = zero_mean_9_cast, y = denom_9_cast);
            tensor<fp16, [1, 320, 1, 1]> var_764_to_fp16 = const()[name = tensor<string, []>("op_764_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19605888)))];
            tensor<fp16, [2, 320, 1, 4096]> var_765_cast = add(x = out_9_cast, y = var_764_to_fp16);
            tensor<fp16, [1, 320, 1, 1]> var_767_to_fp16 = const()[name = tensor<string, []>("op_767_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19606592)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_29_cast = mul(x = var_765_cast, y = var_767_to_fp16);
            tensor<int32, [2]> var_774 = const()[name = tensor<string, []>("op_774"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_776 = const()[name = tensor<string, []>("op_776"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_7_pad_type_0 = const()[name = tensor<string, []>("q_7_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_7_pad_0 = const()[name = tensor<string, []>("q_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19607296)))];
            tensor<fp16, [2, 320, 1, 4096]> q_7_cast = conv(dilations = var_776, groups = var_112, pad = q_7_pad_0, pad_type = q_7_pad_type_0, strides = var_774, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_29_cast);
            tensor<int32, [2]> var_780 = const()[name = tensor<string, []>("op_780"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_782 = const()[name = tensor<string, []>("op_782"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_13_pad_type_0 = const()[name = tensor<string, []>("k_13_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_13_pad_0 = const()[name = tensor<string, []>("k_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1024, 1, 1]> down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [320, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19812160)))];
            tensor<fp16, [2, 320, 1, 77]> k_13_cast = conv(dilations = var_782, groups = var_112, pad = k_13_pad_0, pad_type = k_13_pad_type_0, strides = var_780, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [2]> var_786 = const()[name = tensor<string, []>("op_786"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_788 = const()[name = tensor<string, []>("op_788"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_7_pad_type_0 = const()[name = tensor<string, []>("v_7_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_7_pad_0 = const()[name = tensor<string, []>("v_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1024, 1, 1]> down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [320, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20467584)))];
            tensor<fp16, [2, 320, 1, 77]> v_7_cast = conv(dilations = var_788, groups = var_112, pad = v_7_pad_0, pad_type = v_7_pad_type_0, strides = var_786, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [4]> var_792_begin_0 = const()[name = tensor<string, []>("op_792_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_792_end_0 = const()[name = tensor<string, []>("op_792_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_792_end_mask_0 = const()[name = tensor<string, []>("op_792_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_792_cast = slice_by_index(begin = var_792_begin_0, end = var_792_end_0, end_mask = var_792_end_mask_0, x = q_7_cast);
            tensor<int32, [4]> var_796_begin_0 = const()[name = tensor<string, []>("op_796_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_796_end_0 = const()[name = tensor<string, []>("op_796_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_796_end_mask_0 = const()[name = tensor<string, []>("op_796_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_796_cast = slice_by_index(begin = var_796_begin_0, end = var_796_end_0, end_mask = var_796_end_mask_0, x = q_7_cast);
            tensor<int32, [4]> var_800_begin_0 = const()[name = tensor<string, []>("op_800_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_800_end_0 = const()[name = tensor<string, []>("op_800_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_800_end_mask_0 = const()[name = tensor<string, []>("op_800_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_800_cast = slice_by_index(begin = var_800_begin_0, end = var_800_end_0, end_mask = var_800_end_mask_0, x = q_7_cast);
            tensor<int32, [4]> var_804_begin_0 = const()[name = tensor<string, []>("op_804_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_804_end_0 = const()[name = tensor<string, []>("op_804_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_804_end_mask_0 = const()[name = tensor<string, []>("op_804_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_804_cast = slice_by_index(begin = var_804_begin_0, end = var_804_end_0, end_mask = var_804_end_mask_0, x = q_7_cast);
            tensor<int32, [4]> var_808_begin_0 = const()[name = tensor<string, []>("op_808_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_808_end_0 = const()[name = tensor<string, []>("op_808_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_808_end_mask_0 = const()[name = tensor<string, []>("op_808_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_808_cast = slice_by_index(begin = var_808_begin_0, end = var_808_end_0, end_mask = var_808_end_mask_0, x = q_7_cast);
            tensor<int32, [4]> k_15_perm_0 = const()[name = tensor<string, []>("k_15_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_815_begin_0 = const()[name = tensor<string, []>("op_815_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_815_end_0 = const()[name = tensor<string, []>("op_815_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_815_end_mask_0 = const()[name = tensor<string, []>("op_815_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 320]> transpose_28 = transpose(perm = k_15_perm_0, x = k_13_cast);
            tensor<fp16, [2, 77, 1, 64]> var_815_cast = slice_by_index(begin = var_815_begin_0, end = var_815_end_0, end_mask = var_815_end_mask_0, x = transpose_28);
            tensor<int32, [4]> var_819_begin_0 = const()[name = tensor<string, []>("op_819_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_819_end_0 = const()[name = tensor<string, []>("op_819_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_819_end_mask_0 = const()[name = tensor<string, []>("op_819_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_819_cast = slice_by_index(begin = var_819_begin_0, end = var_819_end_0, end_mask = var_819_end_mask_0, x = transpose_28);
            tensor<int32, [4]> var_823_begin_0 = const()[name = tensor<string, []>("op_823_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_823_end_0 = const()[name = tensor<string, []>("op_823_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_823_end_mask_0 = const()[name = tensor<string, []>("op_823_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_823_cast = slice_by_index(begin = var_823_begin_0, end = var_823_end_0, end_mask = var_823_end_mask_0, x = transpose_28);
            tensor<int32, [4]> var_827_begin_0 = const()[name = tensor<string, []>("op_827_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_827_end_0 = const()[name = tensor<string, []>("op_827_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_827_end_mask_0 = const()[name = tensor<string, []>("op_827_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_827_cast = slice_by_index(begin = var_827_begin_0, end = var_827_end_0, end_mask = var_827_end_mask_0, x = transpose_28);
            tensor<int32, [4]> var_831_begin_0 = const()[name = tensor<string, []>("op_831_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_831_end_0 = const()[name = tensor<string, []>("op_831_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_831_end_mask_0 = const()[name = tensor<string, []>("op_831_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_831_cast = slice_by_index(begin = var_831_begin_0, end = var_831_end_0, end_mask = var_831_end_mask_0, x = transpose_28);
            tensor<int32, [4]> var_833_begin_0 = const()[name = tensor<string, []>("op_833_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_833_end_0 = const()[name = tensor<string, []>("op_833_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_833_end_mask_0 = const()[name = tensor<string, []>("op_833_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_833_cast = slice_by_index(begin = var_833_begin_0, end = var_833_end_0, end_mask = var_833_end_mask_0, x = v_7_cast);
            tensor<int32, [4]> var_837_begin_0 = const()[name = tensor<string, []>("op_837_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_837_end_0 = const()[name = tensor<string, []>("op_837_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_837_end_mask_0 = const()[name = tensor<string, []>("op_837_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_837_cast = slice_by_index(begin = var_837_begin_0, end = var_837_end_0, end_mask = var_837_end_mask_0, x = v_7_cast);
            tensor<int32, [4]> var_841_begin_0 = const()[name = tensor<string, []>("op_841_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_841_end_0 = const()[name = tensor<string, []>("op_841_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_841_end_mask_0 = const()[name = tensor<string, []>("op_841_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_841_cast = slice_by_index(begin = var_841_begin_0, end = var_841_end_0, end_mask = var_841_end_mask_0, x = v_7_cast);
            tensor<int32, [4]> var_845_begin_0 = const()[name = tensor<string, []>("op_845_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_845_end_0 = const()[name = tensor<string, []>("op_845_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_845_end_mask_0 = const()[name = tensor<string, []>("op_845_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_845_cast = slice_by_index(begin = var_845_begin_0, end = var_845_end_0, end_mask = var_845_end_mask_0, x = v_7_cast);
            tensor<int32, [4]> var_849_begin_0 = const()[name = tensor<string, []>("op_849_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_849_end_0 = const()[name = tensor<string, []>("op_849_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_849_end_mask_0 = const()[name = tensor<string, []>("op_849_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_849_cast = slice_by_index(begin = var_849_begin_0, end = var_849_end_0, end_mask = var_849_end_mask_0, x = v_7_cast);
            tensor<string, []> var_853_equation_0 = const()[name = tensor<string, []>("op_853_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_853_cast = einsum(equation = var_853_equation_0, values = (var_815_cast, var_792_cast));
            tensor<fp16, []> var_854_to_fp16 = const()[name = tensor<string, []>("op_854_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_31_cast = mul(x = var_853_cast, y = var_854_to_fp16);
            tensor<string, []> var_857_equation_0 = const()[name = tensor<string, []>("op_857_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_857_cast = einsum(equation = var_857_equation_0, values = (var_819_cast, var_796_cast));
            tensor<fp16, []> var_858_to_fp16 = const()[name = tensor<string, []>("op_858_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_33_cast = mul(x = var_857_cast, y = var_858_to_fp16);
            tensor<string, []> var_861_equation_0 = const()[name = tensor<string, []>("op_861_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_861_cast = einsum(equation = var_861_equation_0, values = (var_823_cast, var_800_cast));
            tensor<fp16, []> var_862_to_fp16 = const()[name = tensor<string, []>("op_862_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_35_cast = mul(x = var_861_cast, y = var_862_to_fp16);
            tensor<string, []> var_865_equation_0 = const()[name = tensor<string, []>("op_865_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_865_cast = einsum(equation = var_865_equation_0, values = (var_827_cast, var_804_cast));
            tensor<fp16, []> var_866_to_fp16 = const()[name = tensor<string, []>("op_866_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_37_cast = mul(x = var_865_cast, y = var_866_to_fp16);
            tensor<string, []> var_869_equation_0 = const()[name = tensor<string, []>("op_869_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_869_cast = einsum(equation = var_869_equation_0, values = (var_831_cast, var_808_cast));
            tensor<fp16, []> var_870_to_fp16 = const()[name = tensor<string, []>("op_870_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_39_cast = mul(x = var_869_cast, y = var_870_to_fp16);
            tensor<fp16, [2, 77, 1, 4096]> var_872_cast = softmax(axis = var_112, x = aw_31_cast);
            tensor<fp16, [2, 77, 1, 4096]> var_873_cast = softmax(axis = var_112, x = aw_33_cast);
            tensor<fp16, [2, 77, 1, 4096]> var_874_cast = softmax(axis = var_112, x = aw_35_cast);
            tensor<fp16, [2, 77, 1, 4096]> var_875_cast = softmax(axis = var_112, x = aw_37_cast);
            tensor<fp16, [2, 77, 1, 4096]> var_876_cast = softmax(axis = var_112, x = aw_39_cast);
            tensor<string, []> var_878_equation_0 = const()[name = tensor<string, []>("op_878_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_878_cast = einsum(equation = var_878_equation_0, values = (var_833_cast, var_872_cast));
            tensor<string, []> var_880_equation_0 = const()[name = tensor<string, []>("op_880_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_880_cast = einsum(equation = var_880_equation_0, values = (var_837_cast, var_873_cast));
            tensor<string, []> var_882_equation_0 = const()[name = tensor<string, []>("op_882_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_882_cast = einsum(equation = var_882_equation_0, values = (var_841_cast, var_874_cast));
            tensor<string, []> var_884_equation_0 = const()[name = tensor<string, []>("op_884_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_884_cast = einsum(equation = var_884_equation_0, values = (var_845_cast, var_875_cast));
            tensor<string, []> var_886_equation_0 = const()[name = tensor<string, []>("op_886_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_886_cast = einsum(equation = var_886_equation_0, values = (var_849_cast, var_876_cast));
            tensor<bool, []> input_53_interleave_0 = const()[name = tensor<string, []>("input_53_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 320, 1, 4096]> input_53_cast = concat(axis = var_112, interleave = input_53_interleave_0, values = (var_878_cast, var_880_cast, var_882_cast, var_884_cast, var_886_cast));
            tensor<int32, [2]> var_892 = const()[name = tensor<string, []>("op_892"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_894 = const()[name = tensor<string, []>("op_894"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_896_pad_type_0 = const()[name = tensor<string, []>("op_896_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_896_pad_0 = const()[name = tensor<string, []>("op_896_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21123008)))];
            tensor<fp16, [320]> down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21327872)))];
            tensor<fp16, [2, 320, 1, 4096]> var_896_cast = conv(bias = down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_894, groups = var_112, pad = var_896_pad_0, pad_type = var_896_pad_type_0, strides = var_892, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_53_cast);
            tensor<fp16, [2, 320, 1, 4096]> inputs_11_cast = add(x = var_896_cast, y = inputs_9_cast);
            tensor<int32, [1]> var_900 = const()[name = tensor<string, []>("op_900"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_11_cast = reduce_mean(axes = var_900, keep_dims = var_107, x = inputs_11_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_11_cast = sub(x = inputs_11_cast, y = channels_mean_11_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_sq_11_cast = mul(x = zero_mean_11_cast, y = zero_mean_11_cast);
            tensor<int32, [1]> var_904 = const()[name = tensor<string, []>("op_904"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_905_cast = reduce_mean(axes = var_904, keep_dims = var_107, x = zero_mean_sq_11_cast);
            tensor<fp16, []> var_906_to_fp16 = const()[name = tensor<string, []>("op_906_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_907_cast = add(x = var_905_cast, y = var_906_to_fp16);
            tensor<fp16, []> denom_11_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_11_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_11_cast = rsqrt(epsilon = denom_11_epsilon_0_to_fp16, x = var_907_cast);
            tensor<fp16, [2, 320, 1, 4096]> out_11_cast = mul(x = zero_mean_11_cast, y = denom_11_cast);
            tensor<fp16, [1, 320, 1, 1]> var_911_to_fp16 = const()[name = tensor<string, []>("op_911_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21328576)))];
            tensor<fp16, [2, 320, 1, 4096]> var_912_cast = add(x = out_11_cast, y = var_911_to_fp16);
            tensor<fp16, [1, 320, 1, 1]> var_914_to_fp16 = const()[name = tensor<string, []>("op_914_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21329280)))];
            tensor<fp16, [2, 320, 1, 4096]> input_55_cast = mul(x = var_912_cast, y = var_914_to_fp16);
            tensor<int32, [2]> var_922 = const()[name = tensor<string, []>("op_922"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_924 = const()[name = tensor<string, []>("op_924"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_926_pad_type_0 = const()[name = tensor<string, []>("op_926_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_926_pad_0 = const()[name = tensor<string, []>("op_926_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [2560, 320, 1, 1]> down_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [2560, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21329984)))];
            tensor<fp16, [2560]> down_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22968448)))];
            tensor<fp16, [2, 2560, 1, 4096]> var_926_cast = conv(bias = down_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_924, groups = var_112, pad = var_926_pad_0, pad_type = var_926_pad_type_0, strides = var_922, weight = down_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_55_cast);
            tensor<int32, [2]> var_927_split_sizes_0 = const()[name = tensor<string, []>("op_927_split_sizes_0"), val = tensor<int32, [2]>([1280, 1280])];
            tensor<int32, []> var_927_axis_0 = const()[name = tensor<string, []>("op_927_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 1280, 1, 4096]> var_927_cast_0, tensor<fp16, [2, 1280, 1, 4096]> var_927_cast_1 = split(axis = var_927_axis_0, split_sizes = var_927_split_sizes_0, x = var_926_cast);
            tensor<string, []> var_929_mode_0 = const()[name = tensor<string, []>("op_929_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 1280, 1, 4096]> var_929_cast = gelu(mode = var_929_mode_0, x = var_927_cast_1);
            tensor<fp16, [2, 1280, 1, 4096]> input_57_cast = mul(x = var_927_cast_0, y = var_929_cast);
            tensor<int32, [2]> var_933 = const()[name = tensor<string, []>("op_933"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_935 = const()[name = tensor<string, []>("op_935"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_937_pad_type_0 = const()[name = tensor<string, []>("op_937_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_937_pad_0 = const()[name = tensor<string, []>("op_937_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1280, 1, 1]> down_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [320, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22973632)))];
            tensor<fp16, [320]> down_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23792896)))];
            tensor<fp16, [2, 320, 1, 4096]> var_937_cast = conv(bias = down_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_935, groups = var_112, pad = var_937_pad_0, pad_type = var_937_pad_type_0, strides = var_933, weight = down_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_57_cast);
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_33_cast = add(x = var_937_cast, y = inputs_11_cast);
            tensor<int32, [4]> var_939 = const()[name = tensor<string, []>("op_939"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> input_59_cast = reshape(shape = var_939, x = hidden_states_33_cast);
            tensor<int32, [2]> var_943 = const()[name = tensor<string, []>("op_943"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_945 = const()[name = tensor<string, []>("op_945"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_35_pad_type_0 = const()[name = tensor<string, []>("hidden_states_35_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_35_pad_0 = const()[name = tensor<string, []>("hidden_states_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> down_blocks_0_attentions_1_proj_out_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_1_proj_out_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23793600)))];
            tensor<fp16, [320]> down_blocks_0_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_attentions_1_proj_out_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23998464)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_35_cast = conv(bias = down_blocks_0_attentions_1_proj_out_bias_to_fp16, dilations = var_945, groups = var_112, pad = hidden_states_35_pad_0, pad_type = hidden_states_35_pad_type_0, strides = var_943, weight = down_blocks_0_attentions_1_proj_out_weight_to_fp16, x = input_59_cast);
            tensor<fp16, [2, 320, 64, 64]> input_61_cast = add(x = hidden_states_35_cast, y = hidden_states_23_cast);
            tensor<int32, [2]> var_952 = const()[name = tensor<string, []>("op_952"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [2]> var_954 = const()[name = tensor<string, []>("op_954"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> input_63_pad_type_0 = const()[name = tensor<string, []>("input_63_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> input_63_pad_0 = const()[name = tensor<string, []>("input_63_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [320, 320, 3, 3]> down_blocks_0_downsamplers_0_conv_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_downsamplers_0_conv_weight_to_fp16"), val = tensor<fp16, [320, 320, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23999168)))];
            tensor<fp16, [320]> down_blocks_0_downsamplers_0_conv_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_0_downsamplers_0_conv_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25842432)))];
            tensor<fp16, [2, 320, 32, 32]> input_63_cast = conv(bias = down_blocks_0_downsamplers_0_conv_bias_to_fp16, dilations = var_954, groups = var_112, pad = input_63_pad_0, pad_type = input_63_pad_type_0, strides = var_952, weight = down_blocks_0_downsamplers_0_conv_weight_to_fp16, x = input_61_cast);
            tensor<bool, []> var_979 = const()[name = tensor<string, []>("op_979"), val = tensor<bool, []>(true)];
            tensor<int32, []> var_984 = const()[name = tensor<string, []>("op_984"), val = tensor<int32, []>(1)];
            tensor<int32, [5]> reshape_24_shape_0 = const()[name = tensor<string, []>("reshape_24_shape_0"), val = tensor<int32, [5]>([2, 32, 10, 32, 32])];
            tensor<fp16, [2, 32, 10, 32, 32]> reshape_24_cast = reshape(shape = reshape_24_shape_0, x = input_63_cast);
            tensor<int32, [3]> reduce_mean_18_axes_0 = const()[name = tensor<string, []>("reduce_mean_18_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_18_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_18_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_18_cast = reduce_mean(axes = reduce_mean_18_axes_0, keep_dims = reduce_mean_18_keep_dims_0, x = reshape_24_cast);
            tensor<fp16, [2, 32, 10, 32, 32]> sub_12_cast = sub(x = reshape_24_cast, y = reduce_mean_18_cast);
            tensor<fp16, [2, 32, 10, 32, 32]> square_6_cast = square(x = sub_12_cast);
            tensor<int32, [3]> reduce_mean_20_axes_0 = const()[name = tensor<string, []>("reduce_mean_20_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_20_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_20_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_20_cast = reduce_mean(axes = reduce_mean_20_axes_0, keep_dims = reduce_mean_20_keep_dims_0, x = square_6_cast);
            tensor<fp16, []> add_12_y_0_to_fp16 = const()[name = tensor<string, []>("add_12_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_12_cast = add(x = reduce_mean_20_cast, y = add_12_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_6_cast = sqrt(x = add_12_cast);
            tensor<fp16, [2, 32, 10, 32, 32]> real_div_6_cast = real_div(x = sub_12_cast, y = sqrt_6_cast);
            tensor<int32, [4]> reshape_25_shape_0 = const()[name = tensor<string, []>("reshape_25_shape_0"), val = tensor<int32, [4]>([2, 320, 32, 32])];
            tensor<fp16, [2, 320, 32, 32]> reshape_25_cast = reshape(shape = reshape_25_shape_0, x = real_div_6_cast);
            tensor<fp16, [320]> add_13_mean_0_to_fp16 = const()[name = tensor<string, []>("add_13_mean_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25843136)))];
            tensor<fp16, [320]> add_13_variance_0_to_fp16 = const()[name = tensor<string, []>("add_13_variance_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25843840)))];
            tensor<fp16, [320]> add_13_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_13_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25844544)))];
            tensor<fp16, [320]> add_13_beta_0_to_fp16 = const()[name = tensor<string, []>("add_13_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25845248)))];
            tensor<fp16, []> add_13_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_13_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 320, 32, 32]> add_13_cast = batch_norm(beta = add_13_beta_0_to_fp16, epsilon = add_13_epsilon_0_to_fp16, gamma = add_13_gamma_0_to_fp16, mean = add_13_mean_0_to_fp16, variance = add_13_variance_0_to_fp16, x = reshape_25_cast);
            tensor<fp16, [2, 320, 32, 32]> input_67_cast = silu(x = add_13_cast);
            tensor<int32, [2]> var_1007 = const()[name = tensor<string, []>("op_1007"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1009 = const()[name = tensor<string, []>("op_1009"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_37_pad_type_0 = const()[name = tensor<string, []>("hidden_states_37_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_37_pad_0 = const()[name = tensor<string, []>("hidden_states_37_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [640, 320, 3, 3]> down_blocks_1_resnets_0_conv1_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_resnets_0_conv1_weight_to_fp16"), val = tensor<fp16, [640, 320, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25845952)))];
            tensor<fp16, [640]> down_blocks_1_resnets_0_conv1_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_resnets_0_conv1_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29532416)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_37_cast = conv(bias = down_blocks_1_resnets_0_conv1_bias_to_fp16, dilations = var_1009, groups = var_984, pad = hidden_states_37_pad_0, pad_type = hidden_states_37_pad_type_0, strides = var_1007, weight = down_blocks_1_resnets_0_conv1_weight_to_fp16, x = input_67_cast);
            tensor<int32, [2]> var_1015 = const()[name = tensor<string, []>("op_1015"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1017 = const()[name = tensor<string, []>("op_1017"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_5_pad_type_0 = const()[name = tensor<string, []>("temb_5_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_5_pad_0 = const()[name = tensor<string, []>("temb_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1280, 1, 1]> down_blocks_1_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [640, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29533760)))];
            tensor<fp16, [640]> down_blocks_1_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31172224)))];
            tensor<fp16, [2, 640, 1, 1]> temb_5_cast = conv(bias = down_blocks_1_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_1017, groups = var_984, pad = temb_5_pad_0, pad_type = temb_5_pad_type_0, strides = var_1015, weight = down_blocks_1_resnets_0_time_emb_proj_weight_to_fp16, x = input_15_cast);
            tensor<fp16, [2, 640, 32, 32]> input_71_cast = add(x = hidden_states_37_cast, y = temb_5_cast);
            tensor<int32, [5]> reshape_28_shape_0 = const()[name = tensor<string, []>("reshape_28_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 32, 32])];
            tensor<fp16, [2, 32, 20, 32, 32]> reshape_28_cast = reshape(shape = reshape_28_shape_0, x = input_71_cast);
            tensor<int32, [3]> reduce_mean_21_axes_0 = const()[name = tensor<string, []>("reduce_mean_21_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_21_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_21_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_21_cast = reduce_mean(axes = reduce_mean_21_axes_0, keep_dims = reduce_mean_21_keep_dims_0, x = reshape_28_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> sub_14_cast = sub(x = reshape_28_cast, y = reduce_mean_21_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> square_7_cast = square(x = sub_14_cast);
            tensor<int32, [3]> reduce_mean_23_axes_0 = const()[name = tensor<string, []>("reduce_mean_23_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_23_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_23_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_23_cast = reduce_mean(axes = reduce_mean_23_axes_0, keep_dims = reduce_mean_23_keep_dims_0, x = square_7_cast);
            tensor<fp16, []> add_14_y_0_to_fp16 = const()[name = tensor<string, []>("add_14_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_14_cast = add(x = reduce_mean_23_cast, y = add_14_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_7_cast = sqrt(x = add_14_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> real_div_7_cast = real_div(x = sub_14_cast, y = sqrt_7_cast);
            tensor<int32, [4]> reshape_29_shape_0 = const()[name = tensor<string, []>("reshape_29_shape_0"), val = tensor<int32, [4]>([2, 640, 32, 32])];
            tensor<fp16, [2, 640, 32, 32]> reshape_29_cast = reshape(shape = reshape_29_shape_0, x = real_div_7_cast);
            tensor<fp16, [640]> add_15_mean_0_to_fp16 = const()[name = tensor<string, []>("add_15_mean_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31173568)))];
            tensor<fp16, [640]> add_15_variance_0_to_fp16 = const()[name = tensor<string, []>("add_15_variance_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31174912)))];
            tensor<fp16, [640]> add_15_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_15_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31176256)))];
            tensor<fp16, [640]> add_15_beta_0_to_fp16 = const()[name = tensor<string, []>("add_15_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31177600)))];
            tensor<fp16, []> add_15_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_15_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 32, 32]> add_15_cast = batch_norm(beta = add_15_beta_0_to_fp16, epsilon = add_15_epsilon_0_to_fp16, gamma = add_15_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_29_cast);
            tensor<fp16, [2, 640, 32, 32]> input_75_cast = silu(x = add_15_cast);
            tensor<int32, [2]> var_1027 = const()[name = tensor<string, []>("op_1027"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1029 = const()[name = tensor<string, []>("op_1029"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_39_pad_type_0 = const()[name = tensor<string, []>("hidden_states_39_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_39_pad_0 = const()[name = tensor<string, []>("hidden_states_39_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [640, 640, 3, 3]> down_blocks_1_resnets_0_conv2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_resnets_0_conv2_weight_to_fp16"), val = tensor<fp16, [640, 640, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31178944)))];
            tensor<fp16, [640]> down_blocks_1_resnets_0_conv2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_resnets_0_conv2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38551808)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_39_cast = conv(bias = down_blocks_1_resnets_0_conv2_bias_to_fp16, dilations = var_1029, groups = var_984, pad = hidden_states_39_pad_0, pad_type = hidden_states_39_pad_type_0, strides = var_1027, weight = down_blocks_1_resnets_0_conv2_weight_to_fp16, x = input_75_cast);
            tensor<int32, [2]> var_1034 = const()[name = tensor<string, []>("op_1034"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1036 = const()[name = tensor<string, []>("op_1036"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> x_1_pad_type_0 = const()[name = tensor<string, []>("x_1_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> x_1_pad_0 = const()[name = tensor<string, []>("x_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 320, 1, 1]> down_blocks_1_resnets_0_conv_shortcut_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_resnets_0_conv_shortcut_weight_to_fp16"), val = tensor<fp16, [640, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38553152)))];
            tensor<fp16, [640]> down_blocks_1_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38962816)))];
            tensor<fp16, [2, 640, 32, 32]> x_1_cast = conv(bias = down_blocks_1_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_1036, groups = var_984, pad = x_1_pad_0, pad_type = x_1_pad_type_0, strides = var_1034, weight = down_blocks_1_resnets_0_conv_shortcut_weight_to_fp16, x = input_63_cast);
            tensor<fp16, [2, 640, 32, 32]> hidden_states_41_cast = add(x = x_1_cast, y = hidden_states_39_cast);
            tensor<int32, [5]> reshape_32_shape_0 = const()[name = tensor<string, []>("reshape_32_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 32, 32])];
            tensor<fp16, [2, 32, 20, 32, 32]> reshape_32_cast = reshape(shape = reshape_32_shape_0, x = hidden_states_41_cast);
            tensor<int32, [3]> reduce_mean_24_axes_0 = const()[name = tensor<string, []>("reduce_mean_24_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_24_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_24_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_24_cast = reduce_mean(axes = reduce_mean_24_axes_0, keep_dims = reduce_mean_24_keep_dims_0, x = reshape_32_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> sub_16_cast = sub(x = reshape_32_cast, y = reduce_mean_24_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> square_8_cast = square(x = sub_16_cast);
            tensor<int32, [3]> reduce_mean_26_axes_0 = const()[name = tensor<string, []>("reduce_mean_26_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_26_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_26_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_26_cast = reduce_mean(axes = reduce_mean_26_axes_0, keep_dims = reduce_mean_26_keep_dims_0, x = square_8_cast);
            tensor<fp16, []> add_16_y_0_to_fp16 = const()[name = tensor<string, []>("add_16_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_16_cast = add(x = reduce_mean_26_cast, y = add_16_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_8_cast = sqrt(x = add_16_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> real_div_8_cast = real_div(x = sub_16_cast, y = sqrt_8_cast);
            tensor<int32, [4]> reshape_33_shape_0 = const()[name = tensor<string, []>("reshape_33_shape_0"), val = tensor<int32, [4]>([2, 640, 32, 32])];
            tensor<fp16, [2, 640, 32, 32]> reshape_33_cast = reshape(shape = reshape_33_shape_0, x = real_div_8_cast);
            tensor<fp16, [640]> add_17_mean_0_to_fp16 = const()[name = tensor<string, []>("add_17_mean_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38964160)))];
            tensor<fp16, [640]> add_17_variance_0_to_fp16 = const()[name = tensor<string, []>("add_17_variance_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38965504)))];
            tensor<fp16, [640]> add_17_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_17_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38966848)))];
            tensor<fp16, [640]> add_17_beta_0_to_fp16 = const()[name = tensor<string, []>("add_17_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38968192)))];
            tensor<fp16, []> add_17_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_17_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 32, 32]> add_17_cast = batch_norm(beta = add_17_beta_0_to_fp16, epsilon = add_17_epsilon_0_to_fp16, gamma = add_17_gamma_0_to_fp16, mean = add_17_mean_0_to_fp16, variance = add_17_variance_0_to_fp16, x = reshape_33_cast);
            tensor<int32, [2]> var_1056 = const()[name = tensor<string, []>("op_1056"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1058 = const()[name = tensor<string, []>("op_1058"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_43_pad_type_0 = const()[name = tensor<string, []>("hidden_states_43_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_43_pad_0 = const()[name = tensor<string, []>("hidden_states_43_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_proj_in_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_proj_in_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38969536)))];
            tensor<fp16, [640]> down_blocks_1_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_proj_in_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39788800)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_43_cast = conv(bias = down_blocks_1_attentions_0_proj_in_bias_to_fp16, dilations = var_1058, groups = var_984, pad = hidden_states_43_pad_0, pad_type = hidden_states_43_pad_type_0, strides = var_1056, weight = down_blocks_1_attentions_0_proj_in_weight_to_fp16, x = add_17_cast);
            tensor<int32, [4]> var_1063 = const()[name = tensor<string, []>("op_1063"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<fp16, [2, 640, 1, 1024]> inputs_13_cast = reshape(shape = var_1063, x = hidden_states_43_cast);
            tensor<int32, [1]> var_1073 = const()[name = tensor<string, []>("op_1073"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_13_cast = reduce_mean(axes = var_1073, keep_dims = var_979, x = inputs_13_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_13_cast = sub(x = inputs_13_cast, y = channels_mean_13_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_sq_13_cast = mul(x = zero_mean_13_cast, y = zero_mean_13_cast);
            tensor<int32, [1]> var_1077 = const()[name = tensor<string, []>("op_1077"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_1078_cast = reduce_mean(axes = var_1077, keep_dims = var_979, x = zero_mean_sq_13_cast);
            tensor<fp16, []> var_1079_to_fp16 = const()[name = tensor<string, []>("op_1079_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_1080_cast = add(x = var_1078_cast, y = var_1079_to_fp16);
            tensor<fp16, []> denom_13_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_13_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_13_cast = rsqrt(epsilon = denom_13_epsilon_0_to_fp16, x = var_1080_cast);
            tensor<fp16, [2, 640, 1, 1024]> out_13_cast = mul(x = zero_mean_13_cast, y = denom_13_cast);
            tensor<fp16, [1, 640, 1, 1]> var_1084_to_fp16 = const()[name = tensor<string, []>("op_1084_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39790144)))];
            tensor<fp16, [2, 640, 1, 1024]> var_1085_cast = add(x = out_13_cast, y = var_1084_to_fp16);
            tensor<fp16, [1, 640, 1, 1]> var_1087_to_fp16 = const()[name = tensor<string, []>("op_1087_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39791488)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_45_cast = mul(x = var_1085_cast, y = var_1087_to_fp16);
            tensor<int32, [2]> var_1094 = const()[name = tensor<string, []>("op_1094"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1096 = const()[name = tensor<string, []>("op_1096"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_9_pad_type_0 = const()[name = tensor<string, []>("q_9_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_9_pad_0 = const()[name = tensor<string, []>("q_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39792832)))];
            tensor<fp16, [2, 640, 1, 1024]> q_9_cast = conv(dilations = var_1096, groups = var_984, pad = q_9_pad_0, pad_type = q_9_pad_type_0, strides = var_1094, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_45_cast);
            tensor<int32, [2]> var_1100 = const()[name = tensor<string, []>("op_1100"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1102 = const()[name = tensor<string, []>("op_1102"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_17_pad_type_0 = const()[name = tensor<string, []>("k_17_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_17_pad_0 = const()[name = tensor<string, []>("k_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40612096)))];
            tensor<fp16, [2, 640, 1, 1024]> k_17_cast = conv(dilations = var_1102, groups = var_984, pad = k_17_pad_0, pad_type = k_17_pad_type_0, strides = var_1100, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_45_cast);
            tensor<int32, [2]> var_1106 = const()[name = tensor<string, []>("op_1106"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1108 = const()[name = tensor<string, []>("op_1108"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_9_pad_type_0 = const()[name = tensor<string, []>("v_9_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_9_pad_0 = const()[name = tensor<string, []>("v_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41431360)))];
            tensor<fp16, [2, 640, 1, 1024]> v_9_cast = conv(dilations = var_1108, groups = var_984, pad = v_9_pad_0, pad_type = v_9_pad_type_0, strides = var_1106, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_45_cast);
            tensor<int32, [4]> var_1112_begin_0 = const()[name = tensor<string, []>("op_1112_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1112_end_0 = const()[name = tensor<string, []>("op_1112_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_1112_end_mask_0 = const()[name = tensor<string, []>("op_1112_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1112_cast = slice_by_index(begin = var_1112_begin_0, end = var_1112_end_0, end_mask = var_1112_end_mask_0, x = q_9_cast);
            tensor<int32, [4]> var_1116_begin_0 = const()[name = tensor<string, []>("op_1116_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_1116_end_0 = const()[name = tensor<string, []>("op_1116_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_1116_end_mask_0 = const()[name = tensor<string, []>("op_1116_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1116_cast = slice_by_index(begin = var_1116_begin_0, end = var_1116_end_0, end_mask = var_1116_end_mask_0, x = q_9_cast);
            tensor<int32, [4]> var_1120_begin_0 = const()[name = tensor<string, []>("op_1120_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_1120_end_0 = const()[name = tensor<string, []>("op_1120_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_1120_end_mask_0 = const()[name = tensor<string, []>("op_1120_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1120_cast = slice_by_index(begin = var_1120_begin_0, end = var_1120_end_0, end_mask = var_1120_end_mask_0, x = q_9_cast);
            tensor<int32, [4]> var_1124_begin_0 = const()[name = tensor<string, []>("op_1124_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_1124_end_0 = const()[name = tensor<string, []>("op_1124_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_1124_end_mask_0 = const()[name = tensor<string, []>("op_1124_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1124_cast = slice_by_index(begin = var_1124_begin_0, end = var_1124_end_0, end_mask = var_1124_end_mask_0, x = q_9_cast);
            tensor<int32, [4]> var_1128_begin_0 = const()[name = tensor<string, []>("op_1128_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_1128_end_0 = const()[name = tensor<string, []>("op_1128_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_1128_end_mask_0 = const()[name = tensor<string, []>("op_1128_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1128_cast = slice_by_index(begin = var_1128_begin_0, end = var_1128_end_0, end_mask = var_1128_end_mask_0, x = q_9_cast);
            tensor<int32, [4]> var_1132_begin_0 = const()[name = tensor<string, []>("op_1132_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_1132_end_0 = const()[name = tensor<string, []>("op_1132_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_1132_end_mask_0 = const()[name = tensor<string, []>("op_1132_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1132_cast = slice_by_index(begin = var_1132_begin_0, end = var_1132_end_0, end_mask = var_1132_end_mask_0, x = q_9_cast);
            tensor<int32, [4]> var_1136_begin_0 = const()[name = tensor<string, []>("op_1136_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_1136_end_0 = const()[name = tensor<string, []>("op_1136_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_1136_end_mask_0 = const()[name = tensor<string, []>("op_1136_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1136_cast = slice_by_index(begin = var_1136_begin_0, end = var_1136_end_0, end_mask = var_1136_end_mask_0, x = q_9_cast);
            tensor<int32, [4]> var_1140_begin_0 = const()[name = tensor<string, []>("op_1140_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_1140_end_0 = const()[name = tensor<string, []>("op_1140_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_1140_end_mask_0 = const()[name = tensor<string, []>("op_1140_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1140_cast = slice_by_index(begin = var_1140_begin_0, end = var_1140_end_0, end_mask = var_1140_end_mask_0, x = q_9_cast);
            tensor<int32, [4]> var_1144_begin_0 = const()[name = tensor<string, []>("op_1144_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_1144_end_0 = const()[name = tensor<string, []>("op_1144_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_1144_end_mask_0 = const()[name = tensor<string, []>("op_1144_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1144_cast = slice_by_index(begin = var_1144_begin_0, end = var_1144_end_0, end_mask = var_1144_end_mask_0, x = q_9_cast);
            tensor<int32, [4]> var_1148_begin_0 = const()[name = tensor<string, []>("op_1148_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_1148_end_0 = const()[name = tensor<string, []>("op_1148_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_1148_end_mask_0 = const()[name = tensor<string, []>("op_1148_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1148_cast = slice_by_index(begin = var_1148_begin_0, end = var_1148_end_0, end_mask = var_1148_end_mask_0, x = q_9_cast);
            tensor<int32, [4]> k_19_perm_0 = const()[name = tensor<string, []>("k_19_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_1155_begin_0 = const()[name = tensor<string, []>("op_1155_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1155_end_0 = const()[name = tensor<string, []>("op_1155_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_1155_end_mask_0 = const()[name = tensor<string, []>("op_1155_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 640]> transpose_27 = transpose(perm = k_19_perm_0, x = k_17_cast);
            tensor<fp16, [2, 1024, 1, 64]> var_1155_cast = slice_by_index(begin = var_1155_begin_0, end = var_1155_end_0, end_mask = var_1155_end_mask_0, x = transpose_27);
            tensor<int32, [4]> var_1159_begin_0 = const()[name = tensor<string, []>("op_1159_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_1159_end_0 = const()[name = tensor<string, []>("op_1159_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_1159_end_mask_0 = const()[name = tensor<string, []>("op_1159_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_1159_cast = slice_by_index(begin = var_1159_begin_0, end = var_1159_end_0, end_mask = var_1159_end_mask_0, x = transpose_27);
            tensor<int32, [4]> var_1163_begin_0 = const()[name = tensor<string, []>("op_1163_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_1163_end_0 = const()[name = tensor<string, []>("op_1163_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_1163_end_mask_0 = const()[name = tensor<string, []>("op_1163_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_1163_cast = slice_by_index(begin = var_1163_begin_0, end = var_1163_end_0, end_mask = var_1163_end_mask_0, x = transpose_27);
            tensor<int32, [4]> var_1167_begin_0 = const()[name = tensor<string, []>("op_1167_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_1167_end_0 = const()[name = tensor<string, []>("op_1167_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_1167_end_mask_0 = const()[name = tensor<string, []>("op_1167_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_1167_cast = slice_by_index(begin = var_1167_begin_0, end = var_1167_end_0, end_mask = var_1167_end_mask_0, x = transpose_27);
            tensor<int32, [4]> var_1171_begin_0 = const()[name = tensor<string, []>("op_1171_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_1171_end_0 = const()[name = tensor<string, []>("op_1171_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_1171_end_mask_0 = const()[name = tensor<string, []>("op_1171_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_1171_cast = slice_by_index(begin = var_1171_begin_0, end = var_1171_end_0, end_mask = var_1171_end_mask_0, x = transpose_27);
            tensor<int32, [4]> var_1175_begin_0 = const()[name = tensor<string, []>("op_1175_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_1175_end_0 = const()[name = tensor<string, []>("op_1175_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_1175_end_mask_0 = const()[name = tensor<string, []>("op_1175_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_1175_cast = slice_by_index(begin = var_1175_begin_0, end = var_1175_end_0, end_mask = var_1175_end_mask_0, x = transpose_27);
            tensor<int32, [4]> var_1179_begin_0 = const()[name = tensor<string, []>("op_1179_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_1179_end_0 = const()[name = tensor<string, []>("op_1179_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_1179_end_mask_0 = const()[name = tensor<string, []>("op_1179_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_1179_cast = slice_by_index(begin = var_1179_begin_0, end = var_1179_end_0, end_mask = var_1179_end_mask_0, x = transpose_27);
            tensor<int32, [4]> var_1183_begin_0 = const()[name = tensor<string, []>("op_1183_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_1183_end_0 = const()[name = tensor<string, []>("op_1183_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_1183_end_mask_0 = const()[name = tensor<string, []>("op_1183_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_1183_cast = slice_by_index(begin = var_1183_begin_0, end = var_1183_end_0, end_mask = var_1183_end_mask_0, x = transpose_27);
            tensor<int32, [4]> var_1187_begin_0 = const()[name = tensor<string, []>("op_1187_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_1187_end_0 = const()[name = tensor<string, []>("op_1187_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_1187_end_mask_0 = const()[name = tensor<string, []>("op_1187_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_1187_cast = slice_by_index(begin = var_1187_begin_0, end = var_1187_end_0, end_mask = var_1187_end_mask_0, x = transpose_27);
            tensor<int32, [4]> var_1191_begin_0 = const()[name = tensor<string, []>("op_1191_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_1191_end_0 = const()[name = tensor<string, []>("op_1191_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_1191_end_mask_0 = const()[name = tensor<string, []>("op_1191_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_1191_cast = slice_by_index(begin = var_1191_begin_0, end = var_1191_end_0, end_mask = var_1191_end_mask_0, x = transpose_27);
            tensor<int32, [4]> var_1193_begin_0 = const()[name = tensor<string, []>("op_1193_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1193_end_0 = const()[name = tensor<string, []>("op_1193_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_1193_end_mask_0 = const()[name = tensor<string, []>("op_1193_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1193_cast = slice_by_index(begin = var_1193_begin_0, end = var_1193_end_0, end_mask = var_1193_end_mask_0, x = v_9_cast);
            tensor<int32, [4]> var_1197_begin_0 = const()[name = tensor<string, []>("op_1197_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_1197_end_0 = const()[name = tensor<string, []>("op_1197_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_1197_end_mask_0 = const()[name = tensor<string, []>("op_1197_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1197_cast = slice_by_index(begin = var_1197_begin_0, end = var_1197_end_0, end_mask = var_1197_end_mask_0, x = v_9_cast);
            tensor<int32, [4]> var_1201_begin_0 = const()[name = tensor<string, []>("op_1201_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_1201_end_0 = const()[name = tensor<string, []>("op_1201_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_1201_end_mask_0 = const()[name = tensor<string, []>("op_1201_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1201_cast = slice_by_index(begin = var_1201_begin_0, end = var_1201_end_0, end_mask = var_1201_end_mask_0, x = v_9_cast);
            tensor<int32, [4]> var_1205_begin_0 = const()[name = tensor<string, []>("op_1205_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_1205_end_0 = const()[name = tensor<string, []>("op_1205_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_1205_end_mask_0 = const()[name = tensor<string, []>("op_1205_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1205_cast = slice_by_index(begin = var_1205_begin_0, end = var_1205_end_0, end_mask = var_1205_end_mask_0, x = v_9_cast);
            tensor<int32, [4]> var_1209_begin_0 = const()[name = tensor<string, []>("op_1209_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_1209_end_0 = const()[name = tensor<string, []>("op_1209_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_1209_end_mask_0 = const()[name = tensor<string, []>("op_1209_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1209_cast = slice_by_index(begin = var_1209_begin_0, end = var_1209_end_0, end_mask = var_1209_end_mask_0, x = v_9_cast);
            tensor<int32, [4]> var_1213_begin_0 = const()[name = tensor<string, []>("op_1213_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_1213_end_0 = const()[name = tensor<string, []>("op_1213_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_1213_end_mask_0 = const()[name = tensor<string, []>("op_1213_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1213_cast = slice_by_index(begin = var_1213_begin_0, end = var_1213_end_0, end_mask = var_1213_end_mask_0, x = v_9_cast);
            tensor<int32, [4]> var_1217_begin_0 = const()[name = tensor<string, []>("op_1217_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_1217_end_0 = const()[name = tensor<string, []>("op_1217_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_1217_end_mask_0 = const()[name = tensor<string, []>("op_1217_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1217_cast = slice_by_index(begin = var_1217_begin_0, end = var_1217_end_0, end_mask = var_1217_end_mask_0, x = v_9_cast);
            tensor<int32, [4]> var_1221_begin_0 = const()[name = tensor<string, []>("op_1221_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_1221_end_0 = const()[name = tensor<string, []>("op_1221_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_1221_end_mask_0 = const()[name = tensor<string, []>("op_1221_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1221_cast = slice_by_index(begin = var_1221_begin_0, end = var_1221_end_0, end_mask = var_1221_end_mask_0, x = v_9_cast);
            tensor<int32, [4]> var_1225_begin_0 = const()[name = tensor<string, []>("op_1225_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_1225_end_0 = const()[name = tensor<string, []>("op_1225_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_1225_end_mask_0 = const()[name = tensor<string, []>("op_1225_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1225_cast = slice_by_index(begin = var_1225_begin_0, end = var_1225_end_0, end_mask = var_1225_end_mask_0, x = v_9_cast);
            tensor<int32, [4]> var_1229_begin_0 = const()[name = tensor<string, []>("op_1229_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_1229_end_0 = const()[name = tensor<string, []>("op_1229_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_1229_end_mask_0 = const()[name = tensor<string, []>("op_1229_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1229_cast = slice_by_index(begin = var_1229_begin_0, end = var_1229_end_0, end_mask = var_1229_end_mask_0, x = v_9_cast);
            tensor<string, []> var_1233_equation_0 = const()[name = tensor<string, []>("op_1233_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_1233_cast = einsum(equation = var_1233_equation_0, values = (var_1155_cast, var_1112_cast));
            tensor<fp16, []> var_1234_to_fp16 = const()[name = tensor<string, []>("op_1234_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_41_cast = mul(x = var_1233_cast, y = var_1234_to_fp16);
            tensor<string, []> var_1237_equation_0 = const()[name = tensor<string, []>("op_1237_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_1237_cast = einsum(equation = var_1237_equation_0, values = (var_1159_cast, var_1116_cast));
            tensor<fp16, []> var_1238_to_fp16 = const()[name = tensor<string, []>("op_1238_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_43_cast = mul(x = var_1237_cast, y = var_1238_to_fp16);
            tensor<string, []> var_1241_equation_0 = const()[name = tensor<string, []>("op_1241_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_1241_cast = einsum(equation = var_1241_equation_0, values = (var_1163_cast, var_1120_cast));
            tensor<fp16, []> var_1242_to_fp16 = const()[name = tensor<string, []>("op_1242_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_45_cast = mul(x = var_1241_cast, y = var_1242_to_fp16);
            tensor<string, []> var_1245_equation_0 = const()[name = tensor<string, []>("op_1245_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_1245_cast = einsum(equation = var_1245_equation_0, values = (var_1167_cast, var_1124_cast));
            tensor<fp16, []> var_1246_to_fp16 = const()[name = tensor<string, []>("op_1246_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_47_cast = mul(x = var_1245_cast, y = var_1246_to_fp16);
            tensor<string, []> var_1249_equation_0 = const()[name = tensor<string, []>("op_1249_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_1249_cast = einsum(equation = var_1249_equation_0, values = (var_1171_cast, var_1128_cast));
            tensor<fp16, []> var_1250_to_fp16 = const()[name = tensor<string, []>("op_1250_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_49_cast = mul(x = var_1249_cast, y = var_1250_to_fp16);
            tensor<string, []> var_1253_equation_0 = const()[name = tensor<string, []>("op_1253_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_1253_cast = einsum(equation = var_1253_equation_0, values = (var_1175_cast, var_1132_cast));
            tensor<fp16, []> var_1254_to_fp16 = const()[name = tensor<string, []>("op_1254_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_51_cast = mul(x = var_1253_cast, y = var_1254_to_fp16);
            tensor<string, []> var_1257_equation_0 = const()[name = tensor<string, []>("op_1257_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_1257_cast = einsum(equation = var_1257_equation_0, values = (var_1179_cast, var_1136_cast));
            tensor<fp16, []> var_1258_to_fp16 = const()[name = tensor<string, []>("op_1258_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_53_cast = mul(x = var_1257_cast, y = var_1258_to_fp16);
            tensor<string, []> var_1261_equation_0 = const()[name = tensor<string, []>("op_1261_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_1261_cast = einsum(equation = var_1261_equation_0, values = (var_1183_cast, var_1140_cast));
            tensor<fp16, []> var_1262_to_fp16 = const()[name = tensor<string, []>("op_1262_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_55_cast = mul(x = var_1261_cast, y = var_1262_to_fp16);
            tensor<string, []> var_1265_equation_0 = const()[name = tensor<string, []>("op_1265_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_1265_cast = einsum(equation = var_1265_equation_0, values = (var_1187_cast, var_1144_cast));
            tensor<fp16, []> var_1266_to_fp16 = const()[name = tensor<string, []>("op_1266_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_57_cast = mul(x = var_1265_cast, y = var_1266_to_fp16);
            tensor<string, []> var_1269_equation_0 = const()[name = tensor<string, []>("op_1269_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_1269_cast = einsum(equation = var_1269_equation_0, values = (var_1191_cast, var_1148_cast));
            tensor<fp16, []> var_1270_to_fp16 = const()[name = tensor<string, []>("op_1270_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_59_cast = mul(x = var_1269_cast, y = var_1270_to_fp16);
            tensor<fp16, [2, 1024, 1, 1024]> var_1272_cast = softmax(axis = var_984, x = aw_41_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_1273_cast = softmax(axis = var_984, x = aw_43_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_1274_cast = softmax(axis = var_984, x = aw_45_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_1275_cast = softmax(axis = var_984, x = aw_47_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_1276_cast = softmax(axis = var_984, x = aw_49_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_1277_cast = softmax(axis = var_984, x = aw_51_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_1278_cast = softmax(axis = var_984, x = aw_53_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_1279_cast = softmax(axis = var_984, x = aw_55_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_1280_cast = softmax(axis = var_984, x = aw_57_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_1281_cast = softmax(axis = var_984, x = aw_59_cast);
            tensor<string, []> var_1283_equation_0 = const()[name = tensor<string, []>("op_1283_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1283_cast = einsum(equation = var_1283_equation_0, values = (var_1193_cast, var_1272_cast));
            tensor<string, []> var_1285_equation_0 = const()[name = tensor<string, []>("op_1285_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1285_cast = einsum(equation = var_1285_equation_0, values = (var_1197_cast, var_1273_cast));
            tensor<string, []> var_1287_equation_0 = const()[name = tensor<string, []>("op_1287_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1287_cast = einsum(equation = var_1287_equation_0, values = (var_1201_cast, var_1274_cast));
            tensor<string, []> var_1289_equation_0 = const()[name = tensor<string, []>("op_1289_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1289_cast = einsum(equation = var_1289_equation_0, values = (var_1205_cast, var_1275_cast));
            tensor<string, []> var_1291_equation_0 = const()[name = tensor<string, []>("op_1291_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1291_cast = einsum(equation = var_1291_equation_0, values = (var_1209_cast, var_1276_cast));
            tensor<string, []> var_1293_equation_0 = const()[name = tensor<string, []>("op_1293_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1293_cast = einsum(equation = var_1293_equation_0, values = (var_1213_cast, var_1277_cast));
            tensor<string, []> var_1295_equation_0 = const()[name = tensor<string, []>("op_1295_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1295_cast = einsum(equation = var_1295_equation_0, values = (var_1217_cast, var_1278_cast));
            tensor<string, []> var_1297_equation_0 = const()[name = tensor<string, []>("op_1297_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1297_cast = einsum(equation = var_1297_equation_0, values = (var_1221_cast, var_1279_cast));
            tensor<string, []> var_1299_equation_0 = const()[name = tensor<string, []>("op_1299_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1299_cast = einsum(equation = var_1299_equation_0, values = (var_1225_cast, var_1280_cast));
            tensor<string, []> var_1301_equation_0 = const()[name = tensor<string, []>("op_1301_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1301_cast = einsum(equation = var_1301_equation_0, values = (var_1229_cast, var_1281_cast));
            tensor<bool, []> input_79_interleave_0 = const()[name = tensor<string, []>("input_79_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 1024]> input_79_cast = concat(axis = var_984, interleave = input_79_interleave_0, values = (var_1283_cast, var_1285_cast, var_1287_cast, var_1289_cast, var_1291_cast, var_1293_cast, var_1295_cast, var_1297_cast, var_1299_cast, var_1301_cast));
            tensor<int32, [2]> var_1307 = const()[name = tensor<string, []>("op_1307"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1309 = const()[name = tensor<string, []>("op_1309"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_1311_pad_type_0 = const()[name = tensor<string, []>("op_1311_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_1311_pad_0 = const()[name = tensor<string, []>("op_1311_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42250624)))];
            tensor<fp16, [640]> down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43069888)))];
            tensor<fp16, [2, 640, 1, 1024]> var_1311_cast = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_1309, groups = var_984, pad = var_1311_pad_0, pad_type = var_1311_pad_type_0, strides = var_1307, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_79_cast);
            tensor<fp16, [2, 640, 1, 1024]> inputs_15_cast = add(x = var_1311_cast, y = inputs_13_cast);
            tensor<int32, [1]> var_1315 = const()[name = tensor<string, []>("op_1315"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_15_cast = reduce_mean(axes = var_1315, keep_dims = var_979, x = inputs_15_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_15_cast = sub(x = inputs_15_cast, y = channels_mean_15_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_sq_15_cast = mul(x = zero_mean_15_cast, y = zero_mean_15_cast);
            tensor<int32, [1]> var_1319 = const()[name = tensor<string, []>("op_1319"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_1320_cast = reduce_mean(axes = var_1319, keep_dims = var_979, x = zero_mean_sq_15_cast);
            tensor<fp16, []> var_1321_to_fp16 = const()[name = tensor<string, []>("op_1321_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_1322_cast = add(x = var_1320_cast, y = var_1321_to_fp16);
            tensor<fp16, []> denom_15_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_15_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_15_cast = rsqrt(epsilon = denom_15_epsilon_0_to_fp16, x = var_1322_cast);
            tensor<fp16, [2, 640, 1, 1024]> out_15_cast = mul(x = zero_mean_15_cast, y = denom_15_cast);
            tensor<fp16, [1, 640, 1, 1]> var_1326_to_fp16 = const()[name = tensor<string, []>("op_1326_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43071232)))];
            tensor<fp16, [2, 640, 1, 1024]> var_1327_cast = add(x = out_15_cast, y = var_1326_to_fp16);
            tensor<fp16, [1, 640, 1, 1]> var_1329_to_fp16 = const()[name = tensor<string, []>("op_1329_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43072576)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_47_cast = mul(x = var_1327_cast, y = var_1329_to_fp16);
            tensor<int32, [2]> var_1336 = const()[name = tensor<string, []>("op_1336"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1338 = const()[name = tensor<string, []>("op_1338"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_11_pad_type_0 = const()[name = tensor<string, []>("q_11_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_11_pad_0 = const()[name = tensor<string, []>("q_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43073920)))];
            tensor<fp16, [2, 640, 1, 1024]> q_11_cast = conv(dilations = var_1338, groups = var_984, pad = q_11_pad_0, pad_type = q_11_pad_type_0, strides = var_1336, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_47_cast);
            tensor<int32, [2]> var_1342 = const()[name = tensor<string, []>("op_1342"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1344 = const()[name = tensor<string, []>("op_1344"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_21_pad_type_0 = const()[name = tensor<string, []>("k_21_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_21_pad_0 = const()[name = tensor<string, []>("k_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1024, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [640, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43893184)))];
            tensor<fp16, [2, 640, 1, 77]> k_21_cast = conv(dilations = var_1344, groups = var_984, pad = k_21_pad_0, pad_type = k_21_pad_type_0, strides = var_1342, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [2]> var_1348 = const()[name = tensor<string, []>("op_1348"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1350 = const()[name = tensor<string, []>("op_1350"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_11_pad_type_0 = const()[name = tensor<string, []>("v_11_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_11_pad_0 = const()[name = tensor<string, []>("v_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1024, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [640, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(45203968)))];
            tensor<fp16, [2, 640, 1, 77]> v_11_cast = conv(dilations = var_1350, groups = var_984, pad = v_11_pad_0, pad_type = v_11_pad_type_0, strides = var_1348, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [4]> var_1354_begin_0 = const()[name = tensor<string, []>("op_1354_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1354_end_0 = const()[name = tensor<string, []>("op_1354_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_1354_end_mask_0 = const()[name = tensor<string, []>("op_1354_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1354_cast = slice_by_index(begin = var_1354_begin_0, end = var_1354_end_0, end_mask = var_1354_end_mask_0, x = q_11_cast);
            tensor<int32, [4]> var_1358_begin_0 = const()[name = tensor<string, []>("op_1358_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_1358_end_0 = const()[name = tensor<string, []>("op_1358_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_1358_end_mask_0 = const()[name = tensor<string, []>("op_1358_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1358_cast = slice_by_index(begin = var_1358_begin_0, end = var_1358_end_0, end_mask = var_1358_end_mask_0, x = q_11_cast);
            tensor<int32, [4]> var_1362_begin_0 = const()[name = tensor<string, []>("op_1362_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_1362_end_0 = const()[name = tensor<string, []>("op_1362_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_1362_end_mask_0 = const()[name = tensor<string, []>("op_1362_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1362_cast = slice_by_index(begin = var_1362_begin_0, end = var_1362_end_0, end_mask = var_1362_end_mask_0, x = q_11_cast);
            tensor<int32, [4]> var_1366_begin_0 = const()[name = tensor<string, []>("op_1366_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_1366_end_0 = const()[name = tensor<string, []>("op_1366_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_1366_end_mask_0 = const()[name = tensor<string, []>("op_1366_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1366_cast = slice_by_index(begin = var_1366_begin_0, end = var_1366_end_0, end_mask = var_1366_end_mask_0, x = q_11_cast);
            tensor<int32, [4]> var_1370_begin_0 = const()[name = tensor<string, []>("op_1370_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_1370_end_0 = const()[name = tensor<string, []>("op_1370_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_1370_end_mask_0 = const()[name = tensor<string, []>("op_1370_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1370_cast = slice_by_index(begin = var_1370_begin_0, end = var_1370_end_0, end_mask = var_1370_end_mask_0, x = q_11_cast);
            tensor<int32, [4]> var_1374_begin_0 = const()[name = tensor<string, []>("op_1374_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_1374_end_0 = const()[name = tensor<string, []>("op_1374_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_1374_end_mask_0 = const()[name = tensor<string, []>("op_1374_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1374_cast = slice_by_index(begin = var_1374_begin_0, end = var_1374_end_0, end_mask = var_1374_end_mask_0, x = q_11_cast);
            tensor<int32, [4]> var_1378_begin_0 = const()[name = tensor<string, []>("op_1378_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_1378_end_0 = const()[name = tensor<string, []>("op_1378_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_1378_end_mask_0 = const()[name = tensor<string, []>("op_1378_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1378_cast = slice_by_index(begin = var_1378_begin_0, end = var_1378_end_0, end_mask = var_1378_end_mask_0, x = q_11_cast);
            tensor<int32, [4]> var_1382_begin_0 = const()[name = tensor<string, []>("op_1382_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_1382_end_0 = const()[name = tensor<string, []>("op_1382_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_1382_end_mask_0 = const()[name = tensor<string, []>("op_1382_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1382_cast = slice_by_index(begin = var_1382_begin_0, end = var_1382_end_0, end_mask = var_1382_end_mask_0, x = q_11_cast);
            tensor<int32, [4]> var_1386_begin_0 = const()[name = tensor<string, []>("op_1386_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_1386_end_0 = const()[name = tensor<string, []>("op_1386_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_1386_end_mask_0 = const()[name = tensor<string, []>("op_1386_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1386_cast = slice_by_index(begin = var_1386_begin_0, end = var_1386_end_0, end_mask = var_1386_end_mask_0, x = q_11_cast);
            tensor<int32, [4]> var_1390_begin_0 = const()[name = tensor<string, []>("op_1390_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_1390_end_0 = const()[name = tensor<string, []>("op_1390_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_1390_end_mask_0 = const()[name = tensor<string, []>("op_1390_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1390_cast = slice_by_index(begin = var_1390_begin_0, end = var_1390_end_0, end_mask = var_1390_end_mask_0, x = q_11_cast);
            tensor<int32, [4]> k_23_perm_0 = const()[name = tensor<string, []>("k_23_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_1397_begin_0 = const()[name = tensor<string, []>("op_1397_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1397_end_0 = const()[name = tensor<string, []>("op_1397_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_1397_end_mask_0 = const()[name = tensor<string, []>("op_1397_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 640]> transpose_26 = transpose(perm = k_23_perm_0, x = k_21_cast);
            tensor<fp16, [2, 77, 1, 64]> var_1397_cast = slice_by_index(begin = var_1397_begin_0, end = var_1397_end_0, end_mask = var_1397_end_mask_0, x = transpose_26);
            tensor<int32, [4]> var_1401_begin_0 = const()[name = tensor<string, []>("op_1401_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_1401_end_0 = const()[name = tensor<string, []>("op_1401_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_1401_end_mask_0 = const()[name = tensor<string, []>("op_1401_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_1401_cast = slice_by_index(begin = var_1401_begin_0, end = var_1401_end_0, end_mask = var_1401_end_mask_0, x = transpose_26);
            tensor<int32, [4]> var_1405_begin_0 = const()[name = tensor<string, []>("op_1405_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_1405_end_0 = const()[name = tensor<string, []>("op_1405_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_1405_end_mask_0 = const()[name = tensor<string, []>("op_1405_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_1405_cast = slice_by_index(begin = var_1405_begin_0, end = var_1405_end_0, end_mask = var_1405_end_mask_0, x = transpose_26);
            tensor<int32, [4]> var_1409_begin_0 = const()[name = tensor<string, []>("op_1409_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_1409_end_0 = const()[name = tensor<string, []>("op_1409_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_1409_end_mask_0 = const()[name = tensor<string, []>("op_1409_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_1409_cast = slice_by_index(begin = var_1409_begin_0, end = var_1409_end_0, end_mask = var_1409_end_mask_0, x = transpose_26);
            tensor<int32, [4]> var_1413_begin_0 = const()[name = tensor<string, []>("op_1413_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_1413_end_0 = const()[name = tensor<string, []>("op_1413_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_1413_end_mask_0 = const()[name = tensor<string, []>("op_1413_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_1413_cast = slice_by_index(begin = var_1413_begin_0, end = var_1413_end_0, end_mask = var_1413_end_mask_0, x = transpose_26);
            tensor<int32, [4]> var_1417_begin_0 = const()[name = tensor<string, []>("op_1417_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_1417_end_0 = const()[name = tensor<string, []>("op_1417_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_1417_end_mask_0 = const()[name = tensor<string, []>("op_1417_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_1417_cast = slice_by_index(begin = var_1417_begin_0, end = var_1417_end_0, end_mask = var_1417_end_mask_0, x = transpose_26);
            tensor<int32, [4]> var_1421_begin_0 = const()[name = tensor<string, []>("op_1421_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_1421_end_0 = const()[name = tensor<string, []>("op_1421_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_1421_end_mask_0 = const()[name = tensor<string, []>("op_1421_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_1421_cast = slice_by_index(begin = var_1421_begin_0, end = var_1421_end_0, end_mask = var_1421_end_mask_0, x = transpose_26);
            tensor<int32, [4]> var_1425_begin_0 = const()[name = tensor<string, []>("op_1425_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_1425_end_0 = const()[name = tensor<string, []>("op_1425_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_1425_end_mask_0 = const()[name = tensor<string, []>("op_1425_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_1425_cast = slice_by_index(begin = var_1425_begin_0, end = var_1425_end_0, end_mask = var_1425_end_mask_0, x = transpose_26);
            tensor<int32, [4]> var_1429_begin_0 = const()[name = tensor<string, []>("op_1429_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_1429_end_0 = const()[name = tensor<string, []>("op_1429_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_1429_end_mask_0 = const()[name = tensor<string, []>("op_1429_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_1429_cast = slice_by_index(begin = var_1429_begin_0, end = var_1429_end_0, end_mask = var_1429_end_mask_0, x = transpose_26);
            tensor<int32, [4]> var_1433_begin_0 = const()[name = tensor<string, []>("op_1433_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_1433_end_0 = const()[name = tensor<string, []>("op_1433_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_1433_end_mask_0 = const()[name = tensor<string, []>("op_1433_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_1433_cast = slice_by_index(begin = var_1433_begin_0, end = var_1433_end_0, end_mask = var_1433_end_mask_0, x = transpose_26);
            tensor<int32, [4]> var_1435_begin_0 = const()[name = tensor<string, []>("op_1435_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1435_end_0 = const()[name = tensor<string, []>("op_1435_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_1435_end_mask_0 = const()[name = tensor<string, []>("op_1435_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1435_cast = slice_by_index(begin = var_1435_begin_0, end = var_1435_end_0, end_mask = var_1435_end_mask_0, x = v_11_cast);
            tensor<int32, [4]> var_1439_begin_0 = const()[name = tensor<string, []>("op_1439_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_1439_end_0 = const()[name = tensor<string, []>("op_1439_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_1439_end_mask_0 = const()[name = tensor<string, []>("op_1439_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1439_cast = slice_by_index(begin = var_1439_begin_0, end = var_1439_end_0, end_mask = var_1439_end_mask_0, x = v_11_cast);
            tensor<int32, [4]> var_1443_begin_0 = const()[name = tensor<string, []>("op_1443_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_1443_end_0 = const()[name = tensor<string, []>("op_1443_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_1443_end_mask_0 = const()[name = tensor<string, []>("op_1443_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1443_cast = slice_by_index(begin = var_1443_begin_0, end = var_1443_end_0, end_mask = var_1443_end_mask_0, x = v_11_cast);
            tensor<int32, [4]> var_1447_begin_0 = const()[name = tensor<string, []>("op_1447_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_1447_end_0 = const()[name = tensor<string, []>("op_1447_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_1447_end_mask_0 = const()[name = tensor<string, []>("op_1447_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1447_cast = slice_by_index(begin = var_1447_begin_0, end = var_1447_end_0, end_mask = var_1447_end_mask_0, x = v_11_cast);
            tensor<int32, [4]> var_1451_begin_0 = const()[name = tensor<string, []>("op_1451_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_1451_end_0 = const()[name = tensor<string, []>("op_1451_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_1451_end_mask_0 = const()[name = tensor<string, []>("op_1451_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1451_cast = slice_by_index(begin = var_1451_begin_0, end = var_1451_end_0, end_mask = var_1451_end_mask_0, x = v_11_cast);
            tensor<int32, [4]> var_1455_begin_0 = const()[name = tensor<string, []>("op_1455_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_1455_end_0 = const()[name = tensor<string, []>("op_1455_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_1455_end_mask_0 = const()[name = tensor<string, []>("op_1455_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1455_cast = slice_by_index(begin = var_1455_begin_0, end = var_1455_end_0, end_mask = var_1455_end_mask_0, x = v_11_cast);
            tensor<int32, [4]> var_1459_begin_0 = const()[name = tensor<string, []>("op_1459_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_1459_end_0 = const()[name = tensor<string, []>("op_1459_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_1459_end_mask_0 = const()[name = tensor<string, []>("op_1459_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1459_cast = slice_by_index(begin = var_1459_begin_0, end = var_1459_end_0, end_mask = var_1459_end_mask_0, x = v_11_cast);
            tensor<int32, [4]> var_1463_begin_0 = const()[name = tensor<string, []>("op_1463_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_1463_end_0 = const()[name = tensor<string, []>("op_1463_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_1463_end_mask_0 = const()[name = tensor<string, []>("op_1463_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1463_cast = slice_by_index(begin = var_1463_begin_0, end = var_1463_end_0, end_mask = var_1463_end_mask_0, x = v_11_cast);
            tensor<int32, [4]> var_1467_begin_0 = const()[name = tensor<string, []>("op_1467_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_1467_end_0 = const()[name = tensor<string, []>("op_1467_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_1467_end_mask_0 = const()[name = tensor<string, []>("op_1467_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1467_cast = slice_by_index(begin = var_1467_begin_0, end = var_1467_end_0, end_mask = var_1467_end_mask_0, x = v_11_cast);
            tensor<int32, [4]> var_1471_begin_0 = const()[name = tensor<string, []>("op_1471_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_1471_end_0 = const()[name = tensor<string, []>("op_1471_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_1471_end_mask_0 = const()[name = tensor<string, []>("op_1471_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1471_cast = slice_by_index(begin = var_1471_begin_0, end = var_1471_end_0, end_mask = var_1471_end_mask_0, x = v_11_cast);
            tensor<string, []> var_1475_equation_0 = const()[name = tensor<string, []>("op_1475_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_1475_cast = einsum(equation = var_1475_equation_0, values = (var_1397_cast, var_1354_cast));
            tensor<fp16, []> var_1476_to_fp16 = const()[name = tensor<string, []>("op_1476_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_61_cast = mul(x = var_1475_cast, y = var_1476_to_fp16);
            tensor<string, []> var_1479_equation_0 = const()[name = tensor<string, []>("op_1479_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_1479_cast = einsum(equation = var_1479_equation_0, values = (var_1401_cast, var_1358_cast));
            tensor<fp16, []> var_1480_to_fp16 = const()[name = tensor<string, []>("op_1480_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_63_cast = mul(x = var_1479_cast, y = var_1480_to_fp16);
            tensor<string, []> var_1483_equation_0 = const()[name = tensor<string, []>("op_1483_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_1483_cast = einsum(equation = var_1483_equation_0, values = (var_1405_cast, var_1362_cast));
            tensor<fp16, []> var_1484_to_fp16 = const()[name = tensor<string, []>("op_1484_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_65_cast = mul(x = var_1483_cast, y = var_1484_to_fp16);
            tensor<string, []> var_1487_equation_0 = const()[name = tensor<string, []>("op_1487_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_1487_cast = einsum(equation = var_1487_equation_0, values = (var_1409_cast, var_1366_cast));
            tensor<fp16, []> var_1488_to_fp16 = const()[name = tensor<string, []>("op_1488_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_67_cast = mul(x = var_1487_cast, y = var_1488_to_fp16);
            tensor<string, []> var_1491_equation_0 = const()[name = tensor<string, []>("op_1491_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_1491_cast = einsum(equation = var_1491_equation_0, values = (var_1413_cast, var_1370_cast));
            tensor<fp16, []> var_1492_to_fp16 = const()[name = tensor<string, []>("op_1492_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_69_cast = mul(x = var_1491_cast, y = var_1492_to_fp16);
            tensor<string, []> var_1495_equation_0 = const()[name = tensor<string, []>("op_1495_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_1495_cast = einsum(equation = var_1495_equation_0, values = (var_1417_cast, var_1374_cast));
            tensor<fp16, []> var_1496_to_fp16 = const()[name = tensor<string, []>("op_1496_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_71_cast = mul(x = var_1495_cast, y = var_1496_to_fp16);
            tensor<string, []> var_1499_equation_0 = const()[name = tensor<string, []>("op_1499_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_1499_cast = einsum(equation = var_1499_equation_0, values = (var_1421_cast, var_1378_cast));
            tensor<fp16, []> var_1500_to_fp16 = const()[name = tensor<string, []>("op_1500_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_73_cast = mul(x = var_1499_cast, y = var_1500_to_fp16);
            tensor<string, []> var_1503_equation_0 = const()[name = tensor<string, []>("op_1503_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_1503_cast = einsum(equation = var_1503_equation_0, values = (var_1425_cast, var_1382_cast));
            tensor<fp16, []> var_1504_to_fp16 = const()[name = tensor<string, []>("op_1504_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_75_cast = mul(x = var_1503_cast, y = var_1504_to_fp16);
            tensor<string, []> var_1507_equation_0 = const()[name = tensor<string, []>("op_1507_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_1507_cast = einsum(equation = var_1507_equation_0, values = (var_1429_cast, var_1386_cast));
            tensor<fp16, []> var_1508_to_fp16 = const()[name = tensor<string, []>("op_1508_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_77_cast = mul(x = var_1507_cast, y = var_1508_to_fp16);
            tensor<string, []> var_1511_equation_0 = const()[name = tensor<string, []>("op_1511_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_1511_cast = einsum(equation = var_1511_equation_0, values = (var_1433_cast, var_1390_cast));
            tensor<fp16, []> var_1512_to_fp16 = const()[name = tensor<string, []>("op_1512_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_79_cast = mul(x = var_1511_cast, y = var_1512_to_fp16);
            tensor<fp16, [2, 77, 1, 1024]> var_1514_cast = softmax(axis = var_984, x = aw_61_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_1515_cast = softmax(axis = var_984, x = aw_63_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_1516_cast = softmax(axis = var_984, x = aw_65_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_1517_cast = softmax(axis = var_984, x = aw_67_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_1518_cast = softmax(axis = var_984, x = aw_69_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_1519_cast = softmax(axis = var_984, x = aw_71_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_1520_cast = softmax(axis = var_984, x = aw_73_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_1521_cast = softmax(axis = var_984, x = aw_75_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_1522_cast = softmax(axis = var_984, x = aw_77_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_1523_cast = softmax(axis = var_984, x = aw_79_cast);
            tensor<string, []> var_1525_equation_0 = const()[name = tensor<string, []>("op_1525_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1525_cast = einsum(equation = var_1525_equation_0, values = (var_1435_cast, var_1514_cast));
            tensor<string, []> var_1527_equation_0 = const()[name = tensor<string, []>("op_1527_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1527_cast = einsum(equation = var_1527_equation_0, values = (var_1439_cast, var_1515_cast));
            tensor<string, []> var_1529_equation_0 = const()[name = tensor<string, []>("op_1529_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1529_cast = einsum(equation = var_1529_equation_0, values = (var_1443_cast, var_1516_cast));
            tensor<string, []> var_1531_equation_0 = const()[name = tensor<string, []>("op_1531_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1531_cast = einsum(equation = var_1531_equation_0, values = (var_1447_cast, var_1517_cast));
            tensor<string, []> var_1533_equation_0 = const()[name = tensor<string, []>("op_1533_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1533_cast = einsum(equation = var_1533_equation_0, values = (var_1451_cast, var_1518_cast));
            tensor<string, []> var_1535_equation_0 = const()[name = tensor<string, []>("op_1535_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1535_cast = einsum(equation = var_1535_equation_0, values = (var_1455_cast, var_1519_cast));
            tensor<string, []> var_1537_equation_0 = const()[name = tensor<string, []>("op_1537_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1537_cast = einsum(equation = var_1537_equation_0, values = (var_1459_cast, var_1520_cast));
            tensor<string, []> var_1539_equation_0 = const()[name = tensor<string, []>("op_1539_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1539_cast = einsum(equation = var_1539_equation_0, values = (var_1463_cast, var_1521_cast));
            tensor<string, []> var_1541_equation_0 = const()[name = tensor<string, []>("op_1541_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1541_cast = einsum(equation = var_1541_equation_0, values = (var_1467_cast, var_1522_cast));
            tensor<string, []> var_1543_equation_0 = const()[name = tensor<string, []>("op_1543_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1543_cast = einsum(equation = var_1543_equation_0, values = (var_1471_cast, var_1523_cast));
            tensor<bool, []> input_81_interleave_0 = const()[name = tensor<string, []>("input_81_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 1024]> input_81_cast = concat(axis = var_984, interleave = input_81_interleave_0, values = (var_1525_cast, var_1527_cast, var_1529_cast, var_1531_cast, var_1533_cast, var_1535_cast, var_1537_cast, var_1539_cast, var_1541_cast, var_1543_cast));
            tensor<int32, [2]> var_1549 = const()[name = tensor<string, []>("op_1549"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1551 = const()[name = tensor<string, []>("op_1551"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_1553_pad_type_0 = const()[name = tensor<string, []>("op_1553_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_1553_pad_0 = const()[name = tensor<string, []>("op_1553_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46514752)))];
            tensor<fp16, [640]> down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(47334016)))];
            tensor<fp16, [2, 640, 1, 1024]> var_1553_cast = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_1551, groups = var_984, pad = var_1553_pad_0, pad_type = var_1553_pad_type_0, strides = var_1549, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_81_cast);
            tensor<fp16, [2, 640, 1, 1024]> inputs_17_cast = add(x = var_1553_cast, y = inputs_15_cast);
            tensor<int32, [1]> var_1557 = const()[name = tensor<string, []>("op_1557"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_17_cast = reduce_mean(axes = var_1557, keep_dims = var_979, x = inputs_17_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_17_cast = sub(x = inputs_17_cast, y = channels_mean_17_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_sq_17_cast = mul(x = zero_mean_17_cast, y = zero_mean_17_cast);
            tensor<int32, [1]> var_1561 = const()[name = tensor<string, []>("op_1561"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_1562_cast = reduce_mean(axes = var_1561, keep_dims = var_979, x = zero_mean_sq_17_cast);
            tensor<fp16, []> var_1563_to_fp16 = const()[name = tensor<string, []>("op_1563_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_1564_cast = add(x = var_1562_cast, y = var_1563_to_fp16);
            tensor<fp16, []> denom_17_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_17_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_17_cast = rsqrt(epsilon = denom_17_epsilon_0_to_fp16, x = var_1564_cast);
            tensor<fp16, [2, 640, 1, 1024]> out_17_cast = mul(x = zero_mean_17_cast, y = denom_17_cast);
            tensor<fp16, [1, 640, 1, 1]> var_1568_to_fp16 = const()[name = tensor<string, []>("op_1568_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(47335360)))];
            tensor<fp16, [2, 640, 1, 1024]> var_1569_cast = add(x = out_17_cast, y = var_1568_to_fp16);
            tensor<fp16, [1, 640, 1, 1]> var_1571_to_fp16 = const()[name = tensor<string, []>("op_1571_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(47336704)))];
            tensor<fp16, [2, 640, 1, 1024]> input_83_cast = mul(x = var_1569_cast, y = var_1571_to_fp16);
            tensor<int32, [2]> var_1579 = const()[name = tensor<string, []>("op_1579"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1581 = const()[name = tensor<string, []>("op_1581"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_1583_pad_type_0 = const()[name = tensor<string, []>("op_1583_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_1583_pad_0 = const()[name = tensor<string, []>("op_1583_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [5120, 640, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [5120, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(47338048)))];
            tensor<fp16, [5120]> down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53891712)))];
            tensor<fp16, [2, 5120, 1, 1024]> var_1583_cast = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_1581, groups = var_984, pad = var_1583_pad_0, pad_type = var_1583_pad_type_0, strides = var_1579, weight = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_83_cast);
            tensor<int32, [2]> var_1584_split_sizes_0 = const()[name = tensor<string, []>("op_1584_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_1584_axis_0 = const()[name = tensor<string, []>("op_1584_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 1024]> var_1584_cast_0, tensor<fp16, [2, 2560, 1, 1024]> var_1584_cast_1 = split(axis = var_1584_axis_0, split_sizes = var_1584_split_sizes_0, x = var_1583_cast);
            tensor<string, []> var_1586_mode_0 = const()[name = tensor<string, []>("op_1586_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 1024]> var_1586_cast = gelu(mode = var_1586_mode_0, x = var_1584_cast_1);
            tensor<fp16, [2, 2560, 1, 1024]> input_85_cast = mul(x = var_1584_cast_0, y = var_1586_cast);
            tensor<int32, [2]> var_1590 = const()[name = tensor<string, []>("op_1590"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1592 = const()[name = tensor<string, []>("op_1592"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_1594_pad_type_0 = const()[name = tensor<string, []>("op_1594_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_1594_pad_0 = const()[name = tensor<string, []>("op_1594_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2560, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [640, 2560, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53902016)))];
            tensor<fp16, [640]> down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57178880)))];
            tensor<fp16, [2, 640, 1, 1024]> var_1594_cast = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_1592, groups = var_984, pad = var_1594_pad_0, pad_type = var_1594_pad_type_0, strides = var_1590, weight = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_85_cast);
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_51_cast = add(x = var_1594_cast, y = inputs_17_cast);
            tensor<int32, [4]> var_1596 = const()[name = tensor<string, []>("op_1596"), val = tensor<int32, [4]>([2, 640, 32, 32])];
            tensor<fp16, [2, 640, 32, 32]> input_87_cast = reshape(shape = var_1596, x = hidden_states_51_cast);
            tensor<int32, [2]> var_1600 = const()[name = tensor<string, []>("op_1600"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1602 = const()[name = tensor<string, []>("op_1602"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_53_pad_type_0 = const()[name = tensor<string, []>("hidden_states_53_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_53_pad_0 = const()[name = tensor<string, []>("hidden_states_53_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_0_proj_out_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_proj_out_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57180224)))];
            tensor<fp16, [640]> down_blocks_1_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_0_proj_out_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57999488)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_53_cast = conv(bias = down_blocks_1_attentions_0_proj_out_bias_to_fp16, dilations = var_1602, groups = var_984, pad = hidden_states_53_pad_0, pad_type = hidden_states_53_pad_type_0, strides = var_1600, weight = down_blocks_1_attentions_0_proj_out_weight_to_fp16, x = input_87_cast);
            tensor<fp16, [2, 640, 32, 32]> input_89_cast = add(x = hidden_states_53_cast, y = hidden_states_41_cast);
            tensor<int32, [5]> reshape_36_shape_0 = const()[name = tensor<string, []>("reshape_36_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 32, 32])];
            tensor<fp16, [2, 32, 20, 32, 32]> reshape_36_cast = reshape(shape = reshape_36_shape_0, x = input_89_cast);
            tensor<int32, [3]> reduce_mean_27_axes_0 = const()[name = tensor<string, []>("reduce_mean_27_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_27_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_27_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_27_cast = reduce_mean(axes = reduce_mean_27_axes_0, keep_dims = reduce_mean_27_keep_dims_0, x = reshape_36_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> sub_18_cast = sub(x = reshape_36_cast, y = reduce_mean_27_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> square_9_cast = square(x = sub_18_cast);
            tensor<int32, [3]> reduce_mean_29_axes_0 = const()[name = tensor<string, []>("reduce_mean_29_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_29_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_29_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_29_cast = reduce_mean(axes = reduce_mean_29_axes_0, keep_dims = reduce_mean_29_keep_dims_0, x = square_9_cast);
            tensor<fp16, []> add_18_y_0_to_fp16 = const()[name = tensor<string, []>("add_18_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_18_cast = add(x = reduce_mean_29_cast, y = add_18_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_9_cast = sqrt(x = add_18_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> real_div_9_cast = real_div(x = sub_18_cast, y = sqrt_9_cast);
            tensor<int32, [4]> reshape_37_shape_0 = const()[name = tensor<string, []>("reshape_37_shape_0"), val = tensor<int32, [4]>([2, 640, 32, 32])];
            tensor<fp16, [2, 640, 32, 32]> reshape_37_cast = reshape(shape = reshape_37_shape_0, x = real_div_9_cast);
            tensor<fp16, [640]> add_19_mean_0_to_fp16 = const()[name = tensor<string, []>("add_19_mean_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58000832)))];
            tensor<fp16, [640]> add_19_variance_0_to_fp16 = const()[name = tensor<string, []>("add_19_variance_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58002176)))];
            tensor<fp16, [640]> add_19_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_19_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58003520)))];
            tensor<fp16, [640]> add_19_beta_0_to_fp16 = const()[name = tensor<string, []>("add_19_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58004864)))];
            tensor<fp16, []> add_19_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_19_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 32, 32]> add_19_cast = batch_norm(beta = add_19_beta_0_to_fp16, epsilon = add_19_epsilon_0_to_fp16, gamma = add_19_gamma_0_to_fp16, mean = add_19_mean_0_to_fp16, variance = add_19_variance_0_to_fp16, x = reshape_37_cast);
            tensor<fp16, [2, 640, 32, 32]> input_93_cast = silu(x = add_19_cast);
            tensor<int32, [2]> var_1617 = const()[name = tensor<string, []>("op_1617"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1619 = const()[name = tensor<string, []>("op_1619"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_55_pad_type_0 = const()[name = tensor<string, []>("hidden_states_55_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_55_pad_0 = const()[name = tensor<string, []>("hidden_states_55_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [640, 640, 3, 3]> down_blocks_1_resnets_1_conv1_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_resnets_1_conv1_weight_to_fp16"), val = tensor<fp16, [640, 640, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58006208)))];
            tensor<fp16, [640]> down_blocks_1_resnets_1_conv1_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_resnets_1_conv1_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65379072)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_55_cast = conv(bias = down_blocks_1_resnets_1_conv1_bias_to_fp16, dilations = var_1619, groups = var_984, pad = hidden_states_55_pad_0, pad_type = hidden_states_55_pad_type_0, strides = var_1617, weight = down_blocks_1_resnets_1_conv1_weight_to_fp16, x = input_93_cast);
            tensor<int32, [2]> var_1625 = const()[name = tensor<string, []>("op_1625"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1627 = const()[name = tensor<string, []>("op_1627"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_7_pad_type_0 = const()[name = tensor<string, []>("temb_7_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_7_pad_0 = const()[name = tensor<string, []>("temb_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1280, 1, 1]> down_blocks_1_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [640, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65380416)))];
            tensor<fp16, [640]> down_blocks_1_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67018880)))];
            tensor<fp16, [2, 640, 1, 1]> temb_7_cast = conv(bias = down_blocks_1_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_1627, groups = var_984, pad = temb_7_pad_0, pad_type = temb_7_pad_type_0, strides = var_1625, weight = down_blocks_1_resnets_1_time_emb_proj_weight_to_fp16, x = input_15_cast);
            tensor<fp16, [2, 640, 32, 32]> input_97_cast = add(x = hidden_states_55_cast, y = temb_7_cast);
            tensor<int32, [5]> reshape_40_shape_0 = const()[name = tensor<string, []>("reshape_40_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 32, 32])];
            tensor<fp16, [2, 32, 20, 32, 32]> reshape_40_cast = reshape(shape = reshape_40_shape_0, x = input_97_cast);
            tensor<int32, [3]> reduce_mean_30_axes_0 = const()[name = tensor<string, []>("reduce_mean_30_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_30_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_30_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_30_cast = reduce_mean(axes = reduce_mean_30_axes_0, keep_dims = reduce_mean_30_keep_dims_0, x = reshape_40_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> sub_20_cast = sub(x = reshape_40_cast, y = reduce_mean_30_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> square_10_cast = square(x = sub_20_cast);
            tensor<int32, [3]> reduce_mean_32_axes_0 = const()[name = tensor<string, []>("reduce_mean_32_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_32_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_32_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_32_cast = reduce_mean(axes = reduce_mean_32_axes_0, keep_dims = reduce_mean_32_keep_dims_0, x = square_10_cast);
            tensor<fp16, []> add_20_y_0_to_fp16 = const()[name = tensor<string, []>("add_20_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_20_cast = add(x = reduce_mean_32_cast, y = add_20_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_10_cast = sqrt(x = add_20_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> real_div_10_cast = real_div(x = sub_20_cast, y = sqrt_10_cast);
            tensor<int32, [4]> reshape_41_shape_0 = const()[name = tensor<string, []>("reshape_41_shape_0"), val = tensor<int32, [4]>([2, 640, 32, 32])];
            tensor<fp16, [2, 640, 32, 32]> reshape_41_cast = reshape(shape = reshape_41_shape_0, x = real_div_10_cast);
            tensor<fp16, [640]> add_21_mean_0_to_fp16 = const()[name = tensor<string, []>("add_21_mean_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67020224)))];
            tensor<fp16, [640]> add_21_variance_0_to_fp16 = const()[name = tensor<string, []>("add_21_variance_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67021568)))];
            tensor<fp16, [640]> add_21_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_21_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67022912)))];
            tensor<fp16, [640]> add_21_beta_0_to_fp16 = const()[name = tensor<string, []>("add_21_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67024256)))];
            tensor<fp16, []> add_21_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_21_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 32, 32]> add_21_cast = batch_norm(beta = add_21_beta_0_to_fp16, epsilon = add_21_epsilon_0_to_fp16, gamma = add_21_gamma_0_to_fp16, mean = add_21_mean_0_to_fp16, variance = add_21_variance_0_to_fp16, x = reshape_41_cast);
            tensor<fp16, [2, 640, 32, 32]> input_101_cast = silu(x = add_21_cast);
            tensor<int32, [2]> var_1637 = const()[name = tensor<string, []>("op_1637"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1639 = const()[name = tensor<string, []>("op_1639"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_57_pad_type_0 = const()[name = tensor<string, []>("hidden_states_57_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_57_pad_0 = const()[name = tensor<string, []>("hidden_states_57_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [640, 640, 3, 3]> down_blocks_1_resnets_1_conv2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_resnets_1_conv2_weight_to_fp16"), val = tensor<fp16, [640, 640, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67025600)))];
            tensor<fp16, [640]> down_blocks_1_resnets_1_conv2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_resnets_1_conv2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74398464)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_57_cast = conv(bias = down_blocks_1_resnets_1_conv2_bias_to_fp16, dilations = var_1639, groups = var_984, pad = hidden_states_57_pad_0, pad_type = hidden_states_57_pad_type_0, strides = var_1637, weight = down_blocks_1_resnets_1_conv2_weight_to_fp16, x = input_101_cast);
            tensor<fp16, [2, 640, 32, 32]> hidden_states_59_cast = add(x = input_89_cast, y = hidden_states_57_cast);
            tensor<int32, [5]> reshape_44_shape_0 = const()[name = tensor<string, []>("reshape_44_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 32, 32])];
            tensor<fp16, [2, 32, 20, 32, 32]> reshape_44_cast = reshape(shape = reshape_44_shape_0, x = hidden_states_59_cast);
            tensor<int32, [3]> reduce_mean_33_axes_0 = const()[name = tensor<string, []>("reduce_mean_33_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_33_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_33_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_33_cast = reduce_mean(axes = reduce_mean_33_axes_0, keep_dims = reduce_mean_33_keep_dims_0, x = reshape_44_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> sub_22_cast = sub(x = reshape_44_cast, y = reduce_mean_33_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> square_11_cast = square(x = sub_22_cast);
            tensor<int32, [3]> reduce_mean_35_axes_0 = const()[name = tensor<string, []>("reduce_mean_35_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_35_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_35_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_35_cast = reduce_mean(axes = reduce_mean_35_axes_0, keep_dims = reduce_mean_35_keep_dims_0, x = square_11_cast);
            tensor<fp16, []> add_22_y_0_to_fp16 = const()[name = tensor<string, []>("add_22_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_22_cast = add(x = reduce_mean_35_cast, y = add_22_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_11_cast = sqrt(x = add_22_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> real_div_11_cast = real_div(x = sub_22_cast, y = sqrt_11_cast);
            tensor<int32, [4]> reshape_45_shape_0 = const()[name = tensor<string, []>("reshape_45_shape_0"), val = tensor<int32, [4]>([2, 640, 32, 32])];
            tensor<fp16, [2, 640, 32, 32]> reshape_45_cast = reshape(shape = reshape_45_shape_0, x = real_div_11_cast);
            tensor<fp16, [640]> add_23_mean_0_to_fp16 = const()[name = tensor<string, []>("add_23_mean_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74399808)))];
            tensor<fp16, [640]> add_23_variance_0_to_fp16 = const()[name = tensor<string, []>("add_23_variance_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74401152)))];
            tensor<fp16, [640]> add_23_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_23_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74402496)))];
            tensor<fp16, [640]> add_23_beta_0_to_fp16 = const()[name = tensor<string, []>("add_23_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74403840)))];
            tensor<fp16, []> add_23_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_23_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 32, 32]> add_23_cast = batch_norm(beta = add_23_beta_0_to_fp16, epsilon = add_23_epsilon_0_to_fp16, gamma = add_23_gamma_0_to_fp16, mean = add_23_mean_0_to_fp16, variance = add_23_variance_0_to_fp16, x = reshape_45_cast);
            tensor<int32, [2]> var_1659 = const()[name = tensor<string, []>("op_1659"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1661 = const()[name = tensor<string, []>("op_1661"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_61_pad_type_0 = const()[name = tensor<string, []>("hidden_states_61_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_61_pad_0 = const()[name = tensor<string, []>("hidden_states_61_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_1_proj_in_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_proj_in_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74405184)))];
            tensor<fp16, [640]> down_blocks_1_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_proj_in_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75224448)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_61_cast = conv(bias = down_blocks_1_attentions_1_proj_in_bias_to_fp16, dilations = var_1661, groups = var_984, pad = hidden_states_61_pad_0, pad_type = hidden_states_61_pad_type_0, strides = var_1659, weight = down_blocks_1_attentions_1_proj_in_weight_to_fp16, x = add_23_cast);
            tensor<int32, [4]> var_1666 = const()[name = tensor<string, []>("op_1666"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<fp16, [2, 640, 1, 1024]> inputs_19_cast = reshape(shape = var_1666, x = hidden_states_61_cast);
            tensor<int32, [1]> var_1676 = const()[name = tensor<string, []>("op_1676"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_19_cast = reduce_mean(axes = var_1676, keep_dims = var_979, x = inputs_19_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_19_cast = sub(x = inputs_19_cast, y = channels_mean_19_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_sq_19_cast = mul(x = zero_mean_19_cast, y = zero_mean_19_cast);
            tensor<int32, [1]> var_1680 = const()[name = tensor<string, []>("op_1680"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_1681_cast = reduce_mean(axes = var_1680, keep_dims = var_979, x = zero_mean_sq_19_cast);
            tensor<fp16, []> var_1682_to_fp16 = const()[name = tensor<string, []>("op_1682_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_1683_cast = add(x = var_1681_cast, y = var_1682_to_fp16);
            tensor<fp16, []> denom_19_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_19_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_19_cast = rsqrt(epsilon = denom_19_epsilon_0_to_fp16, x = var_1683_cast);
            tensor<fp16, [2, 640, 1, 1024]> out_19_cast = mul(x = zero_mean_19_cast, y = denom_19_cast);
            tensor<fp16, [1, 640, 1, 1]> var_1687_to_fp16 = const()[name = tensor<string, []>("op_1687_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75225792)))];
            tensor<fp16, [2, 640, 1, 1024]> var_1688_cast = add(x = out_19_cast, y = var_1687_to_fp16);
            tensor<fp16, [1, 640, 1, 1]> var_1690_to_fp16 = const()[name = tensor<string, []>("op_1690_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75227136)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_63_cast = mul(x = var_1688_cast, y = var_1690_to_fp16);
            tensor<int32, [2]> var_1697 = const()[name = tensor<string, []>("op_1697"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1699 = const()[name = tensor<string, []>("op_1699"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_13_pad_type_0 = const()[name = tensor<string, []>("q_13_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_13_pad_0 = const()[name = tensor<string, []>("q_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75228480)))];
            tensor<fp16, [2, 640, 1, 1024]> q_13_cast = conv(dilations = var_1699, groups = var_984, pad = q_13_pad_0, pad_type = q_13_pad_type_0, strides = var_1697, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_63_cast);
            tensor<int32, [2]> var_1703 = const()[name = tensor<string, []>("op_1703"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1705 = const()[name = tensor<string, []>("op_1705"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_25_pad_type_0 = const()[name = tensor<string, []>("k_25_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_25_pad_0 = const()[name = tensor<string, []>("k_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76047744)))];
            tensor<fp16, [2, 640, 1, 1024]> k_25_cast = conv(dilations = var_1705, groups = var_984, pad = k_25_pad_0, pad_type = k_25_pad_type_0, strides = var_1703, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_63_cast);
            tensor<int32, [2]> var_1709 = const()[name = tensor<string, []>("op_1709"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1711 = const()[name = tensor<string, []>("op_1711"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_13_pad_type_0 = const()[name = tensor<string, []>("v_13_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_13_pad_0 = const()[name = tensor<string, []>("v_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76867008)))];
            tensor<fp16, [2, 640, 1, 1024]> v_13_cast = conv(dilations = var_1711, groups = var_984, pad = v_13_pad_0, pad_type = v_13_pad_type_0, strides = var_1709, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_63_cast);
            tensor<int32, [4]> var_1715_begin_0 = const()[name = tensor<string, []>("op_1715_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1715_end_0 = const()[name = tensor<string, []>("op_1715_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_1715_end_mask_0 = const()[name = tensor<string, []>("op_1715_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1715_cast = slice_by_index(begin = var_1715_begin_0, end = var_1715_end_0, end_mask = var_1715_end_mask_0, x = q_13_cast);
            tensor<int32, [4]> var_1719_begin_0 = const()[name = tensor<string, []>("op_1719_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_1719_end_0 = const()[name = tensor<string, []>("op_1719_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_1719_end_mask_0 = const()[name = tensor<string, []>("op_1719_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1719_cast = slice_by_index(begin = var_1719_begin_0, end = var_1719_end_0, end_mask = var_1719_end_mask_0, x = q_13_cast);
            tensor<int32, [4]> var_1723_begin_0 = const()[name = tensor<string, []>("op_1723_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_1723_end_0 = const()[name = tensor<string, []>("op_1723_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_1723_end_mask_0 = const()[name = tensor<string, []>("op_1723_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1723_cast = slice_by_index(begin = var_1723_begin_0, end = var_1723_end_0, end_mask = var_1723_end_mask_0, x = q_13_cast);
            tensor<int32, [4]> var_1727_begin_0 = const()[name = tensor<string, []>("op_1727_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_1727_end_0 = const()[name = tensor<string, []>("op_1727_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_1727_end_mask_0 = const()[name = tensor<string, []>("op_1727_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1727_cast = slice_by_index(begin = var_1727_begin_0, end = var_1727_end_0, end_mask = var_1727_end_mask_0, x = q_13_cast);
            tensor<int32, [4]> var_1731_begin_0 = const()[name = tensor<string, []>("op_1731_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_1731_end_0 = const()[name = tensor<string, []>("op_1731_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_1731_end_mask_0 = const()[name = tensor<string, []>("op_1731_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1731_cast = slice_by_index(begin = var_1731_begin_0, end = var_1731_end_0, end_mask = var_1731_end_mask_0, x = q_13_cast);
            tensor<int32, [4]> var_1735_begin_0 = const()[name = tensor<string, []>("op_1735_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_1735_end_0 = const()[name = tensor<string, []>("op_1735_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_1735_end_mask_0 = const()[name = tensor<string, []>("op_1735_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1735_cast = slice_by_index(begin = var_1735_begin_0, end = var_1735_end_0, end_mask = var_1735_end_mask_0, x = q_13_cast);
            tensor<int32, [4]> var_1739_begin_0 = const()[name = tensor<string, []>("op_1739_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_1739_end_0 = const()[name = tensor<string, []>("op_1739_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_1739_end_mask_0 = const()[name = tensor<string, []>("op_1739_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1739_cast = slice_by_index(begin = var_1739_begin_0, end = var_1739_end_0, end_mask = var_1739_end_mask_0, x = q_13_cast);
            tensor<int32, [4]> var_1743_begin_0 = const()[name = tensor<string, []>("op_1743_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_1743_end_0 = const()[name = tensor<string, []>("op_1743_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_1743_end_mask_0 = const()[name = tensor<string, []>("op_1743_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1743_cast = slice_by_index(begin = var_1743_begin_0, end = var_1743_end_0, end_mask = var_1743_end_mask_0, x = q_13_cast);
            tensor<int32, [4]> var_1747_begin_0 = const()[name = tensor<string, []>("op_1747_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_1747_end_0 = const()[name = tensor<string, []>("op_1747_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_1747_end_mask_0 = const()[name = tensor<string, []>("op_1747_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1747_cast = slice_by_index(begin = var_1747_begin_0, end = var_1747_end_0, end_mask = var_1747_end_mask_0, x = q_13_cast);
            tensor<int32, [4]> var_1751_begin_0 = const()[name = tensor<string, []>("op_1751_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_1751_end_0 = const()[name = tensor<string, []>("op_1751_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_1751_end_mask_0 = const()[name = tensor<string, []>("op_1751_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1751_cast = slice_by_index(begin = var_1751_begin_0, end = var_1751_end_0, end_mask = var_1751_end_mask_0, x = q_13_cast);
            tensor<int32, [4]> k_27_perm_0 = const()[name = tensor<string, []>("k_27_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_1758_begin_0 = const()[name = tensor<string, []>("op_1758_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1758_end_0 = const()[name = tensor<string, []>("op_1758_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_1758_end_mask_0 = const()[name = tensor<string, []>("op_1758_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 640]> transpose_25 = transpose(perm = k_27_perm_0, x = k_25_cast);
            tensor<fp16, [2, 1024, 1, 64]> var_1758_cast = slice_by_index(begin = var_1758_begin_0, end = var_1758_end_0, end_mask = var_1758_end_mask_0, x = transpose_25);
            tensor<int32, [4]> var_1762_begin_0 = const()[name = tensor<string, []>("op_1762_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_1762_end_0 = const()[name = tensor<string, []>("op_1762_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_1762_end_mask_0 = const()[name = tensor<string, []>("op_1762_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_1762_cast = slice_by_index(begin = var_1762_begin_0, end = var_1762_end_0, end_mask = var_1762_end_mask_0, x = transpose_25);
            tensor<int32, [4]> var_1766_begin_0 = const()[name = tensor<string, []>("op_1766_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_1766_end_0 = const()[name = tensor<string, []>("op_1766_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_1766_end_mask_0 = const()[name = tensor<string, []>("op_1766_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_1766_cast = slice_by_index(begin = var_1766_begin_0, end = var_1766_end_0, end_mask = var_1766_end_mask_0, x = transpose_25);
            tensor<int32, [4]> var_1770_begin_0 = const()[name = tensor<string, []>("op_1770_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_1770_end_0 = const()[name = tensor<string, []>("op_1770_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_1770_end_mask_0 = const()[name = tensor<string, []>("op_1770_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_1770_cast = slice_by_index(begin = var_1770_begin_0, end = var_1770_end_0, end_mask = var_1770_end_mask_0, x = transpose_25);
            tensor<int32, [4]> var_1774_begin_0 = const()[name = tensor<string, []>("op_1774_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_1774_end_0 = const()[name = tensor<string, []>("op_1774_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_1774_end_mask_0 = const()[name = tensor<string, []>("op_1774_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_1774_cast = slice_by_index(begin = var_1774_begin_0, end = var_1774_end_0, end_mask = var_1774_end_mask_0, x = transpose_25);
            tensor<int32, [4]> var_1778_begin_0 = const()[name = tensor<string, []>("op_1778_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_1778_end_0 = const()[name = tensor<string, []>("op_1778_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_1778_end_mask_0 = const()[name = tensor<string, []>("op_1778_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_1778_cast = slice_by_index(begin = var_1778_begin_0, end = var_1778_end_0, end_mask = var_1778_end_mask_0, x = transpose_25);
            tensor<int32, [4]> var_1782_begin_0 = const()[name = tensor<string, []>("op_1782_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_1782_end_0 = const()[name = tensor<string, []>("op_1782_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_1782_end_mask_0 = const()[name = tensor<string, []>("op_1782_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_1782_cast = slice_by_index(begin = var_1782_begin_0, end = var_1782_end_0, end_mask = var_1782_end_mask_0, x = transpose_25);
            tensor<int32, [4]> var_1786_begin_0 = const()[name = tensor<string, []>("op_1786_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_1786_end_0 = const()[name = tensor<string, []>("op_1786_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_1786_end_mask_0 = const()[name = tensor<string, []>("op_1786_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_1786_cast = slice_by_index(begin = var_1786_begin_0, end = var_1786_end_0, end_mask = var_1786_end_mask_0, x = transpose_25);
            tensor<int32, [4]> var_1790_begin_0 = const()[name = tensor<string, []>("op_1790_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_1790_end_0 = const()[name = tensor<string, []>("op_1790_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_1790_end_mask_0 = const()[name = tensor<string, []>("op_1790_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_1790_cast = slice_by_index(begin = var_1790_begin_0, end = var_1790_end_0, end_mask = var_1790_end_mask_0, x = transpose_25);
            tensor<int32, [4]> var_1794_begin_0 = const()[name = tensor<string, []>("op_1794_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_1794_end_0 = const()[name = tensor<string, []>("op_1794_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_1794_end_mask_0 = const()[name = tensor<string, []>("op_1794_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_1794_cast = slice_by_index(begin = var_1794_begin_0, end = var_1794_end_0, end_mask = var_1794_end_mask_0, x = transpose_25);
            tensor<int32, [4]> var_1796_begin_0 = const()[name = tensor<string, []>("op_1796_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1796_end_0 = const()[name = tensor<string, []>("op_1796_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_1796_end_mask_0 = const()[name = tensor<string, []>("op_1796_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1796_cast = slice_by_index(begin = var_1796_begin_0, end = var_1796_end_0, end_mask = var_1796_end_mask_0, x = v_13_cast);
            tensor<int32, [4]> var_1800_begin_0 = const()[name = tensor<string, []>("op_1800_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_1800_end_0 = const()[name = tensor<string, []>("op_1800_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_1800_end_mask_0 = const()[name = tensor<string, []>("op_1800_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1800_cast = slice_by_index(begin = var_1800_begin_0, end = var_1800_end_0, end_mask = var_1800_end_mask_0, x = v_13_cast);
            tensor<int32, [4]> var_1804_begin_0 = const()[name = tensor<string, []>("op_1804_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_1804_end_0 = const()[name = tensor<string, []>("op_1804_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_1804_end_mask_0 = const()[name = tensor<string, []>("op_1804_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1804_cast = slice_by_index(begin = var_1804_begin_0, end = var_1804_end_0, end_mask = var_1804_end_mask_0, x = v_13_cast);
            tensor<int32, [4]> var_1808_begin_0 = const()[name = tensor<string, []>("op_1808_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_1808_end_0 = const()[name = tensor<string, []>("op_1808_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_1808_end_mask_0 = const()[name = tensor<string, []>("op_1808_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1808_cast = slice_by_index(begin = var_1808_begin_0, end = var_1808_end_0, end_mask = var_1808_end_mask_0, x = v_13_cast);
            tensor<int32, [4]> var_1812_begin_0 = const()[name = tensor<string, []>("op_1812_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_1812_end_0 = const()[name = tensor<string, []>("op_1812_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_1812_end_mask_0 = const()[name = tensor<string, []>("op_1812_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1812_cast = slice_by_index(begin = var_1812_begin_0, end = var_1812_end_0, end_mask = var_1812_end_mask_0, x = v_13_cast);
            tensor<int32, [4]> var_1816_begin_0 = const()[name = tensor<string, []>("op_1816_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_1816_end_0 = const()[name = tensor<string, []>("op_1816_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_1816_end_mask_0 = const()[name = tensor<string, []>("op_1816_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1816_cast = slice_by_index(begin = var_1816_begin_0, end = var_1816_end_0, end_mask = var_1816_end_mask_0, x = v_13_cast);
            tensor<int32, [4]> var_1820_begin_0 = const()[name = tensor<string, []>("op_1820_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_1820_end_0 = const()[name = tensor<string, []>("op_1820_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_1820_end_mask_0 = const()[name = tensor<string, []>("op_1820_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1820_cast = slice_by_index(begin = var_1820_begin_0, end = var_1820_end_0, end_mask = var_1820_end_mask_0, x = v_13_cast);
            tensor<int32, [4]> var_1824_begin_0 = const()[name = tensor<string, []>("op_1824_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_1824_end_0 = const()[name = tensor<string, []>("op_1824_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_1824_end_mask_0 = const()[name = tensor<string, []>("op_1824_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1824_cast = slice_by_index(begin = var_1824_begin_0, end = var_1824_end_0, end_mask = var_1824_end_mask_0, x = v_13_cast);
            tensor<int32, [4]> var_1828_begin_0 = const()[name = tensor<string, []>("op_1828_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_1828_end_0 = const()[name = tensor<string, []>("op_1828_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_1828_end_mask_0 = const()[name = tensor<string, []>("op_1828_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1828_cast = slice_by_index(begin = var_1828_begin_0, end = var_1828_end_0, end_mask = var_1828_end_mask_0, x = v_13_cast);
            tensor<int32, [4]> var_1832_begin_0 = const()[name = tensor<string, []>("op_1832_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_1832_end_0 = const()[name = tensor<string, []>("op_1832_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_1832_end_mask_0 = const()[name = tensor<string, []>("op_1832_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1832_cast = slice_by_index(begin = var_1832_begin_0, end = var_1832_end_0, end_mask = var_1832_end_mask_0, x = v_13_cast);
            tensor<string, []> var_1836_equation_0 = const()[name = tensor<string, []>("op_1836_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_1836_cast = einsum(equation = var_1836_equation_0, values = (var_1758_cast, var_1715_cast));
            tensor<fp16, []> var_1837_to_fp16 = const()[name = tensor<string, []>("op_1837_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_81_cast = mul(x = var_1836_cast, y = var_1837_to_fp16);
            tensor<string, []> var_1840_equation_0 = const()[name = tensor<string, []>("op_1840_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_1840_cast = einsum(equation = var_1840_equation_0, values = (var_1762_cast, var_1719_cast));
            tensor<fp16, []> var_1841_to_fp16 = const()[name = tensor<string, []>("op_1841_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_83_cast = mul(x = var_1840_cast, y = var_1841_to_fp16);
            tensor<string, []> var_1844_equation_0 = const()[name = tensor<string, []>("op_1844_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_1844_cast = einsum(equation = var_1844_equation_0, values = (var_1766_cast, var_1723_cast));
            tensor<fp16, []> var_1845_to_fp16 = const()[name = tensor<string, []>("op_1845_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_85_cast = mul(x = var_1844_cast, y = var_1845_to_fp16);
            tensor<string, []> var_1848_equation_0 = const()[name = tensor<string, []>("op_1848_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_1848_cast = einsum(equation = var_1848_equation_0, values = (var_1770_cast, var_1727_cast));
            tensor<fp16, []> var_1849_to_fp16 = const()[name = tensor<string, []>("op_1849_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_87_cast = mul(x = var_1848_cast, y = var_1849_to_fp16);
            tensor<string, []> var_1852_equation_0 = const()[name = tensor<string, []>("op_1852_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_1852_cast = einsum(equation = var_1852_equation_0, values = (var_1774_cast, var_1731_cast));
            tensor<fp16, []> var_1853_to_fp16 = const()[name = tensor<string, []>("op_1853_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_89_cast = mul(x = var_1852_cast, y = var_1853_to_fp16);
            tensor<string, []> var_1856_equation_0 = const()[name = tensor<string, []>("op_1856_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_1856_cast = einsum(equation = var_1856_equation_0, values = (var_1778_cast, var_1735_cast));
            tensor<fp16, []> var_1857_to_fp16 = const()[name = tensor<string, []>("op_1857_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_91_cast = mul(x = var_1856_cast, y = var_1857_to_fp16);
            tensor<string, []> var_1860_equation_0 = const()[name = tensor<string, []>("op_1860_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_1860_cast = einsum(equation = var_1860_equation_0, values = (var_1782_cast, var_1739_cast));
            tensor<fp16, []> var_1861_to_fp16 = const()[name = tensor<string, []>("op_1861_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_93_cast = mul(x = var_1860_cast, y = var_1861_to_fp16);
            tensor<string, []> var_1864_equation_0 = const()[name = tensor<string, []>("op_1864_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_1864_cast = einsum(equation = var_1864_equation_0, values = (var_1786_cast, var_1743_cast));
            tensor<fp16, []> var_1865_to_fp16 = const()[name = tensor<string, []>("op_1865_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_95_cast = mul(x = var_1864_cast, y = var_1865_to_fp16);
            tensor<string, []> var_1868_equation_0 = const()[name = tensor<string, []>("op_1868_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_1868_cast = einsum(equation = var_1868_equation_0, values = (var_1790_cast, var_1747_cast));
            tensor<fp16, []> var_1869_to_fp16 = const()[name = tensor<string, []>("op_1869_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_97_cast = mul(x = var_1868_cast, y = var_1869_to_fp16);
            tensor<string, []> var_1872_equation_0 = const()[name = tensor<string, []>("op_1872_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_1872_cast = einsum(equation = var_1872_equation_0, values = (var_1794_cast, var_1751_cast));
            tensor<fp16, []> var_1873_to_fp16 = const()[name = tensor<string, []>("op_1873_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_99_cast = mul(x = var_1872_cast, y = var_1873_to_fp16);
            tensor<fp16, [2, 1024, 1, 1024]> var_1875_cast = softmax(axis = var_984, x = aw_81_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_1876_cast = softmax(axis = var_984, x = aw_83_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_1877_cast = softmax(axis = var_984, x = aw_85_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_1878_cast = softmax(axis = var_984, x = aw_87_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_1879_cast = softmax(axis = var_984, x = aw_89_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_1880_cast = softmax(axis = var_984, x = aw_91_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_1881_cast = softmax(axis = var_984, x = aw_93_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_1882_cast = softmax(axis = var_984, x = aw_95_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_1883_cast = softmax(axis = var_984, x = aw_97_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_1884_cast = softmax(axis = var_984, x = aw_99_cast);
            tensor<string, []> var_1886_equation_0 = const()[name = tensor<string, []>("op_1886_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1886_cast = einsum(equation = var_1886_equation_0, values = (var_1796_cast, var_1875_cast));
            tensor<string, []> var_1888_equation_0 = const()[name = tensor<string, []>("op_1888_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1888_cast = einsum(equation = var_1888_equation_0, values = (var_1800_cast, var_1876_cast));
            tensor<string, []> var_1890_equation_0 = const()[name = tensor<string, []>("op_1890_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1890_cast = einsum(equation = var_1890_equation_0, values = (var_1804_cast, var_1877_cast));
            tensor<string, []> var_1892_equation_0 = const()[name = tensor<string, []>("op_1892_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1892_cast = einsum(equation = var_1892_equation_0, values = (var_1808_cast, var_1878_cast));
            tensor<string, []> var_1894_equation_0 = const()[name = tensor<string, []>("op_1894_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1894_cast = einsum(equation = var_1894_equation_0, values = (var_1812_cast, var_1879_cast));
            tensor<string, []> var_1896_equation_0 = const()[name = tensor<string, []>("op_1896_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1896_cast = einsum(equation = var_1896_equation_0, values = (var_1816_cast, var_1880_cast));
            tensor<string, []> var_1898_equation_0 = const()[name = tensor<string, []>("op_1898_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1898_cast = einsum(equation = var_1898_equation_0, values = (var_1820_cast, var_1881_cast));
            tensor<string, []> var_1900_equation_0 = const()[name = tensor<string, []>("op_1900_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1900_cast = einsum(equation = var_1900_equation_0, values = (var_1824_cast, var_1882_cast));
            tensor<string, []> var_1902_equation_0 = const()[name = tensor<string, []>("op_1902_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1902_cast = einsum(equation = var_1902_equation_0, values = (var_1828_cast, var_1883_cast));
            tensor<string, []> var_1904_equation_0 = const()[name = tensor<string, []>("op_1904_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_1904_cast = einsum(equation = var_1904_equation_0, values = (var_1832_cast, var_1884_cast));
            tensor<bool, []> input_105_interleave_0 = const()[name = tensor<string, []>("input_105_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 1024]> input_105_cast = concat(axis = var_984, interleave = input_105_interleave_0, values = (var_1886_cast, var_1888_cast, var_1890_cast, var_1892_cast, var_1894_cast, var_1896_cast, var_1898_cast, var_1900_cast, var_1902_cast, var_1904_cast));
            tensor<int32, [2]> var_1910 = const()[name = tensor<string, []>("op_1910"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1912 = const()[name = tensor<string, []>("op_1912"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_1914_pad_type_0 = const()[name = tensor<string, []>("op_1914_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_1914_pad_0 = const()[name = tensor<string, []>("op_1914_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77686272)))];
            tensor<fp16, [640]> down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78505536)))];
            tensor<fp16, [2, 640, 1, 1024]> var_1914_cast = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_1912, groups = var_984, pad = var_1914_pad_0, pad_type = var_1914_pad_type_0, strides = var_1910, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_105_cast);
            tensor<fp16, [2, 640, 1, 1024]> inputs_21_cast = add(x = var_1914_cast, y = inputs_19_cast);
            tensor<int32, [1]> var_1918 = const()[name = tensor<string, []>("op_1918"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_21_cast = reduce_mean(axes = var_1918, keep_dims = var_979, x = inputs_21_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_21_cast = sub(x = inputs_21_cast, y = channels_mean_21_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_sq_21_cast = mul(x = zero_mean_21_cast, y = zero_mean_21_cast);
            tensor<int32, [1]> var_1922 = const()[name = tensor<string, []>("op_1922"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_1923_cast = reduce_mean(axes = var_1922, keep_dims = var_979, x = zero_mean_sq_21_cast);
            tensor<fp16, []> var_1924_to_fp16 = const()[name = tensor<string, []>("op_1924_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_1925_cast = add(x = var_1923_cast, y = var_1924_to_fp16);
            tensor<fp16, []> denom_21_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_21_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_21_cast = rsqrt(epsilon = denom_21_epsilon_0_to_fp16, x = var_1925_cast);
            tensor<fp16, [2, 640, 1, 1024]> out_21_cast = mul(x = zero_mean_21_cast, y = denom_21_cast);
            tensor<fp16, [1, 640, 1, 1]> var_1929_to_fp16 = const()[name = tensor<string, []>("op_1929_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78506880)))];
            tensor<fp16, [2, 640, 1, 1024]> var_1930_cast = add(x = out_21_cast, y = var_1929_to_fp16);
            tensor<fp16, [1, 640, 1, 1]> var_1932_to_fp16 = const()[name = tensor<string, []>("op_1932_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78508224)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_65_cast = mul(x = var_1930_cast, y = var_1932_to_fp16);
            tensor<int32, [2]> var_1939 = const()[name = tensor<string, []>("op_1939"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1941 = const()[name = tensor<string, []>("op_1941"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_15_pad_type_0 = const()[name = tensor<string, []>("q_15_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_15_pad_0 = const()[name = tensor<string, []>("q_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78509568)))];
            tensor<fp16, [2, 640, 1, 1024]> q_15_cast = conv(dilations = var_1941, groups = var_984, pad = q_15_pad_0, pad_type = q_15_pad_type_0, strides = var_1939, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_65_cast);
            tensor<int32, [2]> var_1945 = const()[name = tensor<string, []>("op_1945"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1947 = const()[name = tensor<string, []>("op_1947"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_29_pad_type_0 = const()[name = tensor<string, []>("k_29_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_29_pad_0 = const()[name = tensor<string, []>("k_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1024, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [640, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79328832)))];
            tensor<fp16, [2, 640, 1, 77]> k_29_cast = conv(dilations = var_1947, groups = var_984, pad = k_29_pad_0, pad_type = k_29_pad_type_0, strides = var_1945, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [2]> var_1951 = const()[name = tensor<string, []>("op_1951"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1953 = const()[name = tensor<string, []>("op_1953"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_15_pad_type_0 = const()[name = tensor<string, []>("v_15_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_15_pad_0 = const()[name = tensor<string, []>("v_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1024, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [640, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80639616)))];
            tensor<fp16, [2, 640, 1, 77]> v_15_cast = conv(dilations = var_1953, groups = var_984, pad = v_15_pad_0, pad_type = v_15_pad_type_0, strides = var_1951, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [4]> var_1957_begin_0 = const()[name = tensor<string, []>("op_1957_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1957_end_0 = const()[name = tensor<string, []>("op_1957_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_1957_end_mask_0 = const()[name = tensor<string, []>("op_1957_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1957_cast = slice_by_index(begin = var_1957_begin_0, end = var_1957_end_0, end_mask = var_1957_end_mask_0, x = q_15_cast);
            tensor<int32, [4]> var_1961_begin_0 = const()[name = tensor<string, []>("op_1961_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_1961_end_0 = const()[name = tensor<string, []>("op_1961_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_1961_end_mask_0 = const()[name = tensor<string, []>("op_1961_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1961_cast = slice_by_index(begin = var_1961_begin_0, end = var_1961_end_0, end_mask = var_1961_end_mask_0, x = q_15_cast);
            tensor<int32, [4]> var_1965_begin_0 = const()[name = tensor<string, []>("op_1965_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_1965_end_0 = const()[name = tensor<string, []>("op_1965_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_1965_end_mask_0 = const()[name = tensor<string, []>("op_1965_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1965_cast = slice_by_index(begin = var_1965_begin_0, end = var_1965_end_0, end_mask = var_1965_end_mask_0, x = q_15_cast);
            tensor<int32, [4]> var_1969_begin_0 = const()[name = tensor<string, []>("op_1969_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_1969_end_0 = const()[name = tensor<string, []>("op_1969_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_1969_end_mask_0 = const()[name = tensor<string, []>("op_1969_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1969_cast = slice_by_index(begin = var_1969_begin_0, end = var_1969_end_0, end_mask = var_1969_end_mask_0, x = q_15_cast);
            tensor<int32, [4]> var_1973_begin_0 = const()[name = tensor<string, []>("op_1973_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_1973_end_0 = const()[name = tensor<string, []>("op_1973_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_1973_end_mask_0 = const()[name = tensor<string, []>("op_1973_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1973_cast = slice_by_index(begin = var_1973_begin_0, end = var_1973_end_0, end_mask = var_1973_end_mask_0, x = q_15_cast);
            tensor<int32, [4]> var_1977_begin_0 = const()[name = tensor<string, []>("op_1977_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_1977_end_0 = const()[name = tensor<string, []>("op_1977_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_1977_end_mask_0 = const()[name = tensor<string, []>("op_1977_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1977_cast = slice_by_index(begin = var_1977_begin_0, end = var_1977_end_0, end_mask = var_1977_end_mask_0, x = q_15_cast);
            tensor<int32, [4]> var_1981_begin_0 = const()[name = tensor<string, []>("op_1981_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_1981_end_0 = const()[name = tensor<string, []>("op_1981_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_1981_end_mask_0 = const()[name = tensor<string, []>("op_1981_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1981_cast = slice_by_index(begin = var_1981_begin_0, end = var_1981_end_0, end_mask = var_1981_end_mask_0, x = q_15_cast);
            tensor<int32, [4]> var_1985_begin_0 = const()[name = tensor<string, []>("op_1985_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_1985_end_0 = const()[name = tensor<string, []>("op_1985_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_1985_end_mask_0 = const()[name = tensor<string, []>("op_1985_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1985_cast = slice_by_index(begin = var_1985_begin_0, end = var_1985_end_0, end_mask = var_1985_end_mask_0, x = q_15_cast);
            tensor<int32, [4]> var_1989_begin_0 = const()[name = tensor<string, []>("op_1989_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_1989_end_0 = const()[name = tensor<string, []>("op_1989_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_1989_end_mask_0 = const()[name = tensor<string, []>("op_1989_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1989_cast = slice_by_index(begin = var_1989_begin_0, end = var_1989_end_0, end_mask = var_1989_end_mask_0, x = q_15_cast);
            tensor<int32, [4]> var_1993_begin_0 = const()[name = tensor<string, []>("op_1993_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_1993_end_0 = const()[name = tensor<string, []>("op_1993_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_1993_end_mask_0 = const()[name = tensor<string, []>("op_1993_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_1993_cast = slice_by_index(begin = var_1993_begin_0, end = var_1993_end_0, end_mask = var_1993_end_mask_0, x = q_15_cast);
            tensor<int32, [4]> k_31_perm_0 = const()[name = tensor<string, []>("k_31_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_2000_begin_0 = const()[name = tensor<string, []>("op_2000_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2000_end_0 = const()[name = tensor<string, []>("op_2000_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_2000_end_mask_0 = const()[name = tensor<string, []>("op_2000_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 640]> transpose_24 = transpose(perm = k_31_perm_0, x = k_29_cast);
            tensor<fp16, [2, 77, 1, 64]> var_2000_cast = slice_by_index(begin = var_2000_begin_0, end = var_2000_end_0, end_mask = var_2000_end_mask_0, x = transpose_24);
            tensor<int32, [4]> var_2004_begin_0 = const()[name = tensor<string, []>("op_2004_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_2004_end_0 = const()[name = tensor<string, []>("op_2004_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_2004_end_mask_0 = const()[name = tensor<string, []>("op_2004_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2004_cast = slice_by_index(begin = var_2004_begin_0, end = var_2004_end_0, end_mask = var_2004_end_mask_0, x = transpose_24);
            tensor<int32, [4]> var_2008_begin_0 = const()[name = tensor<string, []>("op_2008_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_2008_end_0 = const()[name = tensor<string, []>("op_2008_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_2008_end_mask_0 = const()[name = tensor<string, []>("op_2008_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2008_cast = slice_by_index(begin = var_2008_begin_0, end = var_2008_end_0, end_mask = var_2008_end_mask_0, x = transpose_24);
            tensor<int32, [4]> var_2012_begin_0 = const()[name = tensor<string, []>("op_2012_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_2012_end_0 = const()[name = tensor<string, []>("op_2012_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_2012_end_mask_0 = const()[name = tensor<string, []>("op_2012_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2012_cast = slice_by_index(begin = var_2012_begin_0, end = var_2012_end_0, end_mask = var_2012_end_mask_0, x = transpose_24);
            tensor<int32, [4]> var_2016_begin_0 = const()[name = tensor<string, []>("op_2016_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_2016_end_0 = const()[name = tensor<string, []>("op_2016_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_2016_end_mask_0 = const()[name = tensor<string, []>("op_2016_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2016_cast = slice_by_index(begin = var_2016_begin_0, end = var_2016_end_0, end_mask = var_2016_end_mask_0, x = transpose_24);
            tensor<int32, [4]> var_2020_begin_0 = const()[name = tensor<string, []>("op_2020_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_2020_end_0 = const()[name = tensor<string, []>("op_2020_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_2020_end_mask_0 = const()[name = tensor<string, []>("op_2020_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2020_cast = slice_by_index(begin = var_2020_begin_0, end = var_2020_end_0, end_mask = var_2020_end_mask_0, x = transpose_24);
            tensor<int32, [4]> var_2024_begin_0 = const()[name = tensor<string, []>("op_2024_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_2024_end_0 = const()[name = tensor<string, []>("op_2024_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_2024_end_mask_0 = const()[name = tensor<string, []>("op_2024_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2024_cast = slice_by_index(begin = var_2024_begin_0, end = var_2024_end_0, end_mask = var_2024_end_mask_0, x = transpose_24);
            tensor<int32, [4]> var_2028_begin_0 = const()[name = tensor<string, []>("op_2028_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_2028_end_0 = const()[name = tensor<string, []>("op_2028_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_2028_end_mask_0 = const()[name = tensor<string, []>("op_2028_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2028_cast = slice_by_index(begin = var_2028_begin_0, end = var_2028_end_0, end_mask = var_2028_end_mask_0, x = transpose_24);
            tensor<int32, [4]> var_2032_begin_0 = const()[name = tensor<string, []>("op_2032_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2032_end_0 = const()[name = tensor<string, []>("op_2032_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_2032_end_mask_0 = const()[name = tensor<string, []>("op_2032_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2032_cast = slice_by_index(begin = var_2032_begin_0, end = var_2032_end_0, end_mask = var_2032_end_mask_0, x = transpose_24);
            tensor<int32, [4]> var_2036_begin_0 = const()[name = tensor<string, []>("op_2036_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_2036_end_0 = const()[name = tensor<string, []>("op_2036_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_2036_end_mask_0 = const()[name = tensor<string, []>("op_2036_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2036_cast = slice_by_index(begin = var_2036_begin_0, end = var_2036_end_0, end_mask = var_2036_end_mask_0, x = transpose_24);
            tensor<int32, [4]> var_2038_begin_0 = const()[name = tensor<string, []>("op_2038_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2038_end_0 = const()[name = tensor<string, []>("op_2038_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_2038_end_mask_0 = const()[name = tensor<string, []>("op_2038_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2038_cast = slice_by_index(begin = var_2038_begin_0, end = var_2038_end_0, end_mask = var_2038_end_mask_0, x = v_15_cast);
            tensor<int32, [4]> var_2042_begin_0 = const()[name = tensor<string, []>("op_2042_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_2042_end_0 = const()[name = tensor<string, []>("op_2042_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_2042_end_mask_0 = const()[name = tensor<string, []>("op_2042_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2042_cast = slice_by_index(begin = var_2042_begin_0, end = var_2042_end_0, end_mask = var_2042_end_mask_0, x = v_15_cast);
            tensor<int32, [4]> var_2046_begin_0 = const()[name = tensor<string, []>("op_2046_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_2046_end_0 = const()[name = tensor<string, []>("op_2046_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_2046_end_mask_0 = const()[name = tensor<string, []>("op_2046_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2046_cast = slice_by_index(begin = var_2046_begin_0, end = var_2046_end_0, end_mask = var_2046_end_mask_0, x = v_15_cast);
            tensor<int32, [4]> var_2050_begin_0 = const()[name = tensor<string, []>("op_2050_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_2050_end_0 = const()[name = tensor<string, []>("op_2050_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_2050_end_mask_0 = const()[name = tensor<string, []>("op_2050_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2050_cast = slice_by_index(begin = var_2050_begin_0, end = var_2050_end_0, end_mask = var_2050_end_mask_0, x = v_15_cast);
            tensor<int32, [4]> var_2054_begin_0 = const()[name = tensor<string, []>("op_2054_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_2054_end_0 = const()[name = tensor<string, []>("op_2054_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_2054_end_mask_0 = const()[name = tensor<string, []>("op_2054_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2054_cast = slice_by_index(begin = var_2054_begin_0, end = var_2054_end_0, end_mask = var_2054_end_mask_0, x = v_15_cast);
            tensor<int32, [4]> var_2058_begin_0 = const()[name = tensor<string, []>("op_2058_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_2058_end_0 = const()[name = tensor<string, []>("op_2058_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_2058_end_mask_0 = const()[name = tensor<string, []>("op_2058_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2058_cast = slice_by_index(begin = var_2058_begin_0, end = var_2058_end_0, end_mask = var_2058_end_mask_0, x = v_15_cast);
            tensor<int32, [4]> var_2062_begin_0 = const()[name = tensor<string, []>("op_2062_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_2062_end_0 = const()[name = tensor<string, []>("op_2062_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_2062_end_mask_0 = const()[name = tensor<string, []>("op_2062_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2062_cast = slice_by_index(begin = var_2062_begin_0, end = var_2062_end_0, end_mask = var_2062_end_mask_0, x = v_15_cast);
            tensor<int32, [4]> var_2066_begin_0 = const()[name = tensor<string, []>("op_2066_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_2066_end_0 = const()[name = tensor<string, []>("op_2066_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_2066_end_mask_0 = const()[name = tensor<string, []>("op_2066_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2066_cast = slice_by_index(begin = var_2066_begin_0, end = var_2066_end_0, end_mask = var_2066_end_mask_0, x = v_15_cast);
            tensor<int32, [4]> var_2070_begin_0 = const()[name = tensor<string, []>("op_2070_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_2070_end_0 = const()[name = tensor<string, []>("op_2070_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_2070_end_mask_0 = const()[name = tensor<string, []>("op_2070_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2070_cast = slice_by_index(begin = var_2070_begin_0, end = var_2070_end_0, end_mask = var_2070_end_mask_0, x = v_15_cast);
            tensor<int32, [4]> var_2074_begin_0 = const()[name = tensor<string, []>("op_2074_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_2074_end_0 = const()[name = tensor<string, []>("op_2074_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_2074_end_mask_0 = const()[name = tensor<string, []>("op_2074_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2074_cast = slice_by_index(begin = var_2074_begin_0, end = var_2074_end_0, end_mask = var_2074_end_mask_0, x = v_15_cast);
            tensor<string, []> var_2078_equation_0 = const()[name = tensor<string, []>("op_2078_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_2078_cast = einsum(equation = var_2078_equation_0, values = (var_2000_cast, var_1957_cast));
            tensor<fp16, []> var_2079_to_fp16 = const()[name = tensor<string, []>("op_2079_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_101_cast = mul(x = var_2078_cast, y = var_2079_to_fp16);
            tensor<string, []> var_2082_equation_0 = const()[name = tensor<string, []>("op_2082_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_2082_cast = einsum(equation = var_2082_equation_0, values = (var_2004_cast, var_1961_cast));
            tensor<fp16, []> var_2083_to_fp16 = const()[name = tensor<string, []>("op_2083_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_103_cast = mul(x = var_2082_cast, y = var_2083_to_fp16);
            tensor<string, []> var_2086_equation_0 = const()[name = tensor<string, []>("op_2086_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_2086_cast = einsum(equation = var_2086_equation_0, values = (var_2008_cast, var_1965_cast));
            tensor<fp16, []> var_2087_to_fp16 = const()[name = tensor<string, []>("op_2087_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_105_cast = mul(x = var_2086_cast, y = var_2087_to_fp16);
            tensor<string, []> var_2090_equation_0 = const()[name = tensor<string, []>("op_2090_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_2090_cast = einsum(equation = var_2090_equation_0, values = (var_2012_cast, var_1969_cast));
            tensor<fp16, []> var_2091_to_fp16 = const()[name = tensor<string, []>("op_2091_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_107_cast = mul(x = var_2090_cast, y = var_2091_to_fp16);
            tensor<string, []> var_2094_equation_0 = const()[name = tensor<string, []>("op_2094_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_2094_cast = einsum(equation = var_2094_equation_0, values = (var_2016_cast, var_1973_cast));
            tensor<fp16, []> var_2095_to_fp16 = const()[name = tensor<string, []>("op_2095_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_109_cast = mul(x = var_2094_cast, y = var_2095_to_fp16);
            tensor<string, []> var_2098_equation_0 = const()[name = tensor<string, []>("op_2098_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_2098_cast = einsum(equation = var_2098_equation_0, values = (var_2020_cast, var_1977_cast));
            tensor<fp16, []> var_2099_to_fp16 = const()[name = tensor<string, []>("op_2099_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_111_cast = mul(x = var_2098_cast, y = var_2099_to_fp16);
            tensor<string, []> var_2102_equation_0 = const()[name = tensor<string, []>("op_2102_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_2102_cast = einsum(equation = var_2102_equation_0, values = (var_2024_cast, var_1981_cast));
            tensor<fp16, []> var_2103_to_fp16 = const()[name = tensor<string, []>("op_2103_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_113_cast = mul(x = var_2102_cast, y = var_2103_to_fp16);
            tensor<string, []> var_2106_equation_0 = const()[name = tensor<string, []>("op_2106_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_2106_cast = einsum(equation = var_2106_equation_0, values = (var_2028_cast, var_1985_cast));
            tensor<fp16, []> var_2107_to_fp16 = const()[name = tensor<string, []>("op_2107_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_115_cast = mul(x = var_2106_cast, y = var_2107_to_fp16);
            tensor<string, []> var_2110_equation_0 = const()[name = tensor<string, []>("op_2110_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_2110_cast = einsum(equation = var_2110_equation_0, values = (var_2032_cast, var_1989_cast));
            tensor<fp16, []> var_2111_to_fp16 = const()[name = tensor<string, []>("op_2111_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_117_cast = mul(x = var_2110_cast, y = var_2111_to_fp16);
            tensor<string, []> var_2114_equation_0 = const()[name = tensor<string, []>("op_2114_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_2114_cast = einsum(equation = var_2114_equation_0, values = (var_2036_cast, var_1993_cast));
            tensor<fp16, []> var_2115_to_fp16 = const()[name = tensor<string, []>("op_2115_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_119_cast = mul(x = var_2114_cast, y = var_2115_to_fp16);
            tensor<fp16, [2, 77, 1, 1024]> var_2117_cast = softmax(axis = var_984, x = aw_101_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_2118_cast = softmax(axis = var_984, x = aw_103_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_2119_cast = softmax(axis = var_984, x = aw_105_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_2120_cast = softmax(axis = var_984, x = aw_107_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_2121_cast = softmax(axis = var_984, x = aw_109_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_2122_cast = softmax(axis = var_984, x = aw_111_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_2123_cast = softmax(axis = var_984, x = aw_113_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_2124_cast = softmax(axis = var_984, x = aw_115_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_2125_cast = softmax(axis = var_984, x = aw_117_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_2126_cast = softmax(axis = var_984, x = aw_119_cast);
            tensor<string, []> var_2128_equation_0 = const()[name = tensor<string, []>("op_2128_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_2128_cast = einsum(equation = var_2128_equation_0, values = (var_2038_cast, var_2117_cast));
            tensor<string, []> var_2130_equation_0 = const()[name = tensor<string, []>("op_2130_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_2130_cast = einsum(equation = var_2130_equation_0, values = (var_2042_cast, var_2118_cast));
            tensor<string, []> var_2132_equation_0 = const()[name = tensor<string, []>("op_2132_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_2132_cast = einsum(equation = var_2132_equation_0, values = (var_2046_cast, var_2119_cast));
            tensor<string, []> var_2134_equation_0 = const()[name = tensor<string, []>("op_2134_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_2134_cast = einsum(equation = var_2134_equation_0, values = (var_2050_cast, var_2120_cast));
            tensor<string, []> var_2136_equation_0 = const()[name = tensor<string, []>("op_2136_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_2136_cast = einsum(equation = var_2136_equation_0, values = (var_2054_cast, var_2121_cast));
            tensor<string, []> var_2138_equation_0 = const()[name = tensor<string, []>("op_2138_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_2138_cast = einsum(equation = var_2138_equation_0, values = (var_2058_cast, var_2122_cast));
            tensor<string, []> var_2140_equation_0 = const()[name = tensor<string, []>("op_2140_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_2140_cast = einsum(equation = var_2140_equation_0, values = (var_2062_cast, var_2123_cast));
            tensor<string, []> var_2142_equation_0 = const()[name = tensor<string, []>("op_2142_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_2142_cast = einsum(equation = var_2142_equation_0, values = (var_2066_cast, var_2124_cast));
            tensor<string, []> var_2144_equation_0 = const()[name = tensor<string, []>("op_2144_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_2144_cast = einsum(equation = var_2144_equation_0, values = (var_2070_cast, var_2125_cast));
            tensor<string, []> var_2146_equation_0 = const()[name = tensor<string, []>("op_2146_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_2146_cast = einsum(equation = var_2146_equation_0, values = (var_2074_cast, var_2126_cast));
            tensor<bool, []> input_107_interleave_0 = const()[name = tensor<string, []>("input_107_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 1024]> input_107_cast = concat(axis = var_984, interleave = input_107_interleave_0, values = (var_2128_cast, var_2130_cast, var_2132_cast, var_2134_cast, var_2136_cast, var_2138_cast, var_2140_cast, var_2142_cast, var_2144_cast, var_2146_cast));
            tensor<int32, [2]> var_2152 = const()[name = tensor<string, []>("op_2152"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2154 = const()[name = tensor<string, []>("op_2154"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_2156_pad_type_0 = const()[name = tensor<string, []>("op_2156_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_2156_pad_0 = const()[name = tensor<string, []>("op_2156_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81950400)))];
            tensor<fp16, [640]> down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82769664)))];
            tensor<fp16, [2, 640, 1, 1024]> var_2156_cast = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_2154, groups = var_984, pad = var_2156_pad_0, pad_type = var_2156_pad_type_0, strides = var_2152, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_107_cast);
            tensor<fp16, [2, 640, 1, 1024]> inputs_23_cast = add(x = var_2156_cast, y = inputs_21_cast);
            tensor<int32, [1]> var_2160 = const()[name = tensor<string, []>("op_2160"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_23_cast = reduce_mean(axes = var_2160, keep_dims = var_979, x = inputs_23_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_23_cast = sub(x = inputs_23_cast, y = channels_mean_23_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_sq_23_cast = mul(x = zero_mean_23_cast, y = zero_mean_23_cast);
            tensor<int32, [1]> var_2164 = const()[name = tensor<string, []>("op_2164"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_2165_cast = reduce_mean(axes = var_2164, keep_dims = var_979, x = zero_mean_sq_23_cast);
            tensor<fp16, []> var_2166_to_fp16 = const()[name = tensor<string, []>("op_2166_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_2167_cast = add(x = var_2165_cast, y = var_2166_to_fp16);
            tensor<fp16, []> denom_23_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_23_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_23_cast = rsqrt(epsilon = denom_23_epsilon_0_to_fp16, x = var_2167_cast);
            tensor<fp16, [2, 640, 1, 1024]> out_23_cast = mul(x = zero_mean_23_cast, y = denom_23_cast);
            tensor<fp16, [1, 640, 1, 1]> var_2171_to_fp16 = const()[name = tensor<string, []>("op_2171_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82771008)))];
            tensor<fp16, [2, 640, 1, 1024]> var_2172_cast = add(x = out_23_cast, y = var_2171_to_fp16);
            tensor<fp16, [1, 640, 1, 1]> var_2174_to_fp16 = const()[name = tensor<string, []>("op_2174_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82772352)))];
            tensor<fp16, [2, 640, 1, 1024]> input_109_cast = mul(x = var_2172_cast, y = var_2174_to_fp16);
            tensor<int32, [2]> var_2182 = const()[name = tensor<string, []>("op_2182"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2184 = const()[name = tensor<string, []>("op_2184"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_2186_pad_type_0 = const()[name = tensor<string, []>("op_2186_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_2186_pad_0 = const()[name = tensor<string, []>("op_2186_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [5120, 640, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [5120, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82773696)))];
            tensor<fp16, [5120]> down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89327360)))];
            tensor<fp16, [2, 5120, 1, 1024]> var_2186_cast = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_2184, groups = var_984, pad = var_2186_pad_0, pad_type = var_2186_pad_type_0, strides = var_2182, weight = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_109_cast);
            tensor<int32, [2]> var_2187_split_sizes_0 = const()[name = tensor<string, []>("op_2187_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_2187_axis_0 = const()[name = tensor<string, []>("op_2187_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 1024]> var_2187_cast_0, tensor<fp16, [2, 2560, 1, 1024]> var_2187_cast_1 = split(axis = var_2187_axis_0, split_sizes = var_2187_split_sizes_0, x = var_2186_cast);
            tensor<string, []> var_2189_mode_0 = const()[name = tensor<string, []>("op_2189_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 1024]> var_2189_cast = gelu(mode = var_2189_mode_0, x = var_2187_cast_1);
            tensor<fp16, [2, 2560, 1, 1024]> input_111_cast = mul(x = var_2187_cast_0, y = var_2189_cast);
            tensor<int32, [2]> var_2193 = const()[name = tensor<string, []>("op_2193"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2195 = const()[name = tensor<string, []>("op_2195"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_2197_pad_type_0 = const()[name = tensor<string, []>("op_2197_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_2197_pad_0 = const()[name = tensor<string, []>("op_2197_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2560, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [640, 2560, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89337664)))];
            tensor<fp16, [640]> down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92614528)))];
            tensor<fp16, [2, 640, 1, 1024]> var_2197_cast = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_2195, groups = var_984, pad = var_2197_pad_0, pad_type = var_2197_pad_type_0, strides = var_2193, weight = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_111_cast);
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_69_cast = add(x = var_2197_cast, y = inputs_23_cast);
            tensor<int32, [4]> var_2199 = const()[name = tensor<string, []>("op_2199"), val = tensor<int32, [4]>([2, 640, 32, 32])];
            tensor<fp16, [2, 640, 32, 32]> input_113_cast = reshape(shape = var_2199, x = hidden_states_69_cast);
            tensor<int32, [2]> var_2203 = const()[name = tensor<string, []>("op_2203"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2205 = const()[name = tensor<string, []>("op_2205"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_71_pad_type_0 = const()[name = tensor<string, []>("hidden_states_71_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_71_pad_0 = const()[name = tensor<string, []>("hidden_states_71_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> down_blocks_1_attentions_1_proj_out_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_proj_out_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92615872)))];
            tensor<fp16, [640]> down_blocks_1_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_attentions_1_proj_out_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93435136)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_71_cast = conv(bias = down_blocks_1_attentions_1_proj_out_bias_to_fp16, dilations = var_2205, groups = var_984, pad = hidden_states_71_pad_0, pad_type = hidden_states_71_pad_type_0, strides = var_2203, weight = down_blocks_1_attentions_1_proj_out_weight_to_fp16, x = input_113_cast);
            tensor<fp16, [2, 640, 32, 32]> input_115_cast = add(x = hidden_states_71_cast, y = hidden_states_59_cast);
            tensor<int32, [2]> var_2212 = const()[name = tensor<string, []>("op_2212"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [2]> var_2214 = const()[name = tensor<string, []>("op_2214"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> input_117_pad_type_0 = const()[name = tensor<string, []>("input_117_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> input_117_pad_0 = const()[name = tensor<string, []>("input_117_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [640, 640, 3, 3]> down_blocks_1_downsamplers_0_conv_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_downsamplers_0_conv_weight_to_fp16"), val = tensor<fp16, [640, 640, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93436480)))];
            tensor<fp16, [640]> down_blocks_1_downsamplers_0_conv_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_1_downsamplers_0_conv_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100809344)))];
            tensor<fp16, [2, 640, 16, 16]> input_117_cast = conv(bias = down_blocks_1_downsamplers_0_conv_bias_to_fp16, dilations = var_2214, groups = var_984, pad = input_117_pad_0, pad_type = input_117_pad_type_0, strides = var_2212, weight = down_blocks_1_downsamplers_0_conv_weight_to_fp16, x = input_115_cast);
            tensor<bool, []> var_2249 = const()[name = tensor<string, []>("op_2249"), val = tensor<bool, []>(true)];
            tensor<int32, []> var_2254 = const()[name = tensor<string, []>("op_2254"), val = tensor<int32, []>(1)];
            tensor<int32, [5]> reshape_48_shape_0 = const()[name = tensor<string, []>("reshape_48_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 16, 16])];
            tensor<fp16, [2, 32, 20, 16, 16]> reshape_48_cast = reshape(shape = reshape_48_shape_0, x = input_117_cast);
            tensor<int32, [3]> reduce_mean_36_axes_0 = const()[name = tensor<string, []>("reduce_mean_36_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_36_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_36_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_36_cast = reduce_mean(axes = reduce_mean_36_axes_0, keep_dims = reduce_mean_36_keep_dims_0, x = reshape_48_cast);
            tensor<fp16, [2, 32, 20, 16, 16]> sub_24_cast = sub(x = reshape_48_cast, y = reduce_mean_36_cast);
            tensor<fp16, [2, 32, 20, 16, 16]> square_12_cast = square(x = sub_24_cast);
            tensor<int32, [3]> reduce_mean_38_axes_0 = const()[name = tensor<string, []>("reduce_mean_38_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_38_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_38_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_38_cast = reduce_mean(axes = reduce_mean_38_axes_0, keep_dims = reduce_mean_38_keep_dims_0, x = square_12_cast);
            tensor<fp16, []> add_24_y_0_to_fp16 = const()[name = tensor<string, []>("add_24_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_24_cast = add(x = reduce_mean_38_cast, y = add_24_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_12_cast = sqrt(x = add_24_cast);
            tensor<fp16, [2, 32, 20, 16, 16]> real_div_12_cast = real_div(x = sub_24_cast, y = sqrt_12_cast);
            tensor<int32, [4]> reshape_49_shape_0 = const()[name = tensor<string, []>("reshape_49_shape_0"), val = tensor<int32, [4]>([2, 640, 16, 16])];
            tensor<fp16, [2, 640, 16, 16]> reshape_49_cast = reshape(shape = reshape_49_shape_0, x = real_div_12_cast);
            tensor<fp16, [640]> add_25_mean_0_to_fp16 = const()[name = tensor<string, []>("add_25_mean_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100810688)))];
            tensor<fp16, [640]> add_25_variance_0_to_fp16 = const()[name = tensor<string, []>("add_25_variance_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100812032)))];
            tensor<fp16, [640]> add_25_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_25_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100813376)))];
            tensor<fp16, [640]> add_25_beta_0_to_fp16 = const()[name = tensor<string, []>("add_25_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100814720)))];
            tensor<fp16, []> add_25_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_25_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 16, 16]> add_25_cast = batch_norm(beta = add_25_beta_0_to_fp16, epsilon = add_25_epsilon_0_to_fp16, gamma = add_25_gamma_0_to_fp16, mean = add_25_mean_0_to_fp16, variance = add_25_variance_0_to_fp16, x = reshape_49_cast);
            tensor<fp16, [2, 640, 16, 16]> input_121_cast = silu(x = add_25_cast);
            tensor<int32, [2]> var_2277 = const()[name = tensor<string, []>("op_2277"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2279 = const()[name = tensor<string, []>("op_2279"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_73_pad_type_0 = const()[name = tensor<string, []>("hidden_states_73_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_73_pad_0 = const()[name = tensor<string, []>("hidden_states_73_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 640, 3, 3]> down_blocks_2_resnets_0_conv1_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_resnets_0_conv1_weight_to_fp16"), val = tensor<fp16, [1280, 640, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100816064)))];
            tensor<fp16, [1280]> down_blocks_2_resnets_0_conv1_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_resnets_0_conv1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115561728)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_73_cast = conv(bias = down_blocks_2_resnets_0_conv1_bias_to_fp16, dilations = var_2279, groups = var_2254, pad = hidden_states_73_pad_0, pad_type = hidden_states_73_pad_type_0, strides = var_2277, weight = down_blocks_2_resnets_0_conv1_weight_to_fp16, x = input_121_cast);
            tensor<int32, [2]> var_2285 = const()[name = tensor<string, []>("op_2285"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2287 = const()[name = tensor<string, []>("op_2287"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_9_pad_type_0 = const()[name = tensor<string, []>("temb_9_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_9_pad_0 = const()[name = tensor<string, []>("temb_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115564352)))];
            tensor<fp16, [1280]> down_blocks_2_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118841216)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_9_cast = conv(bias = down_blocks_2_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_2287, groups = var_2254, pad = temb_9_pad_0, pad_type = temb_9_pad_type_0, strides = var_2285, weight = down_blocks_2_resnets_0_time_emb_proj_weight_to_fp16, x = input_15_cast);
            tensor<fp16, [2, 1280, 16, 16]> input_125_cast = add(x = hidden_states_73_cast, y = temb_9_cast);
            tensor<int32, [5]> reshape_52_shape_0 = const()[name = tensor<string, []>("reshape_52_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 16, 16])];
            tensor<fp16, [2, 32, 40, 16, 16]> reshape_52_cast = reshape(shape = reshape_52_shape_0, x = input_125_cast);
            tensor<int32, [3]> reduce_mean_39_axes_0 = const()[name = tensor<string, []>("reduce_mean_39_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_39_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_39_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_39_cast = reduce_mean(axes = reduce_mean_39_axes_0, keep_dims = reduce_mean_39_keep_dims_0, x = reshape_52_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> sub_26_cast = sub(x = reshape_52_cast, y = reduce_mean_39_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> square_13_cast = square(x = sub_26_cast);
            tensor<int32, [3]> reduce_mean_41_axes_0 = const()[name = tensor<string, []>("reduce_mean_41_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_41_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_41_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_41_cast = reduce_mean(axes = reduce_mean_41_axes_0, keep_dims = reduce_mean_41_keep_dims_0, x = square_13_cast);
            tensor<fp16, []> add_26_y_0_to_fp16 = const()[name = tensor<string, []>("add_26_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_26_cast = add(x = reduce_mean_41_cast, y = add_26_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_13_cast = sqrt(x = add_26_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> real_div_13_cast = real_div(x = sub_26_cast, y = sqrt_13_cast);
            tensor<int32, [4]> reshape_53_shape_0 = const()[name = tensor<string, []>("reshape_53_shape_0"), val = tensor<int32, [4]>([2, 1280, 16, 16])];
            tensor<fp16, [2, 1280, 16, 16]> reshape_53_cast = reshape(shape = reshape_53_shape_0, x = real_div_13_cast);
            tensor<fp16, [1280]> add_27_mean_0_to_fp16 = const()[name = tensor<string, []>("add_27_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118843840)))];
            tensor<fp16, [1280]> add_27_variance_0_to_fp16 = const()[name = tensor<string, []>("add_27_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118846464)))];
            tensor<fp16, [1280]> add_27_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_27_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118849088)))];
            tensor<fp16, [1280]> add_27_beta_0_to_fp16 = const()[name = tensor<string, []>("add_27_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118851712)))];
            tensor<fp16, []> add_27_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_27_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 16, 16]> add_27_cast = batch_norm(beta = add_27_beta_0_to_fp16, epsilon = add_27_epsilon_0_to_fp16, gamma = add_27_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_53_cast);
            tensor<fp16, [2, 1280, 16, 16]> input_129_cast = silu(x = add_27_cast);
            tensor<int32, [2]> var_2297 = const()[name = tensor<string, []>("op_2297"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2299 = const()[name = tensor<string, []>("op_2299"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_75_pad_type_0 = const()[name = tensor<string, []>("hidden_states_75_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_75_pad_0 = const()[name = tensor<string, []>("hidden_states_75_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> down_blocks_2_resnets_0_conv2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_resnets_0_conv2_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118854336)))];
            tensor<fp16, [1280]> down_blocks_2_resnets_0_conv2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_resnets_0_conv2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148345600)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_75_cast = conv(bias = down_blocks_2_resnets_0_conv2_bias_to_fp16, dilations = var_2299, groups = var_2254, pad = hidden_states_75_pad_0, pad_type = hidden_states_75_pad_type_0, strides = var_2297, weight = down_blocks_2_resnets_0_conv2_weight_to_fp16, x = input_129_cast);
            tensor<int32, [2]> var_2304 = const()[name = tensor<string, []>("op_2304"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2306 = const()[name = tensor<string, []>("op_2306"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> x_3_pad_type_0 = const()[name = tensor<string, []>("x_3_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> x_3_pad_0 = const()[name = tensor<string, []>("x_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 640, 1, 1]> down_blocks_2_resnets_0_conv_shortcut_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_resnets_0_conv_shortcut_weight_to_fp16"), val = tensor<fp16, [1280, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148348224)))];
            tensor<fp16, [1280]> down_blocks_2_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149986688)))];
            tensor<fp16, [2, 1280, 16, 16]> x_3_cast = conv(bias = down_blocks_2_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_2306, groups = var_2254, pad = x_3_pad_0, pad_type = x_3_pad_type_0, strides = var_2304, weight = down_blocks_2_resnets_0_conv_shortcut_weight_to_fp16, x = input_117_cast);
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_77_cast = add(x = x_3_cast, y = hidden_states_75_cast);
            tensor<int32, [5]> reshape_56_shape_0 = const()[name = tensor<string, []>("reshape_56_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 16, 16])];
            tensor<fp16, [2, 32, 40, 16, 16]> reshape_56_cast = reshape(shape = reshape_56_shape_0, x = hidden_states_77_cast);
            tensor<int32, [3]> reduce_mean_42_axes_0 = const()[name = tensor<string, []>("reduce_mean_42_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_42_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_42_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_42_cast = reduce_mean(axes = reduce_mean_42_axes_0, keep_dims = reduce_mean_42_keep_dims_0, x = reshape_56_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> sub_28_cast = sub(x = reshape_56_cast, y = reduce_mean_42_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> square_14_cast = square(x = sub_28_cast);
            tensor<int32, [3]> reduce_mean_44_axes_0 = const()[name = tensor<string, []>("reduce_mean_44_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_44_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_44_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_44_cast = reduce_mean(axes = reduce_mean_44_axes_0, keep_dims = reduce_mean_44_keep_dims_0, x = square_14_cast);
            tensor<fp16, []> add_28_y_0_to_fp16 = const()[name = tensor<string, []>("add_28_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_28_cast = add(x = reduce_mean_44_cast, y = add_28_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_14_cast = sqrt(x = add_28_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> real_div_14_cast = real_div(x = sub_28_cast, y = sqrt_14_cast);
            tensor<int32, [4]> reshape_57_shape_0 = const()[name = tensor<string, []>("reshape_57_shape_0"), val = tensor<int32, [4]>([2, 1280, 16, 16])];
            tensor<fp16, [2, 1280, 16, 16]> reshape_57_cast = reshape(shape = reshape_57_shape_0, x = real_div_14_cast);
            tensor<fp16, [1280]> add_29_mean_0_to_fp16 = const()[name = tensor<string, []>("add_29_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149989312)))];
            tensor<fp16, [1280]> add_29_variance_0_to_fp16 = const()[name = tensor<string, []>("add_29_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149991936)))];
            tensor<fp16, [1280]> add_29_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_29_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149994560)))];
            tensor<fp16, [1280]> add_29_beta_0_to_fp16 = const()[name = tensor<string, []>("add_29_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149997184)))];
            tensor<fp16, []> add_29_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_29_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 16, 16]> add_29_cast = batch_norm(beta = add_29_beta_0_to_fp16, epsilon = add_29_epsilon_0_to_fp16, gamma = add_29_gamma_0_to_fp16, mean = add_29_mean_0_to_fp16, variance = add_29_variance_0_to_fp16, x = reshape_57_cast);
            tensor<int32, [2]> var_2326 = const()[name = tensor<string, []>("op_2326"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2328 = const()[name = tensor<string, []>("op_2328"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_79_pad_type_0 = const()[name = tensor<string, []>("hidden_states_79_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_79_pad_0 = const()[name = tensor<string, []>("hidden_states_79_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_proj_in_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_proj_in_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149999808)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_proj_in_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153276672)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_79_cast = conv(bias = down_blocks_2_attentions_0_proj_in_bias_to_fp16, dilations = var_2328, groups = var_2254, pad = hidden_states_79_pad_0, pad_type = hidden_states_79_pad_type_0, strides = var_2326, weight = down_blocks_2_attentions_0_proj_in_weight_to_fp16, x = add_29_cast);
            tensor<int32, [4]> var_2333 = const()[name = tensor<string, []>("op_2333"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<fp16, [2, 1280, 1, 256]> inputs_25_cast = reshape(shape = var_2333, x = hidden_states_79_cast);
            tensor<int32, [1]> var_2343 = const()[name = tensor<string, []>("op_2343"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_25_cast = reduce_mean(axes = var_2343, keep_dims = var_2249, x = inputs_25_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_25_cast = sub(x = inputs_25_cast, y = channels_mean_25_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_sq_25_cast = mul(x = zero_mean_25_cast, y = zero_mean_25_cast);
            tensor<int32, [1]> var_2347 = const()[name = tensor<string, []>("op_2347"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_2348_cast = reduce_mean(axes = var_2347, keep_dims = var_2249, x = zero_mean_sq_25_cast);
            tensor<fp16, []> var_2349_to_fp16 = const()[name = tensor<string, []>("op_2349_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_2350_cast = add(x = var_2348_cast, y = var_2349_to_fp16);
            tensor<fp16, []> denom_25_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_25_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 256]> denom_25_cast = rsqrt(epsilon = denom_25_epsilon_0_to_fp16, x = var_2350_cast);
            tensor<fp16, [2, 1280, 1, 256]> out_25_cast = mul(x = zero_mean_25_cast, y = denom_25_cast);
            tensor<fp16, [1, 1280, 1, 1]> var_2354_to_fp16 = const()[name = tensor<string, []>("op_2354_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153279296)))];
            tensor<fp16, [2, 1280, 1, 256]> var_2355_cast = add(x = out_25_cast, y = var_2354_to_fp16);
            tensor<fp16, [1, 1280, 1, 1]> var_2357_to_fp16 = const()[name = tensor<string, []>("op_2357_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153281920)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_81_cast = mul(x = var_2355_cast, y = var_2357_to_fp16);
            tensor<int32, [2]> var_2364 = const()[name = tensor<string, []>("op_2364"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2366 = const()[name = tensor<string, []>("op_2366"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_17_pad_type_0 = const()[name = tensor<string, []>("q_17_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_17_pad_0 = const()[name = tensor<string, []>("q_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153284544)))];
            tensor<fp16, [2, 1280, 1, 256]> q_17_cast = conv(dilations = var_2366, groups = var_2254, pad = q_17_pad_0, pad_type = q_17_pad_type_0, strides = var_2364, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_81_cast);
            tensor<int32, [2]> var_2370 = const()[name = tensor<string, []>("op_2370"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2372 = const()[name = tensor<string, []>("op_2372"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_33_pad_type_0 = const()[name = tensor<string, []>("k_33_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_33_pad_0 = const()[name = tensor<string, []>("k_33_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(156561408)))];
            tensor<fp16, [2, 1280, 1, 256]> k_33_cast = conv(dilations = var_2372, groups = var_2254, pad = k_33_pad_0, pad_type = k_33_pad_type_0, strides = var_2370, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_81_cast);
            tensor<int32, [2]> var_2376 = const()[name = tensor<string, []>("op_2376"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2378 = const()[name = tensor<string, []>("op_2378"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_17_pad_type_0 = const()[name = tensor<string, []>("v_17_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_17_pad_0 = const()[name = tensor<string, []>("v_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(159838272)))];
            tensor<fp16, [2, 1280, 1, 256]> v_17_cast = conv(dilations = var_2378, groups = var_2254, pad = v_17_pad_0, pad_type = v_17_pad_type_0, strides = var_2376, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_81_cast);
            tensor<int32, [4]> var_2382_begin_0 = const()[name = tensor<string, []>("op_2382_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2382_end_0 = const()[name = tensor<string, []>("op_2382_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_2382_end_mask_0 = const()[name = tensor<string, []>("op_2382_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2382_cast = slice_by_index(begin = var_2382_begin_0, end = var_2382_end_0, end_mask = var_2382_end_mask_0, x = q_17_cast);
            tensor<int32, [4]> var_2386_begin_0 = const()[name = tensor<string, []>("op_2386_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_2386_end_0 = const()[name = tensor<string, []>("op_2386_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_2386_end_mask_0 = const()[name = tensor<string, []>("op_2386_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2386_cast = slice_by_index(begin = var_2386_begin_0, end = var_2386_end_0, end_mask = var_2386_end_mask_0, x = q_17_cast);
            tensor<int32, [4]> var_2390_begin_0 = const()[name = tensor<string, []>("op_2390_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_2390_end_0 = const()[name = tensor<string, []>("op_2390_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_2390_end_mask_0 = const()[name = tensor<string, []>("op_2390_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2390_cast = slice_by_index(begin = var_2390_begin_0, end = var_2390_end_0, end_mask = var_2390_end_mask_0, x = q_17_cast);
            tensor<int32, [4]> var_2394_begin_0 = const()[name = tensor<string, []>("op_2394_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_2394_end_0 = const()[name = tensor<string, []>("op_2394_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_2394_end_mask_0 = const()[name = tensor<string, []>("op_2394_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2394_cast = slice_by_index(begin = var_2394_begin_0, end = var_2394_end_0, end_mask = var_2394_end_mask_0, x = q_17_cast);
            tensor<int32, [4]> var_2398_begin_0 = const()[name = tensor<string, []>("op_2398_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_2398_end_0 = const()[name = tensor<string, []>("op_2398_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_2398_end_mask_0 = const()[name = tensor<string, []>("op_2398_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2398_cast = slice_by_index(begin = var_2398_begin_0, end = var_2398_end_0, end_mask = var_2398_end_mask_0, x = q_17_cast);
            tensor<int32, [4]> var_2402_begin_0 = const()[name = tensor<string, []>("op_2402_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_2402_end_0 = const()[name = tensor<string, []>("op_2402_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_2402_end_mask_0 = const()[name = tensor<string, []>("op_2402_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2402_cast = slice_by_index(begin = var_2402_begin_0, end = var_2402_end_0, end_mask = var_2402_end_mask_0, x = q_17_cast);
            tensor<int32, [4]> var_2406_begin_0 = const()[name = tensor<string, []>("op_2406_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_2406_end_0 = const()[name = tensor<string, []>("op_2406_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_2406_end_mask_0 = const()[name = tensor<string, []>("op_2406_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2406_cast = slice_by_index(begin = var_2406_begin_0, end = var_2406_end_0, end_mask = var_2406_end_mask_0, x = q_17_cast);
            tensor<int32, [4]> var_2410_begin_0 = const()[name = tensor<string, []>("op_2410_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_2410_end_0 = const()[name = tensor<string, []>("op_2410_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_2410_end_mask_0 = const()[name = tensor<string, []>("op_2410_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2410_cast = slice_by_index(begin = var_2410_begin_0, end = var_2410_end_0, end_mask = var_2410_end_mask_0, x = q_17_cast);
            tensor<int32, [4]> var_2414_begin_0 = const()[name = tensor<string, []>("op_2414_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_2414_end_0 = const()[name = tensor<string, []>("op_2414_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_2414_end_mask_0 = const()[name = tensor<string, []>("op_2414_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2414_cast = slice_by_index(begin = var_2414_begin_0, end = var_2414_end_0, end_mask = var_2414_end_mask_0, x = q_17_cast);
            tensor<int32, [4]> var_2418_begin_0 = const()[name = tensor<string, []>("op_2418_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_2418_end_0 = const()[name = tensor<string, []>("op_2418_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_2418_end_mask_0 = const()[name = tensor<string, []>("op_2418_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2418_cast = slice_by_index(begin = var_2418_begin_0, end = var_2418_end_0, end_mask = var_2418_end_mask_0, x = q_17_cast);
            tensor<int32, [4]> var_2422_begin_0 = const()[name = tensor<string, []>("op_2422_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_2422_end_0 = const()[name = tensor<string, []>("op_2422_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_2422_end_mask_0 = const()[name = tensor<string, []>("op_2422_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2422_cast = slice_by_index(begin = var_2422_begin_0, end = var_2422_end_0, end_mask = var_2422_end_mask_0, x = q_17_cast);
            tensor<int32, [4]> var_2426_begin_0 = const()[name = tensor<string, []>("op_2426_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_2426_end_0 = const()[name = tensor<string, []>("op_2426_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_2426_end_mask_0 = const()[name = tensor<string, []>("op_2426_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2426_cast = slice_by_index(begin = var_2426_begin_0, end = var_2426_end_0, end_mask = var_2426_end_mask_0, x = q_17_cast);
            tensor<int32, [4]> var_2430_begin_0 = const()[name = tensor<string, []>("op_2430_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_2430_end_0 = const()[name = tensor<string, []>("op_2430_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_2430_end_mask_0 = const()[name = tensor<string, []>("op_2430_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2430_cast = slice_by_index(begin = var_2430_begin_0, end = var_2430_end_0, end_mask = var_2430_end_mask_0, x = q_17_cast);
            tensor<int32, [4]> var_2434_begin_0 = const()[name = tensor<string, []>("op_2434_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_2434_end_0 = const()[name = tensor<string, []>("op_2434_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_2434_end_mask_0 = const()[name = tensor<string, []>("op_2434_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2434_cast = slice_by_index(begin = var_2434_begin_0, end = var_2434_end_0, end_mask = var_2434_end_mask_0, x = q_17_cast);
            tensor<int32, [4]> var_2438_begin_0 = const()[name = tensor<string, []>("op_2438_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_2438_end_0 = const()[name = tensor<string, []>("op_2438_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_2438_end_mask_0 = const()[name = tensor<string, []>("op_2438_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2438_cast = slice_by_index(begin = var_2438_begin_0, end = var_2438_end_0, end_mask = var_2438_end_mask_0, x = q_17_cast);
            tensor<int32, [4]> var_2442_begin_0 = const()[name = tensor<string, []>("op_2442_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_2442_end_0 = const()[name = tensor<string, []>("op_2442_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_2442_end_mask_0 = const()[name = tensor<string, []>("op_2442_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2442_cast = slice_by_index(begin = var_2442_begin_0, end = var_2442_end_0, end_mask = var_2442_end_mask_0, x = q_17_cast);
            tensor<int32, [4]> var_2446_begin_0 = const()[name = tensor<string, []>("op_2446_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_2446_end_0 = const()[name = tensor<string, []>("op_2446_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_2446_end_mask_0 = const()[name = tensor<string, []>("op_2446_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2446_cast = slice_by_index(begin = var_2446_begin_0, end = var_2446_end_0, end_mask = var_2446_end_mask_0, x = q_17_cast);
            tensor<int32, [4]> var_2450_begin_0 = const()[name = tensor<string, []>("op_2450_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_2450_end_0 = const()[name = tensor<string, []>("op_2450_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_2450_end_mask_0 = const()[name = tensor<string, []>("op_2450_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2450_cast = slice_by_index(begin = var_2450_begin_0, end = var_2450_end_0, end_mask = var_2450_end_mask_0, x = q_17_cast);
            tensor<int32, [4]> var_2454_begin_0 = const()[name = tensor<string, []>("op_2454_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_2454_end_0 = const()[name = tensor<string, []>("op_2454_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_2454_end_mask_0 = const()[name = tensor<string, []>("op_2454_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2454_cast = slice_by_index(begin = var_2454_begin_0, end = var_2454_end_0, end_mask = var_2454_end_mask_0, x = q_17_cast);
            tensor<int32, [4]> var_2458_begin_0 = const()[name = tensor<string, []>("op_2458_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_2458_end_0 = const()[name = tensor<string, []>("op_2458_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_2458_end_mask_0 = const()[name = tensor<string, []>("op_2458_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2458_cast = slice_by_index(begin = var_2458_begin_0, end = var_2458_end_0, end_mask = var_2458_end_mask_0, x = q_17_cast);
            tensor<int32, [4]> k_35_perm_0 = const()[name = tensor<string, []>("k_35_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_2465_begin_0 = const()[name = tensor<string, []>("op_2465_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2465_end_0 = const()[name = tensor<string, []>("op_2465_end_0"), val = tensor<int32, [4]>([2, 256, 1, 64])];
            tensor<bool, [4]> var_2465_end_mask_0 = const()[name = tensor<string, []>("op_2465_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 1280]> transpose_23 = transpose(perm = k_35_perm_0, x = k_33_cast);
            tensor<fp16, [2, 256, 1, 64]> var_2465_cast = slice_by_index(begin = var_2465_begin_0, end = var_2465_end_0, end_mask = var_2465_end_mask_0, x = transpose_23);
            tensor<int32, [4]> var_2469_begin_0 = const()[name = tensor<string, []>("op_2469_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_2469_end_0 = const()[name = tensor<string, []>("op_2469_end_0"), val = tensor<int32, [4]>([2, 256, 1, 128])];
            tensor<bool, [4]> var_2469_end_mask_0 = const()[name = tensor<string, []>("op_2469_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_2469_cast = slice_by_index(begin = var_2469_begin_0, end = var_2469_end_0, end_mask = var_2469_end_mask_0, x = transpose_23);
            tensor<int32, [4]> var_2473_begin_0 = const()[name = tensor<string, []>("op_2473_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_2473_end_0 = const()[name = tensor<string, []>("op_2473_end_0"), val = tensor<int32, [4]>([2, 256, 1, 192])];
            tensor<bool, [4]> var_2473_end_mask_0 = const()[name = tensor<string, []>("op_2473_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_2473_cast = slice_by_index(begin = var_2473_begin_0, end = var_2473_end_0, end_mask = var_2473_end_mask_0, x = transpose_23);
            tensor<int32, [4]> var_2477_begin_0 = const()[name = tensor<string, []>("op_2477_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_2477_end_0 = const()[name = tensor<string, []>("op_2477_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_2477_end_mask_0 = const()[name = tensor<string, []>("op_2477_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_2477_cast = slice_by_index(begin = var_2477_begin_0, end = var_2477_end_0, end_mask = var_2477_end_mask_0, x = transpose_23);
            tensor<int32, [4]> var_2481_begin_0 = const()[name = tensor<string, []>("op_2481_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_2481_end_0 = const()[name = tensor<string, []>("op_2481_end_0"), val = tensor<int32, [4]>([2, 256, 1, 320])];
            tensor<bool, [4]> var_2481_end_mask_0 = const()[name = tensor<string, []>("op_2481_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_2481_cast = slice_by_index(begin = var_2481_begin_0, end = var_2481_end_0, end_mask = var_2481_end_mask_0, x = transpose_23);
            tensor<int32, [4]> var_2485_begin_0 = const()[name = tensor<string, []>("op_2485_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_2485_end_0 = const()[name = tensor<string, []>("op_2485_end_0"), val = tensor<int32, [4]>([2, 256, 1, 384])];
            tensor<bool, [4]> var_2485_end_mask_0 = const()[name = tensor<string, []>("op_2485_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_2485_cast = slice_by_index(begin = var_2485_begin_0, end = var_2485_end_0, end_mask = var_2485_end_mask_0, x = transpose_23);
            tensor<int32, [4]> var_2489_begin_0 = const()[name = tensor<string, []>("op_2489_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_2489_end_0 = const()[name = tensor<string, []>("op_2489_end_0"), val = tensor<int32, [4]>([2, 256, 1, 448])];
            tensor<bool, [4]> var_2489_end_mask_0 = const()[name = tensor<string, []>("op_2489_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_2489_cast = slice_by_index(begin = var_2489_begin_0, end = var_2489_end_0, end_mask = var_2489_end_mask_0, x = transpose_23);
            tensor<int32, [4]> var_2493_begin_0 = const()[name = tensor<string, []>("op_2493_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_2493_end_0 = const()[name = tensor<string, []>("op_2493_end_0"), val = tensor<int32, [4]>([2, 256, 1, 512])];
            tensor<bool, [4]> var_2493_end_mask_0 = const()[name = tensor<string, []>("op_2493_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_2493_cast = slice_by_index(begin = var_2493_begin_0, end = var_2493_end_0, end_mask = var_2493_end_mask_0, x = transpose_23);
            tensor<int32, [4]> var_2497_begin_0 = const()[name = tensor<string, []>("op_2497_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2497_end_0 = const()[name = tensor<string, []>("op_2497_end_0"), val = tensor<int32, [4]>([2, 256, 1, 576])];
            tensor<bool, [4]> var_2497_end_mask_0 = const()[name = tensor<string, []>("op_2497_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_2497_cast = slice_by_index(begin = var_2497_begin_0, end = var_2497_end_0, end_mask = var_2497_end_mask_0, x = transpose_23);
            tensor<int32, [4]> var_2501_begin_0 = const()[name = tensor<string, []>("op_2501_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_2501_end_0 = const()[name = tensor<string, []>("op_2501_end_0"), val = tensor<int32, [4]>([2, 256, 1, 640])];
            tensor<bool, [4]> var_2501_end_mask_0 = const()[name = tensor<string, []>("op_2501_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_2501_cast = slice_by_index(begin = var_2501_begin_0, end = var_2501_end_0, end_mask = var_2501_end_mask_0, x = transpose_23);
            tensor<int32, [4]> var_2505_begin_0 = const()[name = tensor<string, []>("op_2505_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_2505_end_0 = const()[name = tensor<string, []>("op_2505_end_0"), val = tensor<int32, [4]>([2, 256, 1, 704])];
            tensor<bool, [4]> var_2505_end_mask_0 = const()[name = tensor<string, []>("op_2505_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_2505_cast = slice_by_index(begin = var_2505_begin_0, end = var_2505_end_0, end_mask = var_2505_end_mask_0, x = transpose_23);
            tensor<int32, [4]> var_2509_begin_0 = const()[name = tensor<string, []>("op_2509_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_2509_end_0 = const()[name = tensor<string, []>("op_2509_end_0"), val = tensor<int32, [4]>([2, 256, 1, 768])];
            tensor<bool, [4]> var_2509_end_mask_0 = const()[name = tensor<string, []>("op_2509_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_2509_cast = slice_by_index(begin = var_2509_begin_0, end = var_2509_end_0, end_mask = var_2509_end_mask_0, x = transpose_23);
            tensor<int32, [4]> var_2513_begin_0 = const()[name = tensor<string, []>("op_2513_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_2513_end_0 = const()[name = tensor<string, []>("op_2513_end_0"), val = tensor<int32, [4]>([2, 256, 1, 832])];
            tensor<bool, [4]> var_2513_end_mask_0 = const()[name = tensor<string, []>("op_2513_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_2513_cast = slice_by_index(begin = var_2513_begin_0, end = var_2513_end_0, end_mask = var_2513_end_mask_0, x = transpose_23);
            tensor<int32, [4]> var_2517_begin_0 = const()[name = tensor<string, []>("op_2517_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_2517_end_0 = const()[name = tensor<string, []>("op_2517_end_0"), val = tensor<int32, [4]>([2, 256, 1, 896])];
            tensor<bool, [4]> var_2517_end_mask_0 = const()[name = tensor<string, []>("op_2517_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_2517_cast = slice_by_index(begin = var_2517_begin_0, end = var_2517_end_0, end_mask = var_2517_end_mask_0, x = transpose_23);
            tensor<int32, [4]> var_2521_begin_0 = const()[name = tensor<string, []>("op_2521_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_2521_end_0 = const()[name = tensor<string, []>("op_2521_end_0"), val = tensor<int32, [4]>([2, 256, 1, 960])];
            tensor<bool, [4]> var_2521_end_mask_0 = const()[name = tensor<string, []>("op_2521_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_2521_cast = slice_by_index(begin = var_2521_begin_0, end = var_2521_end_0, end_mask = var_2521_end_mask_0, x = transpose_23);
            tensor<int32, [4]> var_2525_begin_0 = const()[name = tensor<string, []>("op_2525_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_2525_end_0 = const()[name = tensor<string, []>("op_2525_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_2525_end_mask_0 = const()[name = tensor<string, []>("op_2525_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_2525_cast = slice_by_index(begin = var_2525_begin_0, end = var_2525_end_0, end_mask = var_2525_end_mask_0, x = transpose_23);
            tensor<int32, [4]> var_2529_begin_0 = const()[name = tensor<string, []>("op_2529_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_2529_end_0 = const()[name = tensor<string, []>("op_2529_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1088])];
            tensor<bool, [4]> var_2529_end_mask_0 = const()[name = tensor<string, []>("op_2529_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_2529_cast = slice_by_index(begin = var_2529_begin_0, end = var_2529_end_0, end_mask = var_2529_end_mask_0, x = transpose_23);
            tensor<int32, [4]> var_2533_begin_0 = const()[name = tensor<string, []>("op_2533_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_2533_end_0 = const()[name = tensor<string, []>("op_2533_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1152])];
            tensor<bool, [4]> var_2533_end_mask_0 = const()[name = tensor<string, []>("op_2533_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_2533_cast = slice_by_index(begin = var_2533_begin_0, end = var_2533_end_0, end_mask = var_2533_end_mask_0, x = transpose_23);
            tensor<int32, [4]> var_2537_begin_0 = const()[name = tensor<string, []>("op_2537_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_2537_end_0 = const()[name = tensor<string, []>("op_2537_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1216])];
            tensor<bool, [4]> var_2537_end_mask_0 = const()[name = tensor<string, []>("op_2537_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_2537_cast = slice_by_index(begin = var_2537_begin_0, end = var_2537_end_0, end_mask = var_2537_end_mask_0, x = transpose_23);
            tensor<int32, [4]> var_2541_begin_0 = const()[name = tensor<string, []>("op_2541_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_2541_end_0 = const()[name = tensor<string, []>("op_2541_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1280])];
            tensor<bool, [4]> var_2541_end_mask_0 = const()[name = tensor<string, []>("op_2541_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_2541_cast = slice_by_index(begin = var_2541_begin_0, end = var_2541_end_0, end_mask = var_2541_end_mask_0, x = transpose_23);
            tensor<int32, [4]> var_2543_begin_0 = const()[name = tensor<string, []>("op_2543_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2543_end_0 = const()[name = tensor<string, []>("op_2543_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_2543_end_mask_0 = const()[name = tensor<string, []>("op_2543_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2543_cast = slice_by_index(begin = var_2543_begin_0, end = var_2543_end_0, end_mask = var_2543_end_mask_0, x = v_17_cast);
            tensor<int32, [4]> var_2547_begin_0 = const()[name = tensor<string, []>("op_2547_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_2547_end_0 = const()[name = tensor<string, []>("op_2547_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_2547_end_mask_0 = const()[name = tensor<string, []>("op_2547_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2547_cast = slice_by_index(begin = var_2547_begin_0, end = var_2547_end_0, end_mask = var_2547_end_mask_0, x = v_17_cast);
            tensor<int32, [4]> var_2551_begin_0 = const()[name = tensor<string, []>("op_2551_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_2551_end_0 = const()[name = tensor<string, []>("op_2551_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_2551_end_mask_0 = const()[name = tensor<string, []>("op_2551_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2551_cast = slice_by_index(begin = var_2551_begin_0, end = var_2551_end_0, end_mask = var_2551_end_mask_0, x = v_17_cast);
            tensor<int32, [4]> var_2555_begin_0 = const()[name = tensor<string, []>("op_2555_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_2555_end_0 = const()[name = tensor<string, []>("op_2555_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_2555_end_mask_0 = const()[name = tensor<string, []>("op_2555_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2555_cast = slice_by_index(begin = var_2555_begin_0, end = var_2555_end_0, end_mask = var_2555_end_mask_0, x = v_17_cast);
            tensor<int32, [4]> var_2559_begin_0 = const()[name = tensor<string, []>("op_2559_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_2559_end_0 = const()[name = tensor<string, []>("op_2559_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_2559_end_mask_0 = const()[name = tensor<string, []>("op_2559_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2559_cast = slice_by_index(begin = var_2559_begin_0, end = var_2559_end_0, end_mask = var_2559_end_mask_0, x = v_17_cast);
            tensor<int32, [4]> var_2563_begin_0 = const()[name = tensor<string, []>("op_2563_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_2563_end_0 = const()[name = tensor<string, []>("op_2563_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_2563_end_mask_0 = const()[name = tensor<string, []>("op_2563_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2563_cast = slice_by_index(begin = var_2563_begin_0, end = var_2563_end_0, end_mask = var_2563_end_mask_0, x = v_17_cast);
            tensor<int32, [4]> var_2567_begin_0 = const()[name = tensor<string, []>("op_2567_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_2567_end_0 = const()[name = tensor<string, []>("op_2567_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_2567_end_mask_0 = const()[name = tensor<string, []>("op_2567_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2567_cast = slice_by_index(begin = var_2567_begin_0, end = var_2567_end_0, end_mask = var_2567_end_mask_0, x = v_17_cast);
            tensor<int32, [4]> var_2571_begin_0 = const()[name = tensor<string, []>("op_2571_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_2571_end_0 = const()[name = tensor<string, []>("op_2571_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_2571_end_mask_0 = const()[name = tensor<string, []>("op_2571_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2571_cast = slice_by_index(begin = var_2571_begin_0, end = var_2571_end_0, end_mask = var_2571_end_mask_0, x = v_17_cast);
            tensor<int32, [4]> var_2575_begin_0 = const()[name = tensor<string, []>("op_2575_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_2575_end_0 = const()[name = tensor<string, []>("op_2575_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_2575_end_mask_0 = const()[name = tensor<string, []>("op_2575_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2575_cast = slice_by_index(begin = var_2575_begin_0, end = var_2575_end_0, end_mask = var_2575_end_mask_0, x = v_17_cast);
            tensor<int32, [4]> var_2579_begin_0 = const()[name = tensor<string, []>("op_2579_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_2579_end_0 = const()[name = tensor<string, []>("op_2579_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_2579_end_mask_0 = const()[name = tensor<string, []>("op_2579_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2579_cast = slice_by_index(begin = var_2579_begin_0, end = var_2579_end_0, end_mask = var_2579_end_mask_0, x = v_17_cast);
            tensor<int32, [4]> var_2583_begin_0 = const()[name = tensor<string, []>("op_2583_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_2583_end_0 = const()[name = tensor<string, []>("op_2583_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_2583_end_mask_0 = const()[name = tensor<string, []>("op_2583_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2583_cast = slice_by_index(begin = var_2583_begin_0, end = var_2583_end_0, end_mask = var_2583_end_mask_0, x = v_17_cast);
            tensor<int32, [4]> var_2587_begin_0 = const()[name = tensor<string, []>("op_2587_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_2587_end_0 = const()[name = tensor<string, []>("op_2587_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_2587_end_mask_0 = const()[name = tensor<string, []>("op_2587_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2587_cast = slice_by_index(begin = var_2587_begin_0, end = var_2587_end_0, end_mask = var_2587_end_mask_0, x = v_17_cast);
            tensor<int32, [4]> var_2591_begin_0 = const()[name = tensor<string, []>("op_2591_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_2591_end_0 = const()[name = tensor<string, []>("op_2591_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_2591_end_mask_0 = const()[name = tensor<string, []>("op_2591_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2591_cast = slice_by_index(begin = var_2591_begin_0, end = var_2591_end_0, end_mask = var_2591_end_mask_0, x = v_17_cast);
            tensor<int32, [4]> var_2595_begin_0 = const()[name = tensor<string, []>("op_2595_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_2595_end_0 = const()[name = tensor<string, []>("op_2595_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_2595_end_mask_0 = const()[name = tensor<string, []>("op_2595_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2595_cast = slice_by_index(begin = var_2595_begin_0, end = var_2595_end_0, end_mask = var_2595_end_mask_0, x = v_17_cast);
            tensor<int32, [4]> var_2599_begin_0 = const()[name = tensor<string, []>("op_2599_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_2599_end_0 = const()[name = tensor<string, []>("op_2599_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_2599_end_mask_0 = const()[name = tensor<string, []>("op_2599_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2599_cast = slice_by_index(begin = var_2599_begin_0, end = var_2599_end_0, end_mask = var_2599_end_mask_0, x = v_17_cast);
            tensor<int32, [4]> var_2603_begin_0 = const()[name = tensor<string, []>("op_2603_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_2603_end_0 = const()[name = tensor<string, []>("op_2603_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_2603_end_mask_0 = const()[name = tensor<string, []>("op_2603_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2603_cast = slice_by_index(begin = var_2603_begin_0, end = var_2603_end_0, end_mask = var_2603_end_mask_0, x = v_17_cast);
            tensor<int32, [4]> var_2607_begin_0 = const()[name = tensor<string, []>("op_2607_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_2607_end_0 = const()[name = tensor<string, []>("op_2607_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_2607_end_mask_0 = const()[name = tensor<string, []>("op_2607_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2607_cast = slice_by_index(begin = var_2607_begin_0, end = var_2607_end_0, end_mask = var_2607_end_mask_0, x = v_17_cast);
            tensor<int32, [4]> var_2611_begin_0 = const()[name = tensor<string, []>("op_2611_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_2611_end_0 = const()[name = tensor<string, []>("op_2611_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_2611_end_mask_0 = const()[name = tensor<string, []>("op_2611_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2611_cast = slice_by_index(begin = var_2611_begin_0, end = var_2611_end_0, end_mask = var_2611_end_mask_0, x = v_17_cast);
            tensor<int32, [4]> var_2615_begin_0 = const()[name = tensor<string, []>("op_2615_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_2615_end_0 = const()[name = tensor<string, []>("op_2615_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_2615_end_mask_0 = const()[name = tensor<string, []>("op_2615_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2615_cast = slice_by_index(begin = var_2615_begin_0, end = var_2615_end_0, end_mask = var_2615_end_mask_0, x = v_17_cast);
            tensor<int32, [4]> var_2619_begin_0 = const()[name = tensor<string, []>("op_2619_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_2619_end_0 = const()[name = tensor<string, []>("op_2619_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_2619_end_mask_0 = const()[name = tensor<string, []>("op_2619_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2619_cast = slice_by_index(begin = var_2619_begin_0, end = var_2619_end_0, end_mask = var_2619_end_mask_0, x = v_17_cast);
            tensor<string, []> var_2623_equation_0 = const()[name = tensor<string, []>("op_2623_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_2623_cast = einsum(equation = var_2623_equation_0, values = (var_2465_cast, var_2382_cast));
            tensor<fp16, []> var_2624_to_fp16 = const()[name = tensor<string, []>("op_2624_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_121_cast = mul(x = var_2623_cast, y = var_2624_to_fp16);
            tensor<string, []> var_2627_equation_0 = const()[name = tensor<string, []>("op_2627_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_2627_cast = einsum(equation = var_2627_equation_0, values = (var_2469_cast, var_2386_cast));
            tensor<fp16, []> var_2628_to_fp16 = const()[name = tensor<string, []>("op_2628_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_123_cast = mul(x = var_2627_cast, y = var_2628_to_fp16);
            tensor<string, []> var_2631_equation_0 = const()[name = tensor<string, []>("op_2631_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_2631_cast = einsum(equation = var_2631_equation_0, values = (var_2473_cast, var_2390_cast));
            tensor<fp16, []> var_2632_to_fp16 = const()[name = tensor<string, []>("op_2632_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_125_cast = mul(x = var_2631_cast, y = var_2632_to_fp16);
            tensor<string, []> var_2635_equation_0 = const()[name = tensor<string, []>("op_2635_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_2635_cast = einsum(equation = var_2635_equation_0, values = (var_2477_cast, var_2394_cast));
            tensor<fp16, []> var_2636_to_fp16 = const()[name = tensor<string, []>("op_2636_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_127_cast = mul(x = var_2635_cast, y = var_2636_to_fp16);
            tensor<string, []> var_2639_equation_0 = const()[name = tensor<string, []>("op_2639_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_2639_cast = einsum(equation = var_2639_equation_0, values = (var_2481_cast, var_2398_cast));
            tensor<fp16, []> var_2640_to_fp16 = const()[name = tensor<string, []>("op_2640_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_129_cast = mul(x = var_2639_cast, y = var_2640_to_fp16);
            tensor<string, []> var_2643_equation_0 = const()[name = tensor<string, []>("op_2643_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_2643_cast = einsum(equation = var_2643_equation_0, values = (var_2485_cast, var_2402_cast));
            tensor<fp16, []> var_2644_to_fp16 = const()[name = tensor<string, []>("op_2644_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_131_cast = mul(x = var_2643_cast, y = var_2644_to_fp16);
            tensor<string, []> var_2647_equation_0 = const()[name = tensor<string, []>("op_2647_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_2647_cast = einsum(equation = var_2647_equation_0, values = (var_2489_cast, var_2406_cast));
            tensor<fp16, []> var_2648_to_fp16 = const()[name = tensor<string, []>("op_2648_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_133_cast = mul(x = var_2647_cast, y = var_2648_to_fp16);
            tensor<string, []> var_2651_equation_0 = const()[name = tensor<string, []>("op_2651_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_2651_cast = einsum(equation = var_2651_equation_0, values = (var_2493_cast, var_2410_cast));
            tensor<fp16, []> var_2652_to_fp16 = const()[name = tensor<string, []>("op_2652_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_135_cast = mul(x = var_2651_cast, y = var_2652_to_fp16);
            tensor<string, []> var_2655_equation_0 = const()[name = tensor<string, []>("op_2655_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_2655_cast = einsum(equation = var_2655_equation_0, values = (var_2497_cast, var_2414_cast));
            tensor<fp16, []> var_2656_to_fp16 = const()[name = tensor<string, []>("op_2656_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_137_cast = mul(x = var_2655_cast, y = var_2656_to_fp16);
            tensor<string, []> var_2659_equation_0 = const()[name = tensor<string, []>("op_2659_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_2659_cast = einsum(equation = var_2659_equation_0, values = (var_2501_cast, var_2418_cast));
            tensor<fp16, []> var_2660_to_fp16 = const()[name = tensor<string, []>("op_2660_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_139_cast = mul(x = var_2659_cast, y = var_2660_to_fp16);
            tensor<string, []> var_2663_equation_0 = const()[name = tensor<string, []>("op_2663_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_2663_cast = einsum(equation = var_2663_equation_0, values = (var_2505_cast, var_2422_cast));
            tensor<fp16, []> var_2664_to_fp16 = const()[name = tensor<string, []>("op_2664_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_141_cast = mul(x = var_2663_cast, y = var_2664_to_fp16);
            tensor<string, []> var_2667_equation_0 = const()[name = tensor<string, []>("op_2667_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_2667_cast = einsum(equation = var_2667_equation_0, values = (var_2509_cast, var_2426_cast));
            tensor<fp16, []> var_2668_to_fp16 = const()[name = tensor<string, []>("op_2668_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_143_cast = mul(x = var_2667_cast, y = var_2668_to_fp16);
            tensor<string, []> var_2671_equation_0 = const()[name = tensor<string, []>("op_2671_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_2671_cast = einsum(equation = var_2671_equation_0, values = (var_2513_cast, var_2430_cast));
            tensor<fp16, []> var_2672_to_fp16 = const()[name = tensor<string, []>("op_2672_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_145_cast = mul(x = var_2671_cast, y = var_2672_to_fp16);
            tensor<string, []> var_2675_equation_0 = const()[name = tensor<string, []>("op_2675_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_2675_cast = einsum(equation = var_2675_equation_0, values = (var_2517_cast, var_2434_cast));
            tensor<fp16, []> var_2676_to_fp16 = const()[name = tensor<string, []>("op_2676_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_147_cast = mul(x = var_2675_cast, y = var_2676_to_fp16);
            tensor<string, []> var_2679_equation_0 = const()[name = tensor<string, []>("op_2679_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_2679_cast = einsum(equation = var_2679_equation_0, values = (var_2521_cast, var_2438_cast));
            tensor<fp16, []> var_2680_to_fp16 = const()[name = tensor<string, []>("op_2680_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_149_cast = mul(x = var_2679_cast, y = var_2680_to_fp16);
            tensor<string, []> var_2683_equation_0 = const()[name = tensor<string, []>("op_2683_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_2683_cast = einsum(equation = var_2683_equation_0, values = (var_2525_cast, var_2442_cast));
            tensor<fp16, []> var_2684_to_fp16 = const()[name = tensor<string, []>("op_2684_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_151_cast = mul(x = var_2683_cast, y = var_2684_to_fp16);
            tensor<string, []> var_2687_equation_0 = const()[name = tensor<string, []>("op_2687_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_2687_cast = einsum(equation = var_2687_equation_0, values = (var_2529_cast, var_2446_cast));
            tensor<fp16, []> var_2688_to_fp16 = const()[name = tensor<string, []>("op_2688_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_153_cast = mul(x = var_2687_cast, y = var_2688_to_fp16);
            tensor<string, []> var_2691_equation_0 = const()[name = tensor<string, []>("op_2691_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_2691_cast = einsum(equation = var_2691_equation_0, values = (var_2533_cast, var_2450_cast));
            tensor<fp16, []> var_2692_to_fp16 = const()[name = tensor<string, []>("op_2692_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_155_cast = mul(x = var_2691_cast, y = var_2692_to_fp16);
            tensor<string, []> var_2695_equation_0 = const()[name = tensor<string, []>("op_2695_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_2695_cast = einsum(equation = var_2695_equation_0, values = (var_2537_cast, var_2454_cast));
            tensor<fp16, []> var_2696_to_fp16 = const()[name = tensor<string, []>("op_2696_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_157_cast = mul(x = var_2695_cast, y = var_2696_to_fp16);
            tensor<string, []> var_2699_equation_0 = const()[name = tensor<string, []>("op_2699_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_2699_cast = einsum(equation = var_2699_equation_0, values = (var_2541_cast, var_2458_cast));
            tensor<fp16, []> var_2700_to_fp16 = const()[name = tensor<string, []>("op_2700_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_159_cast = mul(x = var_2699_cast, y = var_2700_to_fp16);
            tensor<fp16, [2, 256, 1, 256]> var_2702_cast = softmax(axis = var_2254, x = aw_121_cast);
            tensor<fp16, [2, 256, 1, 256]> var_2703_cast = softmax(axis = var_2254, x = aw_123_cast);
            tensor<fp16, [2, 256, 1, 256]> var_2704_cast = softmax(axis = var_2254, x = aw_125_cast);
            tensor<fp16, [2, 256, 1, 256]> var_2705_cast = softmax(axis = var_2254, x = aw_127_cast);
            tensor<fp16, [2, 256, 1, 256]> var_2706_cast = softmax(axis = var_2254, x = aw_129_cast);
            tensor<fp16, [2, 256, 1, 256]> var_2707_cast = softmax(axis = var_2254, x = aw_131_cast);
            tensor<fp16, [2, 256, 1, 256]> var_2708_cast = softmax(axis = var_2254, x = aw_133_cast);
            tensor<fp16, [2, 256, 1, 256]> var_2709_cast = softmax(axis = var_2254, x = aw_135_cast);
            tensor<fp16, [2, 256, 1, 256]> var_2710_cast = softmax(axis = var_2254, x = aw_137_cast);
            tensor<fp16, [2, 256, 1, 256]> var_2711_cast = softmax(axis = var_2254, x = aw_139_cast);
            tensor<fp16, [2, 256, 1, 256]> var_2712_cast = softmax(axis = var_2254, x = aw_141_cast);
            tensor<fp16, [2, 256, 1, 256]> var_2713_cast = softmax(axis = var_2254, x = aw_143_cast);
            tensor<fp16, [2, 256, 1, 256]> var_2714_cast = softmax(axis = var_2254, x = aw_145_cast);
            tensor<fp16, [2, 256, 1, 256]> var_2715_cast = softmax(axis = var_2254, x = aw_147_cast);
            tensor<fp16, [2, 256, 1, 256]> var_2716_cast = softmax(axis = var_2254, x = aw_149_cast);
            tensor<fp16, [2, 256, 1, 256]> var_2717_cast = softmax(axis = var_2254, x = aw_151_cast);
            tensor<fp16, [2, 256, 1, 256]> var_2718_cast = softmax(axis = var_2254, x = aw_153_cast);
            tensor<fp16, [2, 256, 1, 256]> var_2719_cast = softmax(axis = var_2254, x = aw_155_cast);
            tensor<fp16, [2, 256, 1, 256]> var_2720_cast = softmax(axis = var_2254, x = aw_157_cast);
            tensor<fp16, [2, 256, 1, 256]> var_2721_cast = softmax(axis = var_2254, x = aw_159_cast);
            tensor<string, []> var_2723_equation_0 = const()[name = tensor<string, []>("op_2723_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_2723_cast = einsum(equation = var_2723_equation_0, values = (var_2543_cast, var_2702_cast));
            tensor<string, []> var_2725_equation_0 = const()[name = tensor<string, []>("op_2725_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_2725_cast = einsum(equation = var_2725_equation_0, values = (var_2547_cast, var_2703_cast));
            tensor<string, []> var_2727_equation_0 = const()[name = tensor<string, []>("op_2727_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_2727_cast = einsum(equation = var_2727_equation_0, values = (var_2551_cast, var_2704_cast));
            tensor<string, []> var_2729_equation_0 = const()[name = tensor<string, []>("op_2729_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_2729_cast = einsum(equation = var_2729_equation_0, values = (var_2555_cast, var_2705_cast));
            tensor<string, []> var_2731_equation_0 = const()[name = tensor<string, []>("op_2731_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_2731_cast = einsum(equation = var_2731_equation_0, values = (var_2559_cast, var_2706_cast));
            tensor<string, []> var_2733_equation_0 = const()[name = tensor<string, []>("op_2733_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_2733_cast = einsum(equation = var_2733_equation_0, values = (var_2563_cast, var_2707_cast));
            tensor<string, []> var_2735_equation_0 = const()[name = tensor<string, []>("op_2735_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_2735_cast = einsum(equation = var_2735_equation_0, values = (var_2567_cast, var_2708_cast));
            tensor<string, []> var_2737_equation_0 = const()[name = tensor<string, []>("op_2737_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_2737_cast = einsum(equation = var_2737_equation_0, values = (var_2571_cast, var_2709_cast));
            tensor<string, []> var_2739_equation_0 = const()[name = tensor<string, []>("op_2739_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_2739_cast = einsum(equation = var_2739_equation_0, values = (var_2575_cast, var_2710_cast));
            tensor<string, []> var_2741_equation_0 = const()[name = tensor<string, []>("op_2741_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_2741_cast = einsum(equation = var_2741_equation_0, values = (var_2579_cast, var_2711_cast));
            tensor<string, []> var_2743_equation_0 = const()[name = tensor<string, []>("op_2743_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_2743_cast = einsum(equation = var_2743_equation_0, values = (var_2583_cast, var_2712_cast));
            tensor<string, []> var_2745_equation_0 = const()[name = tensor<string, []>("op_2745_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_2745_cast = einsum(equation = var_2745_equation_0, values = (var_2587_cast, var_2713_cast));
            tensor<string, []> var_2747_equation_0 = const()[name = tensor<string, []>("op_2747_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_2747_cast = einsum(equation = var_2747_equation_0, values = (var_2591_cast, var_2714_cast));
            tensor<string, []> var_2749_equation_0 = const()[name = tensor<string, []>("op_2749_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_2749_cast = einsum(equation = var_2749_equation_0, values = (var_2595_cast, var_2715_cast));
            tensor<string, []> var_2751_equation_0 = const()[name = tensor<string, []>("op_2751_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_2751_cast = einsum(equation = var_2751_equation_0, values = (var_2599_cast, var_2716_cast));
            tensor<string, []> var_2753_equation_0 = const()[name = tensor<string, []>("op_2753_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_2753_cast = einsum(equation = var_2753_equation_0, values = (var_2603_cast, var_2717_cast));
            tensor<string, []> var_2755_equation_0 = const()[name = tensor<string, []>("op_2755_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_2755_cast = einsum(equation = var_2755_equation_0, values = (var_2607_cast, var_2718_cast));
            tensor<string, []> var_2757_equation_0 = const()[name = tensor<string, []>("op_2757_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_2757_cast = einsum(equation = var_2757_equation_0, values = (var_2611_cast, var_2719_cast));
            tensor<string, []> var_2759_equation_0 = const()[name = tensor<string, []>("op_2759_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_2759_cast = einsum(equation = var_2759_equation_0, values = (var_2615_cast, var_2720_cast));
            tensor<string, []> var_2761_equation_0 = const()[name = tensor<string, []>("op_2761_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_2761_cast = einsum(equation = var_2761_equation_0, values = (var_2619_cast, var_2721_cast));
            tensor<bool, []> input_133_interleave_0 = const()[name = tensor<string, []>("input_133_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 256]> input_133_cast = concat(axis = var_2254, interleave = input_133_interleave_0, values = (var_2723_cast, var_2725_cast, var_2727_cast, var_2729_cast, var_2731_cast, var_2733_cast, var_2735_cast, var_2737_cast, var_2739_cast, var_2741_cast, var_2743_cast, var_2745_cast, var_2747_cast, var_2749_cast, var_2751_cast, var_2753_cast, var_2755_cast, var_2757_cast, var_2759_cast, var_2761_cast));
            tensor<int32, [2]> var_2767 = const()[name = tensor<string, []>("op_2767"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2769 = const()[name = tensor<string, []>("op_2769"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_2771_pad_type_0 = const()[name = tensor<string, []>("op_2771_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_2771_pad_0 = const()[name = tensor<string, []>("op_2771_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163115136)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166392000)))];
            tensor<fp16, [2, 1280, 1, 256]> var_2771_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_2769, groups = var_2254, pad = var_2771_pad_0, pad_type = var_2771_pad_type_0, strides = var_2767, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_133_cast);
            tensor<fp16, [2, 1280, 1, 256]> inputs_27_cast = add(x = var_2771_cast, y = inputs_25_cast);
            tensor<int32, [1]> var_2775 = const()[name = tensor<string, []>("op_2775"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_27_cast = reduce_mean(axes = var_2775, keep_dims = var_2249, x = inputs_27_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_27_cast = sub(x = inputs_27_cast, y = channels_mean_27_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_sq_27_cast = mul(x = zero_mean_27_cast, y = zero_mean_27_cast);
            tensor<int32, [1]> var_2779 = const()[name = tensor<string, []>("op_2779"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_2780_cast = reduce_mean(axes = var_2779, keep_dims = var_2249, x = zero_mean_sq_27_cast);
            tensor<fp16, []> var_2781_to_fp16 = const()[name = tensor<string, []>("op_2781_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_2782_cast = add(x = var_2780_cast, y = var_2781_to_fp16);
            tensor<fp16, []> denom_27_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_27_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 256]> denom_27_cast = rsqrt(epsilon = denom_27_epsilon_0_to_fp16, x = var_2782_cast);
            tensor<fp16, [2, 1280, 1, 256]> out_27_cast = mul(x = zero_mean_27_cast, y = denom_27_cast);
            tensor<fp16, [1, 1280, 1, 1]> var_2786_to_fp16 = const()[name = tensor<string, []>("op_2786_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166394624)))];
            tensor<fp16, [2, 1280, 1, 256]> var_2787_cast = add(x = out_27_cast, y = var_2786_to_fp16);
            tensor<fp16, [1, 1280, 1, 1]> var_2789_to_fp16 = const()[name = tensor<string, []>("op_2789_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166397248)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_83_cast = mul(x = var_2787_cast, y = var_2789_to_fp16);
            tensor<int32, [2]> var_2796 = const()[name = tensor<string, []>("op_2796"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2798 = const()[name = tensor<string, []>("op_2798"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_19_pad_type_0 = const()[name = tensor<string, []>("q_19_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_19_pad_0 = const()[name = tensor<string, []>("q_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166399872)))];
            tensor<fp16, [2, 1280, 1, 256]> q_19_cast = conv(dilations = var_2798, groups = var_2254, pad = q_19_pad_0, pad_type = q_19_pad_type_0, strides = var_2796, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_83_cast);
            tensor<int32, [2]> var_2802 = const()[name = tensor<string, []>("op_2802"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2804 = const()[name = tensor<string, []>("op_2804"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_37_pad_type_0 = const()[name = tensor<string, []>("k_37_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_37_pad_0 = const()[name = tensor<string, []>("k_37_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1024, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169676736)))];
            tensor<fp16, [2, 1280, 1, 77]> k_37_cast = conv(dilations = var_2804, groups = var_2254, pad = k_37_pad_0, pad_type = k_37_pad_type_0, strides = var_2802, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [2]> var_2808 = const()[name = tensor<string, []>("op_2808"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2810 = const()[name = tensor<string, []>("op_2810"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_19_pad_type_0 = const()[name = tensor<string, []>("v_19_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_19_pad_0 = const()[name = tensor<string, []>("v_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1024, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(172298240)))];
            tensor<fp16, [2, 1280, 1, 77]> v_19_cast = conv(dilations = var_2810, groups = var_2254, pad = v_19_pad_0, pad_type = v_19_pad_type_0, strides = var_2808, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [4]> var_2814_begin_0 = const()[name = tensor<string, []>("op_2814_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2814_end_0 = const()[name = tensor<string, []>("op_2814_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_2814_end_mask_0 = const()[name = tensor<string, []>("op_2814_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2814_cast = slice_by_index(begin = var_2814_begin_0, end = var_2814_end_0, end_mask = var_2814_end_mask_0, x = q_19_cast);
            tensor<int32, [4]> var_2818_begin_0 = const()[name = tensor<string, []>("op_2818_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_2818_end_0 = const()[name = tensor<string, []>("op_2818_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_2818_end_mask_0 = const()[name = tensor<string, []>("op_2818_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2818_cast = slice_by_index(begin = var_2818_begin_0, end = var_2818_end_0, end_mask = var_2818_end_mask_0, x = q_19_cast);
            tensor<int32, [4]> var_2822_begin_0 = const()[name = tensor<string, []>("op_2822_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_2822_end_0 = const()[name = tensor<string, []>("op_2822_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_2822_end_mask_0 = const()[name = tensor<string, []>("op_2822_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2822_cast = slice_by_index(begin = var_2822_begin_0, end = var_2822_end_0, end_mask = var_2822_end_mask_0, x = q_19_cast);
            tensor<int32, [4]> var_2826_begin_0 = const()[name = tensor<string, []>("op_2826_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_2826_end_0 = const()[name = tensor<string, []>("op_2826_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_2826_end_mask_0 = const()[name = tensor<string, []>("op_2826_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2826_cast = slice_by_index(begin = var_2826_begin_0, end = var_2826_end_0, end_mask = var_2826_end_mask_0, x = q_19_cast);
            tensor<int32, [4]> var_2830_begin_0 = const()[name = tensor<string, []>("op_2830_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_2830_end_0 = const()[name = tensor<string, []>("op_2830_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_2830_end_mask_0 = const()[name = tensor<string, []>("op_2830_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2830_cast = slice_by_index(begin = var_2830_begin_0, end = var_2830_end_0, end_mask = var_2830_end_mask_0, x = q_19_cast);
            tensor<int32, [4]> var_2834_begin_0 = const()[name = tensor<string, []>("op_2834_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_2834_end_0 = const()[name = tensor<string, []>("op_2834_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_2834_end_mask_0 = const()[name = tensor<string, []>("op_2834_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2834_cast = slice_by_index(begin = var_2834_begin_0, end = var_2834_end_0, end_mask = var_2834_end_mask_0, x = q_19_cast);
            tensor<int32, [4]> var_2838_begin_0 = const()[name = tensor<string, []>("op_2838_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_2838_end_0 = const()[name = tensor<string, []>("op_2838_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_2838_end_mask_0 = const()[name = tensor<string, []>("op_2838_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2838_cast = slice_by_index(begin = var_2838_begin_0, end = var_2838_end_0, end_mask = var_2838_end_mask_0, x = q_19_cast);
            tensor<int32, [4]> var_2842_begin_0 = const()[name = tensor<string, []>("op_2842_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_2842_end_0 = const()[name = tensor<string, []>("op_2842_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_2842_end_mask_0 = const()[name = tensor<string, []>("op_2842_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2842_cast = slice_by_index(begin = var_2842_begin_0, end = var_2842_end_0, end_mask = var_2842_end_mask_0, x = q_19_cast);
            tensor<int32, [4]> var_2846_begin_0 = const()[name = tensor<string, []>("op_2846_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_2846_end_0 = const()[name = tensor<string, []>("op_2846_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_2846_end_mask_0 = const()[name = tensor<string, []>("op_2846_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2846_cast = slice_by_index(begin = var_2846_begin_0, end = var_2846_end_0, end_mask = var_2846_end_mask_0, x = q_19_cast);
            tensor<int32, [4]> var_2850_begin_0 = const()[name = tensor<string, []>("op_2850_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_2850_end_0 = const()[name = tensor<string, []>("op_2850_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_2850_end_mask_0 = const()[name = tensor<string, []>("op_2850_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2850_cast = slice_by_index(begin = var_2850_begin_0, end = var_2850_end_0, end_mask = var_2850_end_mask_0, x = q_19_cast);
            tensor<int32, [4]> var_2854_begin_0 = const()[name = tensor<string, []>("op_2854_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_2854_end_0 = const()[name = tensor<string, []>("op_2854_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_2854_end_mask_0 = const()[name = tensor<string, []>("op_2854_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2854_cast = slice_by_index(begin = var_2854_begin_0, end = var_2854_end_0, end_mask = var_2854_end_mask_0, x = q_19_cast);
            tensor<int32, [4]> var_2858_begin_0 = const()[name = tensor<string, []>("op_2858_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_2858_end_0 = const()[name = tensor<string, []>("op_2858_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_2858_end_mask_0 = const()[name = tensor<string, []>("op_2858_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2858_cast = slice_by_index(begin = var_2858_begin_0, end = var_2858_end_0, end_mask = var_2858_end_mask_0, x = q_19_cast);
            tensor<int32, [4]> var_2862_begin_0 = const()[name = tensor<string, []>("op_2862_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_2862_end_0 = const()[name = tensor<string, []>("op_2862_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_2862_end_mask_0 = const()[name = tensor<string, []>("op_2862_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2862_cast = slice_by_index(begin = var_2862_begin_0, end = var_2862_end_0, end_mask = var_2862_end_mask_0, x = q_19_cast);
            tensor<int32, [4]> var_2866_begin_0 = const()[name = tensor<string, []>("op_2866_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_2866_end_0 = const()[name = tensor<string, []>("op_2866_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_2866_end_mask_0 = const()[name = tensor<string, []>("op_2866_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2866_cast = slice_by_index(begin = var_2866_begin_0, end = var_2866_end_0, end_mask = var_2866_end_mask_0, x = q_19_cast);
            tensor<int32, [4]> var_2870_begin_0 = const()[name = tensor<string, []>("op_2870_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_2870_end_0 = const()[name = tensor<string, []>("op_2870_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_2870_end_mask_0 = const()[name = tensor<string, []>("op_2870_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2870_cast = slice_by_index(begin = var_2870_begin_0, end = var_2870_end_0, end_mask = var_2870_end_mask_0, x = q_19_cast);
            tensor<int32, [4]> var_2874_begin_0 = const()[name = tensor<string, []>("op_2874_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_2874_end_0 = const()[name = tensor<string, []>("op_2874_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_2874_end_mask_0 = const()[name = tensor<string, []>("op_2874_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2874_cast = slice_by_index(begin = var_2874_begin_0, end = var_2874_end_0, end_mask = var_2874_end_mask_0, x = q_19_cast);
            tensor<int32, [4]> var_2878_begin_0 = const()[name = tensor<string, []>("op_2878_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_2878_end_0 = const()[name = tensor<string, []>("op_2878_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_2878_end_mask_0 = const()[name = tensor<string, []>("op_2878_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2878_cast = slice_by_index(begin = var_2878_begin_0, end = var_2878_end_0, end_mask = var_2878_end_mask_0, x = q_19_cast);
            tensor<int32, [4]> var_2882_begin_0 = const()[name = tensor<string, []>("op_2882_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_2882_end_0 = const()[name = tensor<string, []>("op_2882_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_2882_end_mask_0 = const()[name = tensor<string, []>("op_2882_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2882_cast = slice_by_index(begin = var_2882_begin_0, end = var_2882_end_0, end_mask = var_2882_end_mask_0, x = q_19_cast);
            tensor<int32, [4]> var_2886_begin_0 = const()[name = tensor<string, []>("op_2886_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_2886_end_0 = const()[name = tensor<string, []>("op_2886_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_2886_end_mask_0 = const()[name = tensor<string, []>("op_2886_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2886_cast = slice_by_index(begin = var_2886_begin_0, end = var_2886_end_0, end_mask = var_2886_end_mask_0, x = q_19_cast);
            tensor<int32, [4]> var_2890_begin_0 = const()[name = tensor<string, []>("op_2890_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_2890_end_0 = const()[name = tensor<string, []>("op_2890_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_2890_end_mask_0 = const()[name = tensor<string, []>("op_2890_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_2890_cast = slice_by_index(begin = var_2890_begin_0, end = var_2890_end_0, end_mask = var_2890_end_mask_0, x = q_19_cast);
            tensor<int32, [4]> k_39_perm_0 = const()[name = tensor<string, []>("k_39_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_2897_begin_0 = const()[name = tensor<string, []>("op_2897_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2897_end_0 = const()[name = tensor<string, []>("op_2897_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_2897_end_mask_0 = const()[name = tensor<string, []>("op_2897_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_22 = transpose(perm = k_39_perm_0, x = k_37_cast);
            tensor<fp16, [2, 77, 1, 64]> var_2897_cast = slice_by_index(begin = var_2897_begin_0, end = var_2897_end_0, end_mask = var_2897_end_mask_0, x = transpose_22);
            tensor<int32, [4]> var_2901_begin_0 = const()[name = tensor<string, []>("op_2901_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_2901_end_0 = const()[name = tensor<string, []>("op_2901_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_2901_end_mask_0 = const()[name = tensor<string, []>("op_2901_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2901_cast = slice_by_index(begin = var_2901_begin_0, end = var_2901_end_0, end_mask = var_2901_end_mask_0, x = transpose_22);
            tensor<int32, [4]> var_2905_begin_0 = const()[name = tensor<string, []>("op_2905_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_2905_end_0 = const()[name = tensor<string, []>("op_2905_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_2905_end_mask_0 = const()[name = tensor<string, []>("op_2905_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2905_cast = slice_by_index(begin = var_2905_begin_0, end = var_2905_end_0, end_mask = var_2905_end_mask_0, x = transpose_22);
            tensor<int32, [4]> var_2909_begin_0 = const()[name = tensor<string, []>("op_2909_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_2909_end_0 = const()[name = tensor<string, []>("op_2909_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_2909_end_mask_0 = const()[name = tensor<string, []>("op_2909_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2909_cast = slice_by_index(begin = var_2909_begin_0, end = var_2909_end_0, end_mask = var_2909_end_mask_0, x = transpose_22);
            tensor<int32, [4]> var_2913_begin_0 = const()[name = tensor<string, []>("op_2913_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_2913_end_0 = const()[name = tensor<string, []>("op_2913_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_2913_end_mask_0 = const()[name = tensor<string, []>("op_2913_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2913_cast = slice_by_index(begin = var_2913_begin_0, end = var_2913_end_0, end_mask = var_2913_end_mask_0, x = transpose_22);
            tensor<int32, [4]> var_2917_begin_0 = const()[name = tensor<string, []>("op_2917_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_2917_end_0 = const()[name = tensor<string, []>("op_2917_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_2917_end_mask_0 = const()[name = tensor<string, []>("op_2917_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2917_cast = slice_by_index(begin = var_2917_begin_0, end = var_2917_end_0, end_mask = var_2917_end_mask_0, x = transpose_22);
            tensor<int32, [4]> var_2921_begin_0 = const()[name = tensor<string, []>("op_2921_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_2921_end_0 = const()[name = tensor<string, []>("op_2921_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_2921_end_mask_0 = const()[name = tensor<string, []>("op_2921_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2921_cast = slice_by_index(begin = var_2921_begin_0, end = var_2921_end_0, end_mask = var_2921_end_mask_0, x = transpose_22);
            tensor<int32, [4]> var_2925_begin_0 = const()[name = tensor<string, []>("op_2925_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_2925_end_0 = const()[name = tensor<string, []>("op_2925_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_2925_end_mask_0 = const()[name = tensor<string, []>("op_2925_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2925_cast = slice_by_index(begin = var_2925_begin_0, end = var_2925_end_0, end_mask = var_2925_end_mask_0, x = transpose_22);
            tensor<int32, [4]> var_2929_begin_0 = const()[name = tensor<string, []>("op_2929_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2929_end_0 = const()[name = tensor<string, []>("op_2929_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_2929_end_mask_0 = const()[name = tensor<string, []>("op_2929_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2929_cast = slice_by_index(begin = var_2929_begin_0, end = var_2929_end_0, end_mask = var_2929_end_mask_0, x = transpose_22);
            tensor<int32, [4]> var_2933_begin_0 = const()[name = tensor<string, []>("op_2933_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_2933_end_0 = const()[name = tensor<string, []>("op_2933_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_2933_end_mask_0 = const()[name = tensor<string, []>("op_2933_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2933_cast = slice_by_index(begin = var_2933_begin_0, end = var_2933_end_0, end_mask = var_2933_end_mask_0, x = transpose_22);
            tensor<int32, [4]> var_2937_begin_0 = const()[name = tensor<string, []>("op_2937_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_2937_end_0 = const()[name = tensor<string, []>("op_2937_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_2937_end_mask_0 = const()[name = tensor<string, []>("op_2937_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2937_cast = slice_by_index(begin = var_2937_begin_0, end = var_2937_end_0, end_mask = var_2937_end_mask_0, x = transpose_22);
            tensor<int32, [4]> var_2941_begin_0 = const()[name = tensor<string, []>("op_2941_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_2941_end_0 = const()[name = tensor<string, []>("op_2941_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_2941_end_mask_0 = const()[name = tensor<string, []>("op_2941_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2941_cast = slice_by_index(begin = var_2941_begin_0, end = var_2941_end_0, end_mask = var_2941_end_mask_0, x = transpose_22);
            tensor<int32, [4]> var_2945_begin_0 = const()[name = tensor<string, []>("op_2945_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_2945_end_0 = const()[name = tensor<string, []>("op_2945_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_2945_end_mask_0 = const()[name = tensor<string, []>("op_2945_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2945_cast = slice_by_index(begin = var_2945_begin_0, end = var_2945_end_0, end_mask = var_2945_end_mask_0, x = transpose_22);
            tensor<int32, [4]> var_2949_begin_0 = const()[name = tensor<string, []>("op_2949_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_2949_end_0 = const()[name = tensor<string, []>("op_2949_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_2949_end_mask_0 = const()[name = tensor<string, []>("op_2949_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2949_cast = slice_by_index(begin = var_2949_begin_0, end = var_2949_end_0, end_mask = var_2949_end_mask_0, x = transpose_22);
            tensor<int32, [4]> var_2953_begin_0 = const()[name = tensor<string, []>("op_2953_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_2953_end_0 = const()[name = tensor<string, []>("op_2953_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_2953_end_mask_0 = const()[name = tensor<string, []>("op_2953_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2953_cast = slice_by_index(begin = var_2953_begin_0, end = var_2953_end_0, end_mask = var_2953_end_mask_0, x = transpose_22);
            tensor<int32, [4]> var_2957_begin_0 = const()[name = tensor<string, []>("op_2957_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_2957_end_0 = const()[name = tensor<string, []>("op_2957_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_2957_end_mask_0 = const()[name = tensor<string, []>("op_2957_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2957_cast = slice_by_index(begin = var_2957_begin_0, end = var_2957_end_0, end_mask = var_2957_end_mask_0, x = transpose_22);
            tensor<int32, [4]> var_2961_begin_0 = const()[name = tensor<string, []>("op_2961_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_2961_end_0 = const()[name = tensor<string, []>("op_2961_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_2961_end_mask_0 = const()[name = tensor<string, []>("op_2961_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2961_cast = slice_by_index(begin = var_2961_begin_0, end = var_2961_end_0, end_mask = var_2961_end_mask_0, x = transpose_22);
            tensor<int32, [4]> var_2965_begin_0 = const()[name = tensor<string, []>("op_2965_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_2965_end_0 = const()[name = tensor<string, []>("op_2965_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_2965_end_mask_0 = const()[name = tensor<string, []>("op_2965_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2965_cast = slice_by_index(begin = var_2965_begin_0, end = var_2965_end_0, end_mask = var_2965_end_mask_0, x = transpose_22);
            tensor<int32, [4]> var_2969_begin_0 = const()[name = tensor<string, []>("op_2969_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_2969_end_0 = const()[name = tensor<string, []>("op_2969_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_2969_end_mask_0 = const()[name = tensor<string, []>("op_2969_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2969_cast = slice_by_index(begin = var_2969_begin_0, end = var_2969_end_0, end_mask = var_2969_end_mask_0, x = transpose_22);
            tensor<int32, [4]> var_2973_begin_0 = const()[name = tensor<string, []>("op_2973_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_2973_end_0 = const()[name = tensor<string, []>("op_2973_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_2973_end_mask_0 = const()[name = tensor<string, []>("op_2973_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2973_cast = slice_by_index(begin = var_2973_begin_0, end = var_2973_end_0, end_mask = var_2973_end_mask_0, x = transpose_22);
            tensor<int32, [4]> var_2975_begin_0 = const()[name = tensor<string, []>("op_2975_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2975_end_0 = const()[name = tensor<string, []>("op_2975_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_2975_end_mask_0 = const()[name = tensor<string, []>("op_2975_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2975_cast = slice_by_index(begin = var_2975_begin_0, end = var_2975_end_0, end_mask = var_2975_end_mask_0, x = v_19_cast);
            tensor<int32, [4]> var_2979_begin_0 = const()[name = tensor<string, []>("op_2979_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_2979_end_0 = const()[name = tensor<string, []>("op_2979_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_2979_end_mask_0 = const()[name = tensor<string, []>("op_2979_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2979_cast = slice_by_index(begin = var_2979_begin_0, end = var_2979_end_0, end_mask = var_2979_end_mask_0, x = v_19_cast);
            tensor<int32, [4]> var_2983_begin_0 = const()[name = tensor<string, []>("op_2983_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_2983_end_0 = const()[name = tensor<string, []>("op_2983_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_2983_end_mask_0 = const()[name = tensor<string, []>("op_2983_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2983_cast = slice_by_index(begin = var_2983_begin_0, end = var_2983_end_0, end_mask = var_2983_end_mask_0, x = v_19_cast);
            tensor<int32, [4]> var_2987_begin_0 = const()[name = tensor<string, []>("op_2987_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_2987_end_0 = const()[name = tensor<string, []>("op_2987_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_2987_end_mask_0 = const()[name = tensor<string, []>("op_2987_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2987_cast = slice_by_index(begin = var_2987_begin_0, end = var_2987_end_0, end_mask = var_2987_end_mask_0, x = v_19_cast);
            tensor<int32, [4]> var_2991_begin_0 = const()[name = tensor<string, []>("op_2991_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_2991_end_0 = const()[name = tensor<string, []>("op_2991_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_2991_end_mask_0 = const()[name = tensor<string, []>("op_2991_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2991_cast = slice_by_index(begin = var_2991_begin_0, end = var_2991_end_0, end_mask = var_2991_end_mask_0, x = v_19_cast);
            tensor<int32, [4]> var_2995_begin_0 = const()[name = tensor<string, []>("op_2995_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_2995_end_0 = const()[name = tensor<string, []>("op_2995_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_2995_end_mask_0 = const()[name = tensor<string, []>("op_2995_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2995_cast = slice_by_index(begin = var_2995_begin_0, end = var_2995_end_0, end_mask = var_2995_end_mask_0, x = v_19_cast);
            tensor<int32, [4]> var_2999_begin_0 = const()[name = tensor<string, []>("op_2999_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_2999_end_0 = const()[name = tensor<string, []>("op_2999_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_2999_end_mask_0 = const()[name = tensor<string, []>("op_2999_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2999_cast = slice_by_index(begin = var_2999_begin_0, end = var_2999_end_0, end_mask = var_2999_end_mask_0, x = v_19_cast);
            tensor<int32, [4]> var_3003_begin_0 = const()[name = tensor<string, []>("op_3003_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_3003_end_0 = const()[name = tensor<string, []>("op_3003_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_3003_end_mask_0 = const()[name = tensor<string, []>("op_3003_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3003_cast = slice_by_index(begin = var_3003_begin_0, end = var_3003_end_0, end_mask = var_3003_end_mask_0, x = v_19_cast);
            tensor<int32, [4]> var_3007_begin_0 = const()[name = tensor<string, []>("op_3007_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_3007_end_0 = const()[name = tensor<string, []>("op_3007_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_3007_end_mask_0 = const()[name = tensor<string, []>("op_3007_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3007_cast = slice_by_index(begin = var_3007_begin_0, end = var_3007_end_0, end_mask = var_3007_end_mask_0, x = v_19_cast);
            tensor<int32, [4]> var_3011_begin_0 = const()[name = tensor<string, []>("op_3011_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_3011_end_0 = const()[name = tensor<string, []>("op_3011_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_3011_end_mask_0 = const()[name = tensor<string, []>("op_3011_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3011_cast = slice_by_index(begin = var_3011_begin_0, end = var_3011_end_0, end_mask = var_3011_end_mask_0, x = v_19_cast);
            tensor<int32, [4]> var_3015_begin_0 = const()[name = tensor<string, []>("op_3015_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_3015_end_0 = const()[name = tensor<string, []>("op_3015_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_3015_end_mask_0 = const()[name = tensor<string, []>("op_3015_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3015_cast = slice_by_index(begin = var_3015_begin_0, end = var_3015_end_0, end_mask = var_3015_end_mask_0, x = v_19_cast);
            tensor<int32, [4]> var_3019_begin_0 = const()[name = tensor<string, []>("op_3019_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_3019_end_0 = const()[name = tensor<string, []>("op_3019_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_3019_end_mask_0 = const()[name = tensor<string, []>("op_3019_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3019_cast = slice_by_index(begin = var_3019_begin_0, end = var_3019_end_0, end_mask = var_3019_end_mask_0, x = v_19_cast);
            tensor<int32, [4]> var_3023_begin_0 = const()[name = tensor<string, []>("op_3023_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_3023_end_0 = const()[name = tensor<string, []>("op_3023_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_3023_end_mask_0 = const()[name = tensor<string, []>("op_3023_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3023_cast = slice_by_index(begin = var_3023_begin_0, end = var_3023_end_0, end_mask = var_3023_end_mask_0, x = v_19_cast);
            tensor<int32, [4]> var_3027_begin_0 = const()[name = tensor<string, []>("op_3027_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_3027_end_0 = const()[name = tensor<string, []>("op_3027_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_3027_end_mask_0 = const()[name = tensor<string, []>("op_3027_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3027_cast = slice_by_index(begin = var_3027_begin_0, end = var_3027_end_0, end_mask = var_3027_end_mask_0, x = v_19_cast);
            tensor<int32, [4]> var_3031_begin_0 = const()[name = tensor<string, []>("op_3031_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_3031_end_0 = const()[name = tensor<string, []>("op_3031_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_3031_end_mask_0 = const()[name = tensor<string, []>("op_3031_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3031_cast = slice_by_index(begin = var_3031_begin_0, end = var_3031_end_0, end_mask = var_3031_end_mask_0, x = v_19_cast);
            tensor<int32, [4]> var_3035_begin_0 = const()[name = tensor<string, []>("op_3035_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_3035_end_0 = const()[name = tensor<string, []>("op_3035_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_3035_end_mask_0 = const()[name = tensor<string, []>("op_3035_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3035_cast = slice_by_index(begin = var_3035_begin_0, end = var_3035_end_0, end_mask = var_3035_end_mask_0, x = v_19_cast);
            tensor<int32, [4]> var_3039_begin_0 = const()[name = tensor<string, []>("op_3039_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_3039_end_0 = const()[name = tensor<string, []>("op_3039_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_3039_end_mask_0 = const()[name = tensor<string, []>("op_3039_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3039_cast = slice_by_index(begin = var_3039_begin_0, end = var_3039_end_0, end_mask = var_3039_end_mask_0, x = v_19_cast);
            tensor<int32, [4]> var_3043_begin_0 = const()[name = tensor<string, []>("op_3043_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_3043_end_0 = const()[name = tensor<string, []>("op_3043_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_3043_end_mask_0 = const()[name = tensor<string, []>("op_3043_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3043_cast = slice_by_index(begin = var_3043_begin_0, end = var_3043_end_0, end_mask = var_3043_end_mask_0, x = v_19_cast);
            tensor<int32, [4]> var_3047_begin_0 = const()[name = tensor<string, []>("op_3047_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_3047_end_0 = const()[name = tensor<string, []>("op_3047_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_3047_end_mask_0 = const()[name = tensor<string, []>("op_3047_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3047_cast = slice_by_index(begin = var_3047_begin_0, end = var_3047_end_0, end_mask = var_3047_end_mask_0, x = v_19_cast);
            tensor<int32, [4]> var_3051_begin_0 = const()[name = tensor<string, []>("op_3051_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_3051_end_0 = const()[name = tensor<string, []>("op_3051_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_3051_end_mask_0 = const()[name = tensor<string, []>("op_3051_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3051_cast = slice_by_index(begin = var_3051_begin_0, end = var_3051_end_0, end_mask = var_3051_end_mask_0, x = v_19_cast);
            tensor<string, []> var_3055_equation_0 = const()[name = tensor<string, []>("op_3055_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_3055_cast = einsum(equation = var_3055_equation_0, values = (var_2897_cast, var_2814_cast));
            tensor<fp16, []> var_3056_to_fp16 = const()[name = tensor<string, []>("op_3056_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_161_cast = mul(x = var_3055_cast, y = var_3056_to_fp16);
            tensor<string, []> var_3059_equation_0 = const()[name = tensor<string, []>("op_3059_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_3059_cast = einsum(equation = var_3059_equation_0, values = (var_2901_cast, var_2818_cast));
            tensor<fp16, []> var_3060_to_fp16 = const()[name = tensor<string, []>("op_3060_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_163_cast = mul(x = var_3059_cast, y = var_3060_to_fp16);
            tensor<string, []> var_3063_equation_0 = const()[name = tensor<string, []>("op_3063_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_3063_cast = einsum(equation = var_3063_equation_0, values = (var_2905_cast, var_2822_cast));
            tensor<fp16, []> var_3064_to_fp16 = const()[name = tensor<string, []>("op_3064_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_165_cast = mul(x = var_3063_cast, y = var_3064_to_fp16);
            tensor<string, []> var_3067_equation_0 = const()[name = tensor<string, []>("op_3067_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_3067_cast = einsum(equation = var_3067_equation_0, values = (var_2909_cast, var_2826_cast));
            tensor<fp16, []> var_3068_to_fp16 = const()[name = tensor<string, []>("op_3068_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_167_cast = mul(x = var_3067_cast, y = var_3068_to_fp16);
            tensor<string, []> var_3071_equation_0 = const()[name = tensor<string, []>("op_3071_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_3071_cast = einsum(equation = var_3071_equation_0, values = (var_2913_cast, var_2830_cast));
            tensor<fp16, []> var_3072_to_fp16 = const()[name = tensor<string, []>("op_3072_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_169_cast = mul(x = var_3071_cast, y = var_3072_to_fp16);
            tensor<string, []> var_3075_equation_0 = const()[name = tensor<string, []>("op_3075_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_3075_cast = einsum(equation = var_3075_equation_0, values = (var_2917_cast, var_2834_cast));
            tensor<fp16, []> var_3076_to_fp16 = const()[name = tensor<string, []>("op_3076_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_171_cast = mul(x = var_3075_cast, y = var_3076_to_fp16);
            tensor<string, []> var_3079_equation_0 = const()[name = tensor<string, []>("op_3079_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_3079_cast = einsum(equation = var_3079_equation_0, values = (var_2921_cast, var_2838_cast));
            tensor<fp16, []> var_3080_to_fp16 = const()[name = tensor<string, []>("op_3080_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_173_cast = mul(x = var_3079_cast, y = var_3080_to_fp16);
            tensor<string, []> var_3083_equation_0 = const()[name = tensor<string, []>("op_3083_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_3083_cast = einsum(equation = var_3083_equation_0, values = (var_2925_cast, var_2842_cast));
            tensor<fp16, []> var_3084_to_fp16 = const()[name = tensor<string, []>("op_3084_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_175_cast = mul(x = var_3083_cast, y = var_3084_to_fp16);
            tensor<string, []> var_3087_equation_0 = const()[name = tensor<string, []>("op_3087_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_3087_cast = einsum(equation = var_3087_equation_0, values = (var_2929_cast, var_2846_cast));
            tensor<fp16, []> var_3088_to_fp16 = const()[name = tensor<string, []>("op_3088_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_177_cast = mul(x = var_3087_cast, y = var_3088_to_fp16);
            tensor<string, []> var_3091_equation_0 = const()[name = tensor<string, []>("op_3091_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_3091_cast = einsum(equation = var_3091_equation_0, values = (var_2933_cast, var_2850_cast));
            tensor<fp16, []> var_3092_to_fp16 = const()[name = tensor<string, []>("op_3092_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_179_cast = mul(x = var_3091_cast, y = var_3092_to_fp16);
            tensor<string, []> var_3095_equation_0 = const()[name = tensor<string, []>("op_3095_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_3095_cast = einsum(equation = var_3095_equation_0, values = (var_2937_cast, var_2854_cast));
            tensor<fp16, []> var_3096_to_fp16 = const()[name = tensor<string, []>("op_3096_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_181_cast = mul(x = var_3095_cast, y = var_3096_to_fp16);
            tensor<string, []> var_3099_equation_0 = const()[name = tensor<string, []>("op_3099_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_3099_cast = einsum(equation = var_3099_equation_0, values = (var_2941_cast, var_2858_cast));
            tensor<fp16, []> var_3100_to_fp16 = const()[name = tensor<string, []>("op_3100_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_183_cast = mul(x = var_3099_cast, y = var_3100_to_fp16);
            tensor<string, []> var_3103_equation_0 = const()[name = tensor<string, []>("op_3103_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_3103_cast = einsum(equation = var_3103_equation_0, values = (var_2945_cast, var_2862_cast));
            tensor<fp16, []> var_3104_to_fp16 = const()[name = tensor<string, []>("op_3104_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_185_cast = mul(x = var_3103_cast, y = var_3104_to_fp16);
            tensor<string, []> var_3107_equation_0 = const()[name = tensor<string, []>("op_3107_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_3107_cast = einsum(equation = var_3107_equation_0, values = (var_2949_cast, var_2866_cast));
            tensor<fp16, []> var_3108_to_fp16 = const()[name = tensor<string, []>("op_3108_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_187_cast = mul(x = var_3107_cast, y = var_3108_to_fp16);
            tensor<string, []> var_3111_equation_0 = const()[name = tensor<string, []>("op_3111_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_3111_cast = einsum(equation = var_3111_equation_0, values = (var_2953_cast, var_2870_cast));
            tensor<fp16, []> var_3112_to_fp16 = const()[name = tensor<string, []>("op_3112_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_189_cast = mul(x = var_3111_cast, y = var_3112_to_fp16);
            tensor<string, []> var_3115_equation_0 = const()[name = tensor<string, []>("op_3115_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_3115_cast = einsum(equation = var_3115_equation_0, values = (var_2957_cast, var_2874_cast));
            tensor<fp16, []> var_3116_to_fp16 = const()[name = tensor<string, []>("op_3116_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_191_cast = mul(x = var_3115_cast, y = var_3116_to_fp16);
            tensor<string, []> var_3119_equation_0 = const()[name = tensor<string, []>("op_3119_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_3119_cast = einsum(equation = var_3119_equation_0, values = (var_2961_cast, var_2878_cast));
            tensor<fp16, []> var_3120_to_fp16 = const()[name = tensor<string, []>("op_3120_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_193_cast = mul(x = var_3119_cast, y = var_3120_to_fp16);
            tensor<string, []> var_3123_equation_0 = const()[name = tensor<string, []>("op_3123_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_3123_cast = einsum(equation = var_3123_equation_0, values = (var_2965_cast, var_2882_cast));
            tensor<fp16, []> var_3124_to_fp16 = const()[name = tensor<string, []>("op_3124_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_195_cast = mul(x = var_3123_cast, y = var_3124_to_fp16);
            tensor<string, []> var_3127_equation_0 = const()[name = tensor<string, []>("op_3127_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_3127_cast = einsum(equation = var_3127_equation_0, values = (var_2969_cast, var_2886_cast));
            tensor<fp16, []> var_3128_to_fp16 = const()[name = tensor<string, []>("op_3128_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_197_cast = mul(x = var_3127_cast, y = var_3128_to_fp16);
            tensor<string, []> var_3131_equation_0 = const()[name = tensor<string, []>("op_3131_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_3131_cast = einsum(equation = var_3131_equation_0, values = (var_2973_cast, var_2890_cast));
            tensor<fp16, []> var_3132_to_fp16 = const()[name = tensor<string, []>("op_3132_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_199_cast = mul(x = var_3131_cast, y = var_3132_to_fp16);
            tensor<fp16, [2, 77, 1, 256]> var_3134_cast = softmax(axis = var_2254, x = aw_161_cast);
            tensor<fp16, [2, 77, 1, 256]> var_3135_cast = softmax(axis = var_2254, x = aw_163_cast);
            tensor<fp16, [2, 77, 1, 256]> var_3136_cast = softmax(axis = var_2254, x = aw_165_cast);
            tensor<fp16, [2, 77, 1, 256]> var_3137_cast = softmax(axis = var_2254, x = aw_167_cast);
            tensor<fp16, [2, 77, 1, 256]> var_3138_cast = softmax(axis = var_2254, x = aw_169_cast);
            tensor<fp16, [2, 77, 1, 256]> var_3139_cast = softmax(axis = var_2254, x = aw_171_cast);
            tensor<fp16, [2, 77, 1, 256]> var_3140_cast = softmax(axis = var_2254, x = aw_173_cast);
            tensor<fp16, [2, 77, 1, 256]> var_3141_cast = softmax(axis = var_2254, x = aw_175_cast);
            tensor<fp16, [2, 77, 1, 256]> var_3142_cast = softmax(axis = var_2254, x = aw_177_cast);
            tensor<fp16, [2, 77, 1, 256]> var_3143_cast = softmax(axis = var_2254, x = aw_179_cast);
            tensor<fp16, [2, 77, 1, 256]> var_3144_cast = softmax(axis = var_2254, x = aw_181_cast);
            tensor<fp16, [2, 77, 1, 256]> var_3145_cast = softmax(axis = var_2254, x = aw_183_cast);
            tensor<fp16, [2, 77, 1, 256]> var_3146_cast = softmax(axis = var_2254, x = aw_185_cast);
            tensor<fp16, [2, 77, 1, 256]> var_3147_cast = softmax(axis = var_2254, x = aw_187_cast);
            tensor<fp16, [2, 77, 1, 256]> var_3148_cast = softmax(axis = var_2254, x = aw_189_cast);
            tensor<fp16, [2, 77, 1, 256]> var_3149_cast = softmax(axis = var_2254, x = aw_191_cast);
            tensor<fp16, [2, 77, 1, 256]> var_3150_cast = softmax(axis = var_2254, x = aw_193_cast);
            tensor<fp16, [2, 77, 1, 256]> var_3151_cast = softmax(axis = var_2254, x = aw_195_cast);
            tensor<fp16, [2, 77, 1, 256]> var_3152_cast = softmax(axis = var_2254, x = aw_197_cast);
            tensor<fp16, [2, 77, 1, 256]> var_3153_cast = softmax(axis = var_2254, x = aw_199_cast);
            tensor<string, []> var_3155_equation_0 = const()[name = tensor<string, []>("op_3155_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3155_cast = einsum(equation = var_3155_equation_0, values = (var_2975_cast, var_3134_cast));
            tensor<string, []> var_3157_equation_0 = const()[name = tensor<string, []>("op_3157_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3157_cast = einsum(equation = var_3157_equation_0, values = (var_2979_cast, var_3135_cast));
            tensor<string, []> var_3159_equation_0 = const()[name = tensor<string, []>("op_3159_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3159_cast = einsum(equation = var_3159_equation_0, values = (var_2983_cast, var_3136_cast));
            tensor<string, []> var_3161_equation_0 = const()[name = tensor<string, []>("op_3161_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3161_cast = einsum(equation = var_3161_equation_0, values = (var_2987_cast, var_3137_cast));
            tensor<string, []> var_3163_equation_0 = const()[name = tensor<string, []>("op_3163_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3163_cast = einsum(equation = var_3163_equation_0, values = (var_2991_cast, var_3138_cast));
            tensor<string, []> var_3165_equation_0 = const()[name = tensor<string, []>("op_3165_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3165_cast = einsum(equation = var_3165_equation_0, values = (var_2995_cast, var_3139_cast));
            tensor<string, []> var_3167_equation_0 = const()[name = tensor<string, []>("op_3167_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3167_cast = einsum(equation = var_3167_equation_0, values = (var_2999_cast, var_3140_cast));
            tensor<string, []> var_3169_equation_0 = const()[name = tensor<string, []>("op_3169_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3169_cast = einsum(equation = var_3169_equation_0, values = (var_3003_cast, var_3141_cast));
            tensor<string, []> var_3171_equation_0 = const()[name = tensor<string, []>("op_3171_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3171_cast = einsum(equation = var_3171_equation_0, values = (var_3007_cast, var_3142_cast));
            tensor<string, []> var_3173_equation_0 = const()[name = tensor<string, []>("op_3173_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3173_cast = einsum(equation = var_3173_equation_0, values = (var_3011_cast, var_3143_cast));
            tensor<string, []> var_3175_equation_0 = const()[name = tensor<string, []>("op_3175_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3175_cast = einsum(equation = var_3175_equation_0, values = (var_3015_cast, var_3144_cast));
            tensor<string, []> var_3177_equation_0 = const()[name = tensor<string, []>("op_3177_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3177_cast = einsum(equation = var_3177_equation_0, values = (var_3019_cast, var_3145_cast));
            tensor<string, []> var_3179_equation_0 = const()[name = tensor<string, []>("op_3179_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3179_cast = einsum(equation = var_3179_equation_0, values = (var_3023_cast, var_3146_cast));
            tensor<string, []> var_3181_equation_0 = const()[name = tensor<string, []>("op_3181_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3181_cast = einsum(equation = var_3181_equation_0, values = (var_3027_cast, var_3147_cast));
            tensor<string, []> var_3183_equation_0 = const()[name = tensor<string, []>("op_3183_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3183_cast = einsum(equation = var_3183_equation_0, values = (var_3031_cast, var_3148_cast));
            tensor<string, []> var_3185_equation_0 = const()[name = tensor<string, []>("op_3185_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3185_cast = einsum(equation = var_3185_equation_0, values = (var_3035_cast, var_3149_cast));
            tensor<string, []> var_3187_equation_0 = const()[name = tensor<string, []>("op_3187_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3187_cast = einsum(equation = var_3187_equation_0, values = (var_3039_cast, var_3150_cast));
            tensor<string, []> var_3189_equation_0 = const()[name = tensor<string, []>("op_3189_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3189_cast = einsum(equation = var_3189_equation_0, values = (var_3043_cast, var_3151_cast));
            tensor<string, []> var_3191_equation_0 = const()[name = tensor<string, []>("op_3191_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3191_cast = einsum(equation = var_3191_equation_0, values = (var_3047_cast, var_3152_cast));
            tensor<string, []> var_3193_equation_0 = const()[name = tensor<string, []>("op_3193_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3193_cast = einsum(equation = var_3193_equation_0, values = (var_3051_cast, var_3153_cast));
            tensor<bool, []> input_135_interleave_0 = const()[name = tensor<string, []>("input_135_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 256]> input_135_cast = concat(axis = var_2254, interleave = input_135_interleave_0, values = (var_3155_cast, var_3157_cast, var_3159_cast, var_3161_cast, var_3163_cast, var_3165_cast, var_3167_cast, var_3169_cast, var_3171_cast, var_3173_cast, var_3175_cast, var_3177_cast, var_3179_cast, var_3181_cast, var_3183_cast, var_3185_cast, var_3187_cast, var_3189_cast, var_3191_cast, var_3193_cast));
            tensor<int32, [2]> var_3199 = const()[name = tensor<string, []>("op_3199"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3201 = const()[name = tensor<string, []>("op_3201"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_3203_pad_type_0 = const()[name = tensor<string, []>("op_3203_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_3203_pad_0 = const()[name = tensor<string, []>("op_3203_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(174919744)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178196608)))];
            tensor<fp16, [2, 1280, 1, 256]> var_3203_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_3201, groups = var_2254, pad = var_3203_pad_0, pad_type = var_3203_pad_type_0, strides = var_3199, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_135_cast);
            tensor<fp16, [2, 1280, 1, 256]> inputs_29_cast = add(x = var_3203_cast, y = inputs_27_cast);
            tensor<int32, [1]> var_3207 = const()[name = tensor<string, []>("op_3207"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_29_cast = reduce_mean(axes = var_3207, keep_dims = var_2249, x = inputs_29_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_29_cast = sub(x = inputs_29_cast, y = channels_mean_29_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_sq_29_cast = mul(x = zero_mean_29_cast, y = zero_mean_29_cast);
            tensor<int32, [1]> var_3211 = const()[name = tensor<string, []>("op_3211"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_3212_cast = reduce_mean(axes = var_3211, keep_dims = var_2249, x = zero_mean_sq_29_cast);
            tensor<fp16, []> var_3213_to_fp16 = const()[name = tensor<string, []>("op_3213_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_3214_cast = add(x = var_3212_cast, y = var_3213_to_fp16);
            tensor<fp16, []> denom_29_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_29_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 256]> denom_29_cast = rsqrt(epsilon = denom_29_epsilon_0_to_fp16, x = var_3214_cast);
            tensor<fp16, [2, 1280, 1, 256]> out_29_cast = mul(x = zero_mean_29_cast, y = denom_29_cast);
            tensor<fp16, [1, 1280, 1, 1]> var_3218_to_fp16 = const()[name = tensor<string, []>("op_3218_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178199232)))];
            tensor<fp16, [2, 1280, 1, 256]> var_3219_cast = add(x = out_29_cast, y = var_3218_to_fp16);
            tensor<fp16, [1, 1280, 1, 1]> var_3221_to_fp16 = const()[name = tensor<string, []>("op_3221_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178201856)))];
            tensor<fp16, [2, 1280, 1, 256]> input_137_cast = mul(x = var_3219_cast, y = var_3221_to_fp16);
            tensor<int32, [2]> var_3229 = const()[name = tensor<string, []>("op_3229"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3231 = const()[name = tensor<string, []>("op_3231"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_3233_pad_type_0 = const()[name = tensor<string, []>("op_3233_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_3233_pad_0 = const()[name = tensor<string, []>("op_3233_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178204480)))];
            tensor<fp16, [10240]> down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(204418944)))];
            tensor<fp16, [2, 10240, 1, 256]> var_3233_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_3231, groups = var_2254, pad = var_3233_pad_0, pad_type = var_3233_pad_type_0, strides = var_3229, weight = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_137_cast);
            tensor<int32, [2]> var_3234_split_sizes_0 = const()[name = tensor<string, []>("op_3234_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_3234_axis_0 = const()[name = tensor<string, []>("op_3234_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 256]> var_3234_cast_0, tensor<fp16, [2, 5120, 1, 256]> var_3234_cast_1 = split(axis = var_3234_axis_0, split_sizes = var_3234_split_sizes_0, x = var_3233_cast);
            tensor<string, []> var_3236_mode_0 = const()[name = tensor<string, []>("op_3236_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 256]> var_3236_cast = gelu(mode = var_3236_mode_0, x = var_3234_cast_1);
            tensor<fp16, [2, 5120, 1, 256]> input_139_cast = mul(x = var_3234_cast_0, y = var_3236_cast);
            tensor<int32, [2]> var_3240 = const()[name = tensor<string, []>("op_3240"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3242 = const()[name = tensor<string, []>("op_3242"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_3244_pad_type_0 = const()[name = tensor<string, []>("op_3244_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_3244_pad_0 = const()[name = tensor<string, []>("op_3244_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(204439488)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217546752)))];
            tensor<fp16, [2, 1280, 1, 256]> var_3244_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_3242, groups = var_2254, pad = var_3244_pad_0, pad_type = var_3244_pad_type_0, strides = var_3240, weight = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_139_cast);
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_87_cast = add(x = var_3244_cast, y = inputs_29_cast);
            tensor<int32, [4]> var_3246 = const()[name = tensor<string, []>("op_3246"), val = tensor<int32, [4]>([2, 1280, 16, 16])];
            tensor<fp16, [2, 1280, 16, 16]> input_141_cast = reshape(shape = var_3246, x = hidden_states_87_cast);
            tensor<int32, [2]> var_3250 = const()[name = tensor<string, []>("op_3250"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3252 = const()[name = tensor<string, []>("op_3252"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_89_pad_type_0 = const()[name = tensor<string, []>("hidden_states_89_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_89_pad_0 = const()[name = tensor<string, []>("hidden_states_89_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_0_proj_out_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_proj_out_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217549376)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_0_proj_out_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(220826240)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_89_cast = conv(bias = down_blocks_2_attentions_0_proj_out_bias_to_fp16, dilations = var_3252, groups = var_2254, pad = hidden_states_89_pad_0, pad_type = hidden_states_89_pad_type_0, strides = var_3250, weight = down_blocks_2_attentions_0_proj_out_weight_to_fp16, x = input_141_cast);
            tensor<fp16, [2, 1280, 16, 16]> input_143_cast = add(x = hidden_states_89_cast, y = hidden_states_77_cast);
            tensor<int32, [5]> reshape_60_shape_0 = const()[name = tensor<string, []>("reshape_60_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 16, 16])];
            tensor<fp16, [2, 32, 40, 16, 16]> reshape_60_cast = reshape(shape = reshape_60_shape_0, x = input_143_cast);
            tensor<int32, [3]> reduce_mean_45_axes_0 = const()[name = tensor<string, []>("reduce_mean_45_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_45_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_45_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_45_cast = reduce_mean(axes = reduce_mean_45_axes_0, keep_dims = reduce_mean_45_keep_dims_0, x = reshape_60_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> sub_30_cast = sub(x = reshape_60_cast, y = reduce_mean_45_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> square_15_cast = square(x = sub_30_cast);
            tensor<int32, [3]> reduce_mean_47_axes_0 = const()[name = tensor<string, []>("reduce_mean_47_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_47_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_47_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_47_cast = reduce_mean(axes = reduce_mean_47_axes_0, keep_dims = reduce_mean_47_keep_dims_0, x = square_15_cast);
            tensor<fp16, []> add_30_y_0_to_fp16 = const()[name = tensor<string, []>("add_30_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_30_cast = add(x = reduce_mean_47_cast, y = add_30_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_15_cast = sqrt(x = add_30_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> real_div_15_cast = real_div(x = sub_30_cast, y = sqrt_15_cast);
            tensor<int32, [4]> reshape_61_shape_0 = const()[name = tensor<string, []>("reshape_61_shape_0"), val = tensor<int32, [4]>([2, 1280, 16, 16])];
            tensor<fp16, [2, 1280, 16, 16]> reshape_61_cast = reshape(shape = reshape_61_shape_0, x = real_div_15_cast);
            tensor<fp16, [1280]> add_31_mean_0_to_fp16 = const()[name = tensor<string, []>("add_31_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(220828864)))];
            tensor<fp16, [1280]> add_31_variance_0_to_fp16 = const()[name = tensor<string, []>("add_31_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(220831488)))];
            tensor<fp16, [1280]> add_31_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_31_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(220834112)))];
            tensor<fp16, [1280]> add_31_beta_0_to_fp16 = const()[name = tensor<string, []>("add_31_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(220836736)))];
            tensor<fp16, []> add_31_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_31_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 16, 16]> add_31_cast = batch_norm(beta = add_31_beta_0_to_fp16, epsilon = add_31_epsilon_0_to_fp16, gamma = add_31_gamma_0_to_fp16, mean = add_31_mean_0_to_fp16, variance = add_31_variance_0_to_fp16, x = reshape_61_cast);
            tensor<fp16, [2, 1280, 16, 16]> input_147_cast = silu(x = add_31_cast);
            tensor<int32, [2]> var_3267 = const()[name = tensor<string, []>("op_3267"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3269 = const()[name = tensor<string, []>("op_3269"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_91_pad_type_0 = const()[name = tensor<string, []>("hidden_states_91_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_91_pad_0 = const()[name = tensor<string, []>("hidden_states_91_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> down_blocks_2_resnets_1_conv1_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_resnets_1_conv1_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(220839360)))];
            tensor<fp16, [1280]> down_blocks_2_resnets_1_conv1_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_resnets_1_conv1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(250330624)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_91_cast = conv(bias = down_blocks_2_resnets_1_conv1_bias_to_fp16, dilations = var_3269, groups = var_2254, pad = hidden_states_91_pad_0, pad_type = hidden_states_91_pad_type_0, strides = var_3267, weight = down_blocks_2_resnets_1_conv1_weight_to_fp16, x = input_147_cast);
            tensor<int32, [2]> var_3275 = const()[name = tensor<string, []>("op_3275"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3277 = const()[name = tensor<string, []>("op_3277"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_11_pad_type_0 = const()[name = tensor<string, []>("temb_11_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_11_pad_0 = const()[name = tensor<string, []>("temb_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(250333248)))];
            tensor<fp16, [1280]> down_blocks_2_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(253610112)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_11_cast = conv(bias = down_blocks_2_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_3277, groups = var_2254, pad = temb_11_pad_0, pad_type = temb_11_pad_type_0, strides = var_3275, weight = down_blocks_2_resnets_1_time_emb_proj_weight_to_fp16, x = input_15_cast);
            tensor<fp16, [2, 1280, 16, 16]> input_151_cast = add(x = hidden_states_91_cast, y = temb_11_cast);
            tensor<int32, [5]> reshape_64_shape_0 = const()[name = tensor<string, []>("reshape_64_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 16, 16])];
            tensor<fp16, [2, 32, 40, 16, 16]> reshape_64_cast = reshape(shape = reshape_64_shape_0, x = input_151_cast);
            tensor<int32, [3]> reduce_mean_48_axes_0 = const()[name = tensor<string, []>("reduce_mean_48_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_48_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_48_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_48_cast = reduce_mean(axes = reduce_mean_48_axes_0, keep_dims = reduce_mean_48_keep_dims_0, x = reshape_64_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> sub_32_cast = sub(x = reshape_64_cast, y = reduce_mean_48_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> square_16_cast = square(x = sub_32_cast);
            tensor<int32, [3]> reduce_mean_50_axes_0 = const()[name = tensor<string, []>("reduce_mean_50_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_50_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_50_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_50_cast = reduce_mean(axes = reduce_mean_50_axes_0, keep_dims = reduce_mean_50_keep_dims_0, x = square_16_cast);
            tensor<fp16, []> add_32_y_0_to_fp16 = const()[name = tensor<string, []>("add_32_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_32_cast = add(x = reduce_mean_50_cast, y = add_32_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_16_cast = sqrt(x = add_32_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> real_div_16_cast = real_div(x = sub_32_cast, y = sqrt_16_cast);
            tensor<int32, [4]> reshape_65_shape_0 = const()[name = tensor<string, []>("reshape_65_shape_0"), val = tensor<int32, [4]>([2, 1280, 16, 16])];
            tensor<fp16, [2, 1280, 16, 16]> reshape_65_cast = reshape(shape = reshape_65_shape_0, x = real_div_16_cast);
            tensor<fp16, [1280]> add_33_mean_0_to_fp16 = const()[name = tensor<string, []>("add_33_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(253612736)))];
            tensor<fp16, [1280]> add_33_variance_0_to_fp16 = const()[name = tensor<string, []>("add_33_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(253615360)))];
            tensor<fp16, [1280]> add_33_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_33_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(253617984)))];
            tensor<fp16, [1280]> add_33_beta_0_to_fp16 = const()[name = tensor<string, []>("add_33_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(253620608)))];
            tensor<fp16, []> add_33_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_33_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 16, 16]> add_33_cast = batch_norm(beta = add_33_beta_0_to_fp16, epsilon = add_33_epsilon_0_to_fp16, gamma = add_33_gamma_0_to_fp16, mean = add_33_mean_0_to_fp16, variance = add_33_variance_0_to_fp16, x = reshape_65_cast);
            tensor<fp16, [2, 1280, 16, 16]> input_155_cast = silu(x = add_33_cast);
            tensor<int32, [2]> var_3287 = const()[name = tensor<string, []>("op_3287"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3289 = const()[name = tensor<string, []>("op_3289"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_93_pad_type_0 = const()[name = tensor<string, []>("hidden_states_93_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_93_pad_0 = const()[name = tensor<string, []>("hidden_states_93_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> down_blocks_2_resnets_1_conv2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_resnets_1_conv2_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(253623232)))];
            tensor<fp16, [1280]> down_blocks_2_resnets_1_conv2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_resnets_1_conv2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(283114496)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_93_cast = conv(bias = down_blocks_2_resnets_1_conv2_bias_to_fp16, dilations = var_3289, groups = var_2254, pad = hidden_states_93_pad_0, pad_type = hidden_states_93_pad_type_0, strides = var_3287, weight = down_blocks_2_resnets_1_conv2_weight_to_fp16, x = input_155_cast);
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_95_cast = add(x = input_143_cast, y = hidden_states_93_cast);
            tensor<int32, [5]> reshape_68_shape_0 = const()[name = tensor<string, []>("reshape_68_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 16, 16])];
            tensor<fp16, [2, 32, 40, 16, 16]> reshape_68_cast = reshape(shape = reshape_68_shape_0, x = hidden_states_95_cast);
            tensor<int32, [3]> reduce_mean_51_axes_0 = const()[name = tensor<string, []>("reduce_mean_51_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_51_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_51_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_51_cast = reduce_mean(axes = reduce_mean_51_axes_0, keep_dims = reduce_mean_51_keep_dims_0, x = reshape_68_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> sub_34_cast = sub(x = reshape_68_cast, y = reduce_mean_51_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> square_17_cast = square(x = sub_34_cast);
            tensor<int32, [3]> reduce_mean_53_axes_0 = const()[name = tensor<string, []>("reduce_mean_53_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_53_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_53_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_53_cast = reduce_mean(axes = reduce_mean_53_axes_0, keep_dims = reduce_mean_53_keep_dims_0, x = square_17_cast);
            tensor<fp16, []> add_34_y_0_to_fp16 = const()[name = tensor<string, []>("add_34_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_34_cast = add(x = reduce_mean_53_cast, y = add_34_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_17_cast = sqrt(x = add_34_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> real_div_17_cast = real_div(x = sub_34_cast, y = sqrt_17_cast);
            tensor<int32, [4]> reshape_69_shape_0 = const()[name = tensor<string, []>("reshape_69_shape_0"), val = tensor<int32, [4]>([2, 1280, 16, 16])];
            tensor<fp16, [2, 1280, 16, 16]> reshape_69_cast = reshape(shape = reshape_69_shape_0, x = real_div_17_cast);
            tensor<fp16, [1280]> add_35_mean_0_to_fp16 = const()[name = tensor<string, []>("add_35_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(283117120)))];
            tensor<fp16, [1280]> add_35_variance_0_to_fp16 = const()[name = tensor<string, []>("add_35_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(283119744)))];
            tensor<fp16, [1280]> add_35_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_35_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(283122368)))];
            tensor<fp16, [1280]> add_35_beta_0_to_fp16 = const()[name = tensor<string, []>("add_35_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(283124992)))];
            tensor<fp16, []> add_35_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_35_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 16, 16]> add_35_cast = batch_norm(beta = add_35_beta_0_to_fp16, epsilon = add_35_epsilon_0_to_fp16, gamma = add_35_gamma_0_to_fp16, mean = add_35_mean_0_to_fp16, variance = add_35_variance_0_to_fp16, x = reshape_69_cast);
            tensor<int32, [2]> var_3309 = const()[name = tensor<string, []>("op_3309"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3311 = const()[name = tensor<string, []>("op_3311"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_97_pad_type_0 = const()[name = tensor<string, []>("hidden_states_97_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_97_pad_0 = const()[name = tensor<string, []>("hidden_states_97_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_proj_in_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_proj_in_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(283127616)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_proj_in_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(286404480)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_97_cast = conv(bias = down_blocks_2_attentions_1_proj_in_bias_to_fp16, dilations = var_3311, groups = var_2254, pad = hidden_states_97_pad_0, pad_type = hidden_states_97_pad_type_0, strides = var_3309, weight = down_blocks_2_attentions_1_proj_in_weight_to_fp16, x = add_35_cast);
            tensor<int32, [4]> var_3316 = const()[name = tensor<string, []>("op_3316"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<fp16, [2, 1280, 1, 256]> inputs_31_cast = reshape(shape = var_3316, x = hidden_states_97_cast);
            tensor<int32, [1]> var_3326 = const()[name = tensor<string, []>("op_3326"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_31_cast = reduce_mean(axes = var_3326, keep_dims = var_2249, x = inputs_31_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_31_cast = sub(x = inputs_31_cast, y = channels_mean_31_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_sq_31_cast = mul(x = zero_mean_31_cast, y = zero_mean_31_cast);
            tensor<int32, [1]> var_3330 = const()[name = tensor<string, []>("op_3330"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_3331_cast = reduce_mean(axes = var_3330, keep_dims = var_2249, x = zero_mean_sq_31_cast);
            tensor<fp16, []> var_3332_to_fp16 = const()[name = tensor<string, []>("op_3332_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_3333_cast = add(x = var_3331_cast, y = var_3332_to_fp16);
            tensor<fp16, []> denom_31_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_31_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 256]> denom_31_cast = rsqrt(epsilon = denom_31_epsilon_0_to_fp16, x = var_3333_cast);
            tensor<fp16, [2, 1280, 1, 256]> out_31_cast = mul(x = zero_mean_31_cast, y = denom_31_cast);
            tensor<fp16, [1, 1280, 1, 1]> var_3337_to_fp16 = const()[name = tensor<string, []>("op_3337_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(286407104)))];
            tensor<fp16, [2, 1280, 1, 256]> var_3338_cast = add(x = out_31_cast, y = var_3337_to_fp16);
            tensor<fp16, [1, 1280, 1, 1]> var_3340_to_fp16 = const()[name = tensor<string, []>("op_3340_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(286409728)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_99_cast = mul(x = var_3338_cast, y = var_3340_to_fp16);
            tensor<int32, [2]> var_3347 = const()[name = tensor<string, []>("op_3347"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3349 = const()[name = tensor<string, []>("op_3349"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_21_pad_type_0 = const()[name = tensor<string, []>("q_21_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_21_pad_0 = const()[name = tensor<string, []>("q_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(286412352)))];
            tensor<fp16, [2, 1280, 1, 256]> q_21_cast = conv(dilations = var_3349, groups = var_2254, pad = q_21_pad_0, pad_type = q_21_pad_type_0, strides = var_3347, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_99_cast);
            tensor<int32, [2]> var_3353 = const()[name = tensor<string, []>("op_3353"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3355 = const()[name = tensor<string, []>("op_3355"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_41_pad_type_0 = const()[name = tensor<string, []>("k_41_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_41_pad_0 = const()[name = tensor<string, []>("k_41_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(289689216)))];
            tensor<fp16, [2, 1280, 1, 256]> k_41_cast = conv(dilations = var_3355, groups = var_2254, pad = k_41_pad_0, pad_type = k_41_pad_type_0, strides = var_3353, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_99_cast);
            tensor<int32, [2]> var_3359 = const()[name = tensor<string, []>("op_3359"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3361 = const()[name = tensor<string, []>("op_3361"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_21_pad_type_0 = const()[name = tensor<string, []>("v_21_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_21_pad_0 = const()[name = tensor<string, []>("v_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(292966080)))];
            tensor<fp16, [2, 1280, 1, 256]> v_21_cast = conv(dilations = var_3361, groups = var_2254, pad = v_21_pad_0, pad_type = v_21_pad_type_0, strides = var_3359, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_99_cast);
            tensor<int32, [4]> var_3365_begin_0 = const()[name = tensor<string, []>("op_3365_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3365_end_0 = const()[name = tensor<string, []>("op_3365_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_3365_end_mask_0 = const()[name = tensor<string, []>("op_3365_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3365_cast = slice_by_index(begin = var_3365_begin_0, end = var_3365_end_0, end_mask = var_3365_end_mask_0, x = q_21_cast);
            tensor<int32, [4]> var_3369_begin_0 = const()[name = tensor<string, []>("op_3369_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_3369_end_0 = const()[name = tensor<string, []>("op_3369_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_3369_end_mask_0 = const()[name = tensor<string, []>("op_3369_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3369_cast = slice_by_index(begin = var_3369_begin_0, end = var_3369_end_0, end_mask = var_3369_end_mask_0, x = q_21_cast);
            tensor<int32, [4]> var_3373_begin_0 = const()[name = tensor<string, []>("op_3373_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_3373_end_0 = const()[name = tensor<string, []>("op_3373_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_3373_end_mask_0 = const()[name = tensor<string, []>("op_3373_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3373_cast = slice_by_index(begin = var_3373_begin_0, end = var_3373_end_0, end_mask = var_3373_end_mask_0, x = q_21_cast);
            tensor<int32, [4]> var_3377_begin_0 = const()[name = tensor<string, []>("op_3377_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_3377_end_0 = const()[name = tensor<string, []>("op_3377_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_3377_end_mask_0 = const()[name = tensor<string, []>("op_3377_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3377_cast = slice_by_index(begin = var_3377_begin_0, end = var_3377_end_0, end_mask = var_3377_end_mask_0, x = q_21_cast);
            tensor<int32, [4]> var_3381_begin_0 = const()[name = tensor<string, []>("op_3381_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_3381_end_0 = const()[name = tensor<string, []>("op_3381_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_3381_end_mask_0 = const()[name = tensor<string, []>("op_3381_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3381_cast = slice_by_index(begin = var_3381_begin_0, end = var_3381_end_0, end_mask = var_3381_end_mask_0, x = q_21_cast);
            tensor<int32, [4]> var_3385_begin_0 = const()[name = tensor<string, []>("op_3385_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_3385_end_0 = const()[name = tensor<string, []>("op_3385_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_3385_end_mask_0 = const()[name = tensor<string, []>("op_3385_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3385_cast = slice_by_index(begin = var_3385_begin_0, end = var_3385_end_0, end_mask = var_3385_end_mask_0, x = q_21_cast);
            tensor<int32, [4]> var_3389_begin_0 = const()[name = tensor<string, []>("op_3389_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_3389_end_0 = const()[name = tensor<string, []>("op_3389_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_3389_end_mask_0 = const()[name = tensor<string, []>("op_3389_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3389_cast = slice_by_index(begin = var_3389_begin_0, end = var_3389_end_0, end_mask = var_3389_end_mask_0, x = q_21_cast);
            tensor<int32, [4]> var_3393_begin_0 = const()[name = tensor<string, []>("op_3393_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_3393_end_0 = const()[name = tensor<string, []>("op_3393_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_3393_end_mask_0 = const()[name = tensor<string, []>("op_3393_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3393_cast = slice_by_index(begin = var_3393_begin_0, end = var_3393_end_0, end_mask = var_3393_end_mask_0, x = q_21_cast);
            tensor<int32, [4]> var_3397_begin_0 = const()[name = tensor<string, []>("op_3397_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_3397_end_0 = const()[name = tensor<string, []>("op_3397_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_3397_end_mask_0 = const()[name = tensor<string, []>("op_3397_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3397_cast = slice_by_index(begin = var_3397_begin_0, end = var_3397_end_0, end_mask = var_3397_end_mask_0, x = q_21_cast);
            tensor<int32, [4]> var_3401_begin_0 = const()[name = tensor<string, []>("op_3401_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_3401_end_0 = const()[name = tensor<string, []>("op_3401_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_3401_end_mask_0 = const()[name = tensor<string, []>("op_3401_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3401_cast = slice_by_index(begin = var_3401_begin_0, end = var_3401_end_0, end_mask = var_3401_end_mask_0, x = q_21_cast);
            tensor<int32, [4]> var_3405_begin_0 = const()[name = tensor<string, []>("op_3405_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_3405_end_0 = const()[name = tensor<string, []>("op_3405_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_3405_end_mask_0 = const()[name = tensor<string, []>("op_3405_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3405_cast = slice_by_index(begin = var_3405_begin_0, end = var_3405_end_0, end_mask = var_3405_end_mask_0, x = q_21_cast);
            tensor<int32, [4]> var_3409_begin_0 = const()[name = tensor<string, []>("op_3409_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_3409_end_0 = const()[name = tensor<string, []>("op_3409_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_3409_end_mask_0 = const()[name = tensor<string, []>("op_3409_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3409_cast = slice_by_index(begin = var_3409_begin_0, end = var_3409_end_0, end_mask = var_3409_end_mask_0, x = q_21_cast);
            tensor<int32, [4]> var_3413_begin_0 = const()[name = tensor<string, []>("op_3413_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_3413_end_0 = const()[name = tensor<string, []>("op_3413_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_3413_end_mask_0 = const()[name = tensor<string, []>("op_3413_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3413_cast = slice_by_index(begin = var_3413_begin_0, end = var_3413_end_0, end_mask = var_3413_end_mask_0, x = q_21_cast);
            tensor<int32, [4]> var_3417_begin_0 = const()[name = tensor<string, []>("op_3417_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_3417_end_0 = const()[name = tensor<string, []>("op_3417_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_3417_end_mask_0 = const()[name = tensor<string, []>("op_3417_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3417_cast = slice_by_index(begin = var_3417_begin_0, end = var_3417_end_0, end_mask = var_3417_end_mask_0, x = q_21_cast);
            tensor<int32, [4]> var_3421_begin_0 = const()[name = tensor<string, []>("op_3421_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_3421_end_0 = const()[name = tensor<string, []>("op_3421_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_3421_end_mask_0 = const()[name = tensor<string, []>("op_3421_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3421_cast = slice_by_index(begin = var_3421_begin_0, end = var_3421_end_0, end_mask = var_3421_end_mask_0, x = q_21_cast);
            tensor<int32, [4]> var_3425_begin_0 = const()[name = tensor<string, []>("op_3425_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_3425_end_0 = const()[name = tensor<string, []>("op_3425_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_3425_end_mask_0 = const()[name = tensor<string, []>("op_3425_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3425_cast = slice_by_index(begin = var_3425_begin_0, end = var_3425_end_0, end_mask = var_3425_end_mask_0, x = q_21_cast);
            tensor<int32, [4]> var_3429_begin_0 = const()[name = tensor<string, []>("op_3429_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_3429_end_0 = const()[name = tensor<string, []>("op_3429_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_3429_end_mask_0 = const()[name = tensor<string, []>("op_3429_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3429_cast = slice_by_index(begin = var_3429_begin_0, end = var_3429_end_0, end_mask = var_3429_end_mask_0, x = q_21_cast);
            tensor<int32, [4]> var_3433_begin_0 = const()[name = tensor<string, []>("op_3433_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_3433_end_0 = const()[name = tensor<string, []>("op_3433_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_3433_end_mask_0 = const()[name = tensor<string, []>("op_3433_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3433_cast = slice_by_index(begin = var_3433_begin_0, end = var_3433_end_0, end_mask = var_3433_end_mask_0, x = q_21_cast);
            tensor<int32, [4]> var_3437_begin_0 = const()[name = tensor<string, []>("op_3437_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_3437_end_0 = const()[name = tensor<string, []>("op_3437_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_3437_end_mask_0 = const()[name = tensor<string, []>("op_3437_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3437_cast = slice_by_index(begin = var_3437_begin_0, end = var_3437_end_0, end_mask = var_3437_end_mask_0, x = q_21_cast);
            tensor<int32, [4]> var_3441_begin_0 = const()[name = tensor<string, []>("op_3441_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_3441_end_0 = const()[name = tensor<string, []>("op_3441_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_3441_end_mask_0 = const()[name = tensor<string, []>("op_3441_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3441_cast = slice_by_index(begin = var_3441_begin_0, end = var_3441_end_0, end_mask = var_3441_end_mask_0, x = q_21_cast);
            tensor<int32, [4]> k_43_perm_0 = const()[name = tensor<string, []>("k_43_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_3448_begin_0 = const()[name = tensor<string, []>("op_3448_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3448_end_0 = const()[name = tensor<string, []>("op_3448_end_0"), val = tensor<int32, [4]>([2, 256, 1, 64])];
            tensor<bool, [4]> var_3448_end_mask_0 = const()[name = tensor<string, []>("op_3448_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 1280]> transpose_21 = transpose(perm = k_43_perm_0, x = k_41_cast);
            tensor<fp16, [2, 256, 1, 64]> var_3448_cast = slice_by_index(begin = var_3448_begin_0, end = var_3448_end_0, end_mask = var_3448_end_mask_0, x = transpose_21);
            tensor<int32, [4]> var_3452_begin_0 = const()[name = tensor<string, []>("op_3452_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_3452_end_0 = const()[name = tensor<string, []>("op_3452_end_0"), val = tensor<int32, [4]>([2, 256, 1, 128])];
            tensor<bool, [4]> var_3452_end_mask_0 = const()[name = tensor<string, []>("op_3452_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_3452_cast = slice_by_index(begin = var_3452_begin_0, end = var_3452_end_0, end_mask = var_3452_end_mask_0, x = transpose_21);
            tensor<int32, [4]> var_3456_begin_0 = const()[name = tensor<string, []>("op_3456_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_3456_end_0 = const()[name = tensor<string, []>("op_3456_end_0"), val = tensor<int32, [4]>([2, 256, 1, 192])];
            tensor<bool, [4]> var_3456_end_mask_0 = const()[name = tensor<string, []>("op_3456_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_3456_cast = slice_by_index(begin = var_3456_begin_0, end = var_3456_end_0, end_mask = var_3456_end_mask_0, x = transpose_21);
            tensor<int32, [4]> var_3460_begin_0 = const()[name = tensor<string, []>("op_3460_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_3460_end_0 = const()[name = tensor<string, []>("op_3460_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_3460_end_mask_0 = const()[name = tensor<string, []>("op_3460_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_3460_cast = slice_by_index(begin = var_3460_begin_0, end = var_3460_end_0, end_mask = var_3460_end_mask_0, x = transpose_21);
            tensor<int32, [4]> var_3464_begin_0 = const()[name = tensor<string, []>("op_3464_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_3464_end_0 = const()[name = tensor<string, []>("op_3464_end_0"), val = tensor<int32, [4]>([2, 256, 1, 320])];
            tensor<bool, [4]> var_3464_end_mask_0 = const()[name = tensor<string, []>("op_3464_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_3464_cast = slice_by_index(begin = var_3464_begin_0, end = var_3464_end_0, end_mask = var_3464_end_mask_0, x = transpose_21);
            tensor<int32, [4]> var_3468_begin_0 = const()[name = tensor<string, []>("op_3468_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_3468_end_0 = const()[name = tensor<string, []>("op_3468_end_0"), val = tensor<int32, [4]>([2, 256, 1, 384])];
            tensor<bool, [4]> var_3468_end_mask_0 = const()[name = tensor<string, []>("op_3468_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_3468_cast = slice_by_index(begin = var_3468_begin_0, end = var_3468_end_0, end_mask = var_3468_end_mask_0, x = transpose_21);
            tensor<int32, [4]> var_3472_begin_0 = const()[name = tensor<string, []>("op_3472_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_3472_end_0 = const()[name = tensor<string, []>("op_3472_end_0"), val = tensor<int32, [4]>([2, 256, 1, 448])];
            tensor<bool, [4]> var_3472_end_mask_0 = const()[name = tensor<string, []>("op_3472_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_3472_cast = slice_by_index(begin = var_3472_begin_0, end = var_3472_end_0, end_mask = var_3472_end_mask_0, x = transpose_21);
            tensor<int32, [4]> var_3476_begin_0 = const()[name = tensor<string, []>("op_3476_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_3476_end_0 = const()[name = tensor<string, []>("op_3476_end_0"), val = tensor<int32, [4]>([2, 256, 1, 512])];
            tensor<bool, [4]> var_3476_end_mask_0 = const()[name = tensor<string, []>("op_3476_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_3476_cast = slice_by_index(begin = var_3476_begin_0, end = var_3476_end_0, end_mask = var_3476_end_mask_0, x = transpose_21);
            tensor<int32, [4]> var_3480_begin_0 = const()[name = tensor<string, []>("op_3480_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3480_end_0 = const()[name = tensor<string, []>("op_3480_end_0"), val = tensor<int32, [4]>([2, 256, 1, 576])];
            tensor<bool, [4]> var_3480_end_mask_0 = const()[name = tensor<string, []>("op_3480_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_3480_cast = slice_by_index(begin = var_3480_begin_0, end = var_3480_end_0, end_mask = var_3480_end_mask_0, x = transpose_21);
            tensor<int32, [4]> var_3484_begin_0 = const()[name = tensor<string, []>("op_3484_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_3484_end_0 = const()[name = tensor<string, []>("op_3484_end_0"), val = tensor<int32, [4]>([2, 256, 1, 640])];
            tensor<bool, [4]> var_3484_end_mask_0 = const()[name = tensor<string, []>("op_3484_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_3484_cast = slice_by_index(begin = var_3484_begin_0, end = var_3484_end_0, end_mask = var_3484_end_mask_0, x = transpose_21);
            tensor<int32, [4]> var_3488_begin_0 = const()[name = tensor<string, []>("op_3488_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_3488_end_0 = const()[name = tensor<string, []>("op_3488_end_0"), val = tensor<int32, [4]>([2, 256, 1, 704])];
            tensor<bool, [4]> var_3488_end_mask_0 = const()[name = tensor<string, []>("op_3488_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_3488_cast = slice_by_index(begin = var_3488_begin_0, end = var_3488_end_0, end_mask = var_3488_end_mask_0, x = transpose_21);
            tensor<int32, [4]> var_3492_begin_0 = const()[name = tensor<string, []>("op_3492_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_3492_end_0 = const()[name = tensor<string, []>("op_3492_end_0"), val = tensor<int32, [4]>([2, 256, 1, 768])];
            tensor<bool, [4]> var_3492_end_mask_0 = const()[name = tensor<string, []>("op_3492_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_3492_cast = slice_by_index(begin = var_3492_begin_0, end = var_3492_end_0, end_mask = var_3492_end_mask_0, x = transpose_21);
            tensor<int32, [4]> var_3496_begin_0 = const()[name = tensor<string, []>("op_3496_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_3496_end_0 = const()[name = tensor<string, []>("op_3496_end_0"), val = tensor<int32, [4]>([2, 256, 1, 832])];
            tensor<bool, [4]> var_3496_end_mask_0 = const()[name = tensor<string, []>("op_3496_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_3496_cast = slice_by_index(begin = var_3496_begin_0, end = var_3496_end_0, end_mask = var_3496_end_mask_0, x = transpose_21);
            tensor<int32, [4]> var_3500_begin_0 = const()[name = tensor<string, []>("op_3500_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_3500_end_0 = const()[name = tensor<string, []>("op_3500_end_0"), val = tensor<int32, [4]>([2, 256, 1, 896])];
            tensor<bool, [4]> var_3500_end_mask_0 = const()[name = tensor<string, []>("op_3500_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_3500_cast = slice_by_index(begin = var_3500_begin_0, end = var_3500_end_0, end_mask = var_3500_end_mask_0, x = transpose_21);
            tensor<int32, [4]> var_3504_begin_0 = const()[name = tensor<string, []>("op_3504_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_3504_end_0 = const()[name = tensor<string, []>("op_3504_end_0"), val = tensor<int32, [4]>([2, 256, 1, 960])];
            tensor<bool, [4]> var_3504_end_mask_0 = const()[name = tensor<string, []>("op_3504_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_3504_cast = slice_by_index(begin = var_3504_begin_0, end = var_3504_end_0, end_mask = var_3504_end_mask_0, x = transpose_21);
            tensor<int32, [4]> var_3508_begin_0 = const()[name = tensor<string, []>("op_3508_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_3508_end_0 = const()[name = tensor<string, []>("op_3508_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_3508_end_mask_0 = const()[name = tensor<string, []>("op_3508_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_3508_cast = slice_by_index(begin = var_3508_begin_0, end = var_3508_end_0, end_mask = var_3508_end_mask_0, x = transpose_21);
            tensor<int32, [4]> var_3512_begin_0 = const()[name = tensor<string, []>("op_3512_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_3512_end_0 = const()[name = tensor<string, []>("op_3512_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1088])];
            tensor<bool, [4]> var_3512_end_mask_0 = const()[name = tensor<string, []>("op_3512_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_3512_cast = slice_by_index(begin = var_3512_begin_0, end = var_3512_end_0, end_mask = var_3512_end_mask_0, x = transpose_21);
            tensor<int32, [4]> var_3516_begin_0 = const()[name = tensor<string, []>("op_3516_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_3516_end_0 = const()[name = tensor<string, []>("op_3516_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1152])];
            tensor<bool, [4]> var_3516_end_mask_0 = const()[name = tensor<string, []>("op_3516_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_3516_cast = slice_by_index(begin = var_3516_begin_0, end = var_3516_end_0, end_mask = var_3516_end_mask_0, x = transpose_21);
            tensor<int32, [4]> var_3520_begin_0 = const()[name = tensor<string, []>("op_3520_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_3520_end_0 = const()[name = tensor<string, []>("op_3520_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1216])];
            tensor<bool, [4]> var_3520_end_mask_0 = const()[name = tensor<string, []>("op_3520_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_3520_cast = slice_by_index(begin = var_3520_begin_0, end = var_3520_end_0, end_mask = var_3520_end_mask_0, x = transpose_21);
            tensor<int32, [4]> var_3524_begin_0 = const()[name = tensor<string, []>("op_3524_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_3524_end_0 = const()[name = tensor<string, []>("op_3524_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1280])];
            tensor<bool, [4]> var_3524_end_mask_0 = const()[name = tensor<string, []>("op_3524_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_3524_cast = slice_by_index(begin = var_3524_begin_0, end = var_3524_end_0, end_mask = var_3524_end_mask_0, x = transpose_21);
            tensor<int32, [4]> var_3526_begin_0 = const()[name = tensor<string, []>("op_3526_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3526_end_0 = const()[name = tensor<string, []>("op_3526_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_3526_end_mask_0 = const()[name = tensor<string, []>("op_3526_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3526_cast = slice_by_index(begin = var_3526_begin_0, end = var_3526_end_0, end_mask = var_3526_end_mask_0, x = v_21_cast);
            tensor<int32, [4]> var_3530_begin_0 = const()[name = tensor<string, []>("op_3530_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_3530_end_0 = const()[name = tensor<string, []>("op_3530_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_3530_end_mask_0 = const()[name = tensor<string, []>("op_3530_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3530_cast = slice_by_index(begin = var_3530_begin_0, end = var_3530_end_0, end_mask = var_3530_end_mask_0, x = v_21_cast);
            tensor<int32, [4]> var_3534_begin_0 = const()[name = tensor<string, []>("op_3534_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_3534_end_0 = const()[name = tensor<string, []>("op_3534_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_3534_end_mask_0 = const()[name = tensor<string, []>("op_3534_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3534_cast = slice_by_index(begin = var_3534_begin_0, end = var_3534_end_0, end_mask = var_3534_end_mask_0, x = v_21_cast);
            tensor<int32, [4]> var_3538_begin_0 = const()[name = tensor<string, []>("op_3538_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_3538_end_0 = const()[name = tensor<string, []>("op_3538_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_3538_end_mask_0 = const()[name = tensor<string, []>("op_3538_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3538_cast = slice_by_index(begin = var_3538_begin_0, end = var_3538_end_0, end_mask = var_3538_end_mask_0, x = v_21_cast);
            tensor<int32, [4]> var_3542_begin_0 = const()[name = tensor<string, []>("op_3542_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_3542_end_0 = const()[name = tensor<string, []>("op_3542_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_3542_end_mask_0 = const()[name = tensor<string, []>("op_3542_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3542_cast = slice_by_index(begin = var_3542_begin_0, end = var_3542_end_0, end_mask = var_3542_end_mask_0, x = v_21_cast);
            tensor<int32, [4]> var_3546_begin_0 = const()[name = tensor<string, []>("op_3546_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_3546_end_0 = const()[name = tensor<string, []>("op_3546_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_3546_end_mask_0 = const()[name = tensor<string, []>("op_3546_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3546_cast = slice_by_index(begin = var_3546_begin_0, end = var_3546_end_0, end_mask = var_3546_end_mask_0, x = v_21_cast);
            tensor<int32, [4]> var_3550_begin_0 = const()[name = tensor<string, []>("op_3550_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_3550_end_0 = const()[name = tensor<string, []>("op_3550_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_3550_end_mask_0 = const()[name = tensor<string, []>("op_3550_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3550_cast = slice_by_index(begin = var_3550_begin_0, end = var_3550_end_0, end_mask = var_3550_end_mask_0, x = v_21_cast);
            tensor<int32, [4]> var_3554_begin_0 = const()[name = tensor<string, []>("op_3554_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_3554_end_0 = const()[name = tensor<string, []>("op_3554_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_3554_end_mask_0 = const()[name = tensor<string, []>("op_3554_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3554_cast = slice_by_index(begin = var_3554_begin_0, end = var_3554_end_0, end_mask = var_3554_end_mask_0, x = v_21_cast);
            tensor<int32, [4]> var_3558_begin_0 = const()[name = tensor<string, []>("op_3558_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_3558_end_0 = const()[name = tensor<string, []>("op_3558_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_3558_end_mask_0 = const()[name = tensor<string, []>("op_3558_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3558_cast = slice_by_index(begin = var_3558_begin_0, end = var_3558_end_0, end_mask = var_3558_end_mask_0, x = v_21_cast);
            tensor<int32, [4]> var_3562_begin_0 = const()[name = tensor<string, []>("op_3562_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_3562_end_0 = const()[name = tensor<string, []>("op_3562_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_3562_end_mask_0 = const()[name = tensor<string, []>("op_3562_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3562_cast = slice_by_index(begin = var_3562_begin_0, end = var_3562_end_0, end_mask = var_3562_end_mask_0, x = v_21_cast);
            tensor<int32, [4]> var_3566_begin_0 = const()[name = tensor<string, []>("op_3566_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_3566_end_0 = const()[name = tensor<string, []>("op_3566_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_3566_end_mask_0 = const()[name = tensor<string, []>("op_3566_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3566_cast = slice_by_index(begin = var_3566_begin_0, end = var_3566_end_0, end_mask = var_3566_end_mask_0, x = v_21_cast);
            tensor<int32, [4]> var_3570_begin_0 = const()[name = tensor<string, []>("op_3570_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_3570_end_0 = const()[name = tensor<string, []>("op_3570_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_3570_end_mask_0 = const()[name = tensor<string, []>("op_3570_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3570_cast = slice_by_index(begin = var_3570_begin_0, end = var_3570_end_0, end_mask = var_3570_end_mask_0, x = v_21_cast);
            tensor<int32, [4]> var_3574_begin_0 = const()[name = tensor<string, []>("op_3574_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_3574_end_0 = const()[name = tensor<string, []>("op_3574_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_3574_end_mask_0 = const()[name = tensor<string, []>("op_3574_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3574_cast = slice_by_index(begin = var_3574_begin_0, end = var_3574_end_0, end_mask = var_3574_end_mask_0, x = v_21_cast);
            tensor<int32, [4]> var_3578_begin_0 = const()[name = tensor<string, []>("op_3578_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_3578_end_0 = const()[name = tensor<string, []>("op_3578_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_3578_end_mask_0 = const()[name = tensor<string, []>("op_3578_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3578_cast = slice_by_index(begin = var_3578_begin_0, end = var_3578_end_0, end_mask = var_3578_end_mask_0, x = v_21_cast);
            tensor<int32, [4]> var_3582_begin_0 = const()[name = tensor<string, []>("op_3582_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_3582_end_0 = const()[name = tensor<string, []>("op_3582_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_3582_end_mask_0 = const()[name = tensor<string, []>("op_3582_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3582_cast = slice_by_index(begin = var_3582_begin_0, end = var_3582_end_0, end_mask = var_3582_end_mask_0, x = v_21_cast);
            tensor<int32, [4]> var_3586_begin_0 = const()[name = tensor<string, []>("op_3586_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_3586_end_0 = const()[name = tensor<string, []>("op_3586_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_3586_end_mask_0 = const()[name = tensor<string, []>("op_3586_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3586_cast = slice_by_index(begin = var_3586_begin_0, end = var_3586_end_0, end_mask = var_3586_end_mask_0, x = v_21_cast);
            tensor<int32, [4]> var_3590_begin_0 = const()[name = tensor<string, []>("op_3590_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_3590_end_0 = const()[name = tensor<string, []>("op_3590_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_3590_end_mask_0 = const()[name = tensor<string, []>("op_3590_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3590_cast = slice_by_index(begin = var_3590_begin_0, end = var_3590_end_0, end_mask = var_3590_end_mask_0, x = v_21_cast);
            tensor<int32, [4]> var_3594_begin_0 = const()[name = tensor<string, []>("op_3594_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_3594_end_0 = const()[name = tensor<string, []>("op_3594_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_3594_end_mask_0 = const()[name = tensor<string, []>("op_3594_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3594_cast = slice_by_index(begin = var_3594_begin_0, end = var_3594_end_0, end_mask = var_3594_end_mask_0, x = v_21_cast);
            tensor<int32, [4]> var_3598_begin_0 = const()[name = tensor<string, []>("op_3598_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_3598_end_0 = const()[name = tensor<string, []>("op_3598_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_3598_end_mask_0 = const()[name = tensor<string, []>("op_3598_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3598_cast = slice_by_index(begin = var_3598_begin_0, end = var_3598_end_0, end_mask = var_3598_end_mask_0, x = v_21_cast);
            tensor<int32, [4]> var_3602_begin_0 = const()[name = tensor<string, []>("op_3602_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_3602_end_0 = const()[name = tensor<string, []>("op_3602_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_3602_end_mask_0 = const()[name = tensor<string, []>("op_3602_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3602_cast = slice_by_index(begin = var_3602_begin_0, end = var_3602_end_0, end_mask = var_3602_end_mask_0, x = v_21_cast);
            tensor<string, []> var_3606_equation_0 = const()[name = tensor<string, []>("op_3606_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_3606_cast = einsum(equation = var_3606_equation_0, values = (var_3448_cast, var_3365_cast));
            tensor<fp16, []> var_3607_to_fp16 = const()[name = tensor<string, []>("op_3607_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_201_cast = mul(x = var_3606_cast, y = var_3607_to_fp16);
            tensor<string, []> var_3610_equation_0 = const()[name = tensor<string, []>("op_3610_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_3610_cast = einsum(equation = var_3610_equation_0, values = (var_3452_cast, var_3369_cast));
            tensor<fp16, []> var_3611_to_fp16 = const()[name = tensor<string, []>("op_3611_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_203_cast = mul(x = var_3610_cast, y = var_3611_to_fp16);
            tensor<string, []> var_3614_equation_0 = const()[name = tensor<string, []>("op_3614_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_3614_cast = einsum(equation = var_3614_equation_0, values = (var_3456_cast, var_3373_cast));
            tensor<fp16, []> var_3615_to_fp16 = const()[name = tensor<string, []>("op_3615_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_205_cast = mul(x = var_3614_cast, y = var_3615_to_fp16);
            tensor<string, []> var_3618_equation_0 = const()[name = tensor<string, []>("op_3618_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_3618_cast = einsum(equation = var_3618_equation_0, values = (var_3460_cast, var_3377_cast));
            tensor<fp16, []> var_3619_to_fp16 = const()[name = tensor<string, []>("op_3619_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_207_cast = mul(x = var_3618_cast, y = var_3619_to_fp16);
            tensor<string, []> var_3622_equation_0 = const()[name = tensor<string, []>("op_3622_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_3622_cast = einsum(equation = var_3622_equation_0, values = (var_3464_cast, var_3381_cast));
            tensor<fp16, []> var_3623_to_fp16 = const()[name = tensor<string, []>("op_3623_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_209_cast = mul(x = var_3622_cast, y = var_3623_to_fp16);
            tensor<string, []> var_3626_equation_0 = const()[name = tensor<string, []>("op_3626_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_3626_cast = einsum(equation = var_3626_equation_0, values = (var_3468_cast, var_3385_cast));
            tensor<fp16, []> var_3627_to_fp16 = const()[name = tensor<string, []>("op_3627_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_211_cast = mul(x = var_3626_cast, y = var_3627_to_fp16);
            tensor<string, []> var_3630_equation_0 = const()[name = tensor<string, []>("op_3630_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_3630_cast = einsum(equation = var_3630_equation_0, values = (var_3472_cast, var_3389_cast));
            tensor<fp16, []> var_3631_to_fp16 = const()[name = tensor<string, []>("op_3631_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_213_cast = mul(x = var_3630_cast, y = var_3631_to_fp16);
            tensor<string, []> var_3634_equation_0 = const()[name = tensor<string, []>("op_3634_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_3634_cast = einsum(equation = var_3634_equation_0, values = (var_3476_cast, var_3393_cast));
            tensor<fp16, []> var_3635_to_fp16 = const()[name = tensor<string, []>("op_3635_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_215_cast = mul(x = var_3634_cast, y = var_3635_to_fp16);
            tensor<string, []> var_3638_equation_0 = const()[name = tensor<string, []>("op_3638_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_3638_cast = einsum(equation = var_3638_equation_0, values = (var_3480_cast, var_3397_cast));
            tensor<fp16, []> var_3639_to_fp16 = const()[name = tensor<string, []>("op_3639_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_217_cast = mul(x = var_3638_cast, y = var_3639_to_fp16);
            tensor<string, []> var_3642_equation_0 = const()[name = tensor<string, []>("op_3642_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_3642_cast = einsum(equation = var_3642_equation_0, values = (var_3484_cast, var_3401_cast));
            tensor<fp16, []> var_3643_to_fp16 = const()[name = tensor<string, []>("op_3643_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_219_cast = mul(x = var_3642_cast, y = var_3643_to_fp16);
            tensor<string, []> var_3646_equation_0 = const()[name = tensor<string, []>("op_3646_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_3646_cast = einsum(equation = var_3646_equation_0, values = (var_3488_cast, var_3405_cast));
            tensor<fp16, []> var_3647_to_fp16 = const()[name = tensor<string, []>("op_3647_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_221_cast = mul(x = var_3646_cast, y = var_3647_to_fp16);
            tensor<string, []> var_3650_equation_0 = const()[name = tensor<string, []>("op_3650_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_3650_cast = einsum(equation = var_3650_equation_0, values = (var_3492_cast, var_3409_cast));
            tensor<fp16, []> var_3651_to_fp16 = const()[name = tensor<string, []>("op_3651_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_223_cast = mul(x = var_3650_cast, y = var_3651_to_fp16);
            tensor<string, []> var_3654_equation_0 = const()[name = tensor<string, []>("op_3654_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_3654_cast = einsum(equation = var_3654_equation_0, values = (var_3496_cast, var_3413_cast));
            tensor<fp16, []> var_3655_to_fp16 = const()[name = tensor<string, []>("op_3655_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_225_cast = mul(x = var_3654_cast, y = var_3655_to_fp16);
            tensor<string, []> var_3658_equation_0 = const()[name = tensor<string, []>("op_3658_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_3658_cast = einsum(equation = var_3658_equation_0, values = (var_3500_cast, var_3417_cast));
            tensor<fp16, []> var_3659_to_fp16 = const()[name = tensor<string, []>("op_3659_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_227_cast = mul(x = var_3658_cast, y = var_3659_to_fp16);
            tensor<string, []> var_3662_equation_0 = const()[name = tensor<string, []>("op_3662_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_3662_cast = einsum(equation = var_3662_equation_0, values = (var_3504_cast, var_3421_cast));
            tensor<fp16, []> var_3663_to_fp16 = const()[name = tensor<string, []>("op_3663_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_229_cast = mul(x = var_3662_cast, y = var_3663_to_fp16);
            tensor<string, []> var_3666_equation_0 = const()[name = tensor<string, []>("op_3666_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_3666_cast = einsum(equation = var_3666_equation_0, values = (var_3508_cast, var_3425_cast));
            tensor<fp16, []> var_3667_to_fp16 = const()[name = tensor<string, []>("op_3667_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_231_cast = mul(x = var_3666_cast, y = var_3667_to_fp16);
            tensor<string, []> var_3670_equation_0 = const()[name = tensor<string, []>("op_3670_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_3670_cast = einsum(equation = var_3670_equation_0, values = (var_3512_cast, var_3429_cast));
            tensor<fp16, []> var_3671_to_fp16 = const()[name = tensor<string, []>("op_3671_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_233_cast = mul(x = var_3670_cast, y = var_3671_to_fp16);
            tensor<string, []> var_3674_equation_0 = const()[name = tensor<string, []>("op_3674_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_3674_cast = einsum(equation = var_3674_equation_0, values = (var_3516_cast, var_3433_cast));
            tensor<fp16, []> var_3675_to_fp16 = const()[name = tensor<string, []>("op_3675_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_235_cast = mul(x = var_3674_cast, y = var_3675_to_fp16);
            tensor<string, []> var_3678_equation_0 = const()[name = tensor<string, []>("op_3678_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_3678_cast = einsum(equation = var_3678_equation_0, values = (var_3520_cast, var_3437_cast));
            tensor<fp16, []> var_3679_to_fp16 = const()[name = tensor<string, []>("op_3679_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_237_cast = mul(x = var_3678_cast, y = var_3679_to_fp16);
            tensor<string, []> var_3682_equation_0 = const()[name = tensor<string, []>("op_3682_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_3682_cast = einsum(equation = var_3682_equation_0, values = (var_3524_cast, var_3441_cast));
            tensor<fp16, []> var_3683_to_fp16 = const()[name = tensor<string, []>("op_3683_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_239_cast = mul(x = var_3682_cast, y = var_3683_to_fp16);
            tensor<fp16, [2, 256, 1, 256]> var_3685_cast = softmax(axis = var_2254, x = aw_201_cast);
            tensor<fp16, [2, 256, 1, 256]> var_3686_cast = softmax(axis = var_2254, x = aw_203_cast);
            tensor<fp16, [2, 256, 1, 256]> var_3687_cast = softmax(axis = var_2254, x = aw_205_cast);
            tensor<fp16, [2, 256, 1, 256]> var_3688_cast = softmax(axis = var_2254, x = aw_207_cast);
            tensor<fp16, [2, 256, 1, 256]> var_3689_cast = softmax(axis = var_2254, x = aw_209_cast);
            tensor<fp16, [2, 256, 1, 256]> var_3690_cast = softmax(axis = var_2254, x = aw_211_cast);
            tensor<fp16, [2, 256, 1, 256]> var_3691_cast = softmax(axis = var_2254, x = aw_213_cast);
            tensor<fp16, [2, 256, 1, 256]> var_3692_cast = softmax(axis = var_2254, x = aw_215_cast);
            tensor<fp16, [2, 256, 1, 256]> var_3693_cast = softmax(axis = var_2254, x = aw_217_cast);
            tensor<fp16, [2, 256, 1, 256]> var_3694_cast = softmax(axis = var_2254, x = aw_219_cast);
            tensor<fp16, [2, 256, 1, 256]> var_3695_cast = softmax(axis = var_2254, x = aw_221_cast);
            tensor<fp16, [2, 256, 1, 256]> var_3696_cast = softmax(axis = var_2254, x = aw_223_cast);
            tensor<fp16, [2, 256, 1, 256]> var_3697_cast = softmax(axis = var_2254, x = aw_225_cast);
            tensor<fp16, [2, 256, 1, 256]> var_3698_cast = softmax(axis = var_2254, x = aw_227_cast);
            tensor<fp16, [2, 256, 1, 256]> var_3699_cast = softmax(axis = var_2254, x = aw_229_cast);
            tensor<fp16, [2, 256, 1, 256]> var_3700_cast = softmax(axis = var_2254, x = aw_231_cast);
            tensor<fp16, [2, 256, 1, 256]> var_3701_cast = softmax(axis = var_2254, x = aw_233_cast);
            tensor<fp16, [2, 256, 1, 256]> var_3702_cast = softmax(axis = var_2254, x = aw_235_cast);
            tensor<fp16, [2, 256, 1, 256]> var_3703_cast = softmax(axis = var_2254, x = aw_237_cast);
            tensor<fp16, [2, 256, 1, 256]> var_3704_cast = softmax(axis = var_2254, x = aw_239_cast);
            tensor<string, []> var_3706_equation_0 = const()[name = tensor<string, []>("op_3706_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3706_cast = einsum(equation = var_3706_equation_0, values = (var_3526_cast, var_3685_cast));
            tensor<string, []> var_3708_equation_0 = const()[name = tensor<string, []>("op_3708_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3708_cast = einsum(equation = var_3708_equation_0, values = (var_3530_cast, var_3686_cast));
            tensor<string, []> var_3710_equation_0 = const()[name = tensor<string, []>("op_3710_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3710_cast = einsum(equation = var_3710_equation_0, values = (var_3534_cast, var_3687_cast));
            tensor<string, []> var_3712_equation_0 = const()[name = tensor<string, []>("op_3712_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3712_cast = einsum(equation = var_3712_equation_0, values = (var_3538_cast, var_3688_cast));
            tensor<string, []> var_3714_equation_0 = const()[name = tensor<string, []>("op_3714_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3714_cast = einsum(equation = var_3714_equation_0, values = (var_3542_cast, var_3689_cast));
            tensor<string, []> var_3716_equation_0 = const()[name = tensor<string, []>("op_3716_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3716_cast = einsum(equation = var_3716_equation_0, values = (var_3546_cast, var_3690_cast));
            tensor<string, []> var_3718_equation_0 = const()[name = tensor<string, []>("op_3718_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3718_cast = einsum(equation = var_3718_equation_0, values = (var_3550_cast, var_3691_cast));
            tensor<string, []> var_3720_equation_0 = const()[name = tensor<string, []>("op_3720_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3720_cast = einsum(equation = var_3720_equation_0, values = (var_3554_cast, var_3692_cast));
            tensor<string, []> var_3722_equation_0 = const()[name = tensor<string, []>("op_3722_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3722_cast = einsum(equation = var_3722_equation_0, values = (var_3558_cast, var_3693_cast));
            tensor<string, []> var_3724_equation_0 = const()[name = tensor<string, []>("op_3724_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3724_cast = einsum(equation = var_3724_equation_0, values = (var_3562_cast, var_3694_cast));
            tensor<string, []> var_3726_equation_0 = const()[name = tensor<string, []>("op_3726_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3726_cast = einsum(equation = var_3726_equation_0, values = (var_3566_cast, var_3695_cast));
            tensor<string, []> var_3728_equation_0 = const()[name = tensor<string, []>("op_3728_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3728_cast = einsum(equation = var_3728_equation_0, values = (var_3570_cast, var_3696_cast));
            tensor<string, []> var_3730_equation_0 = const()[name = tensor<string, []>("op_3730_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3730_cast = einsum(equation = var_3730_equation_0, values = (var_3574_cast, var_3697_cast));
            tensor<string, []> var_3732_equation_0 = const()[name = tensor<string, []>("op_3732_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3732_cast = einsum(equation = var_3732_equation_0, values = (var_3578_cast, var_3698_cast));
            tensor<string, []> var_3734_equation_0 = const()[name = tensor<string, []>("op_3734_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3734_cast = einsum(equation = var_3734_equation_0, values = (var_3582_cast, var_3699_cast));
            tensor<string, []> var_3736_equation_0 = const()[name = tensor<string, []>("op_3736_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3736_cast = einsum(equation = var_3736_equation_0, values = (var_3586_cast, var_3700_cast));
            tensor<string, []> var_3738_equation_0 = const()[name = tensor<string, []>("op_3738_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3738_cast = einsum(equation = var_3738_equation_0, values = (var_3590_cast, var_3701_cast));
            tensor<string, []> var_3740_equation_0 = const()[name = tensor<string, []>("op_3740_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3740_cast = einsum(equation = var_3740_equation_0, values = (var_3594_cast, var_3702_cast));
            tensor<string, []> var_3742_equation_0 = const()[name = tensor<string, []>("op_3742_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3742_cast = einsum(equation = var_3742_equation_0, values = (var_3598_cast, var_3703_cast));
            tensor<string, []> var_3744_equation_0 = const()[name = tensor<string, []>("op_3744_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_3744_cast = einsum(equation = var_3744_equation_0, values = (var_3602_cast, var_3704_cast));
            tensor<bool, []> input_159_interleave_0 = const()[name = tensor<string, []>("input_159_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 256]> input_159_cast = concat(axis = var_2254, interleave = input_159_interleave_0, values = (var_3706_cast, var_3708_cast, var_3710_cast, var_3712_cast, var_3714_cast, var_3716_cast, var_3718_cast, var_3720_cast, var_3722_cast, var_3724_cast, var_3726_cast, var_3728_cast, var_3730_cast, var_3732_cast, var_3734_cast, var_3736_cast, var_3738_cast, var_3740_cast, var_3742_cast, var_3744_cast));
            tensor<int32, [2]> var_3750 = const()[name = tensor<string, []>("op_3750"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3752 = const()[name = tensor<string, []>("op_3752"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_3754_pad_type_0 = const()[name = tensor<string, []>("op_3754_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_3754_pad_0 = const()[name = tensor<string, []>("op_3754_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(296242944)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(299519808)))];
            tensor<fp16, [2, 1280, 1, 256]> var_3754_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_3752, groups = var_2254, pad = var_3754_pad_0, pad_type = var_3754_pad_type_0, strides = var_3750, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_159_cast);
            tensor<fp16, [2, 1280, 1, 256]> inputs_33_cast = add(x = var_3754_cast, y = inputs_31_cast);
            tensor<int32, [1]> var_3758 = const()[name = tensor<string, []>("op_3758"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_33_cast = reduce_mean(axes = var_3758, keep_dims = var_2249, x = inputs_33_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_33_cast = sub(x = inputs_33_cast, y = channels_mean_33_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_sq_33_cast = mul(x = zero_mean_33_cast, y = zero_mean_33_cast);
            tensor<int32, [1]> var_3762 = const()[name = tensor<string, []>("op_3762"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_3763_cast = reduce_mean(axes = var_3762, keep_dims = var_2249, x = zero_mean_sq_33_cast);
            tensor<fp16, []> var_3764_to_fp16 = const()[name = tensor<string, []>("op_3764_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_3765_cast = add(x = var_3763_cast, y = var_3764_to_fp16);
            tensor<fp16, []> denom_33_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_33_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 256]> denom_33_cast = rsqrt(epsilon = denom_33_epsilon_0_to_fp16, x = var_3765_cast);
            tensor<fp16, [2, 1280, 1, 256]> out_33_cast = mul(x = zero_mean_33_cast, y = denom_33_cast);
            tensor<fp16, [1, 1280, 1, 1]> var_3769_to_fp16 = const()[name = tensor<string, []>("op_3769_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(299522432)))];
            tensor<fp16, [2, 1280, 1, 256]> var_3770_cast = add(x = out_33_cast, y = var_3769_to_fp16);
            tensor<fp16, [1, 1280, 1, 1]> var_3772_to_fp16 = const()[name = tensor<string, []>("op_3772_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(299525056)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_101_cast = mul(x = var_3770_cast, y = var_3772_to_fp16);
            tensor<int32, [2]> var_3779 = const()[name = tensor<string, []>("op_3779"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3781 = const()[name = tensor<string, []>("op_3781"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_23_pad_type_0 = const()[name = tensor<string, []>("q_23_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_23_pad_0 = const()[name = tensor<string, []>("q_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(299527680)))];
            tensor<fp16, [2, 1280, 1, 256]> q_23_cast = conv(dilations = var_3781, groups = var_2254, pad = q_23_pad_0, pad_type = q_23_pad_type_0, strides = var_3779, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_101_cast);
            tensor<int32, [2]> var_3785 = const()[name = tensor<string, []>("op_3785"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3787 = const()[name = tensor<string, []>("op_3787"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_45_pad_type_0 = const()[name = tensor<string, []>("k_45_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_45_pad_0 = const()[name = tensor<string, []>("k_45_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1024, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(302804544)))];
            tensor<fp16, [2, 1280, 1, 77]> k_45_cast = conv(dilations = var_3787, groups = var_2254, pad = k_45_pad_0, pad_type = k_45_pad_type_0, strides = var_3785, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [2]> var_3791 = const()[name = tensor<string, []>("op_3791"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3793 = const()[name = tensor<string, []>("op_3793"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_23_pad_type_0 = const()[name = tensor<string, []>("v_23_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_23_pad_0 = const()[name = tensor<string, []>("v_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1024, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(305426048)))];
            tensor<fp16, [2, 1280, 1, 77]> v_23_cast = conv(dilations = var_3793, groups = var_2254, pad = v_23_pad_0, pad_type = v_23_pad_type_0, strides = var_3791, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [4]> var_3797_begin_0 = const()[name = tensor<string, []>("op_3797_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3797_end_0 = const()[name = tensor<string, []>("op_3797_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_3797_end_mask_0 = const()[name = tensor<string, []>("op_3797_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3797_cast = slice_by_index(begin = var_3797_begin_0, end = var_3797_end_0, end_mask = var_3797_end_mask_0, x = q_23_cast);
            tensor<int32, [4]> var_3801_begin_0 = const()[name = tensor<string, []>("op_3801_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_3801_end_0 = const()[name = tensor<string, []>("op_3801_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_3801_end_mask_0 = const()[name = tensor<string, []>("op_3801_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3801_cast = slice_by_index(begin = var_3801_begin_0, end = var_3801_end_0, end_mask = var_3801_end_mask_0, x = q_23_cast);
            tensor<int32, [4]> var_3805_begin_0 = const()[name = tensor<string, []>("op_3805_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_3805_end_0 = const()[name = tensor<string, []>("op_3805_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_3805_end_mask_0 = const()[name = tensor<string, []>("op_3805_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3805_cast = slice_by_index(begin = var_3805_begin_0, end = var_3805_end_0, end_mask = var_3805_end_mask_0, x = q_23_cast);
            tensor<int32, [4]> var_3809_begin_0 = const()[name = tensor<string, []>("op_3809_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_3809_end_0 = const()[name = tensor<string, []>("op_3809_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_3809_end_mask_0 = const()[name = tensor<string, []>("op_3809_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3809_cast = slice_by_index(begin = var_3809_begin_0, end = var_3809_end_0, end_mask = var_3809_end_mask_0, x = q_23_cast);
            tensor<int32, [4]> var_3813_begin_0 = const()[name = tensor<string, []>("op_3813_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_3813_end_0 = const()[name = tensor<string, []>("op_3813_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_3813_end_mask_0 = const()[name = tensor<string, []>("op_3813_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3813_cast = slice_by_index(begin = var_3813_begin_0, end = var_3813_end_0, end_mask = var_3813_end_mask_0, x = q_23_cast);
            tensor<int32, [4]> var_3817_begin_0 = const()[name = tensor<string, []>("op_3817_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_3817_end_0 = const()[name = tensor<string, []>("op_3817_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_3817_end_mask_0 = const()[name = tensor<string, []>("op_3817_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3817_cast = slice_by_index(begin = var_3817_begin_0, end = var_3817_end_0, end_mask = var_3817_end_mask_0, x = q_23_cast);
            tensor<int32, [4]> var_3821_begin_0 = const()[name = tensor<string, []>("op_3821_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_3821_end_0 = const()[name = tensor<string, []>("op_3821_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_3821_end_mask_0 = const()[name = tensor<string, []>("op_3821_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3821_cast = slice_by_index(begin = var_3821_begin_0, end = var_3821_end_0, end_mask = var_3821_end_mask_0, x = q_23_cast);
            tensor<int32, [4]> var_3825_begin_0 = const()[name = tensor<string, []>("op_3825_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_3825_end_0 = const()[name = tensor<string, []>("op_3825_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_3825_end_mask_0 = const()[name = tensor<string, []>("op_3825_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3825_cast = slice_by_index(begin = var_3825_begin_0, end = var_3825_end_0, end_mask = var_3825_end_mask_0, x = q_23_cast);
            tensor<int32, [4]> var_3829_begin_0 = const()[name = tensor<string, []>("op_3829_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_3829_end_0 = const()[name = tensor<string, []>("op_3829_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_3829_end_mask_0 = const()[name = tensor<string, []>("op_3829_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3829_cast = slice_by_index(begin = var_3829_begin_0, end = var_3829_end_0, end_mask = var_3829_end_mask_0, x = q_23_cast);
            tensor<int32, [4]> var_3833_begin_0 = const()[name = tensor<string, []>("op_3833_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_3833_end_0 = const()[name = tensor<string, []>("op_3833_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_3833_end_mask_0 = const()[name = tensor<string, []>("op_3833_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3833_cast = slice_by_index(begin = var_3833_begin_0, end = var_3833_end_0, end_mask = var_3833_end_mask_0, x = q_23_cast);
            tensor<int32, [4]> var_3837_begin_0 = const()[name = tensor<string, []>("op_3837_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_3837_end_0 = const()[name = tensor<string, []>("op_3837_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_3837_end_mask_0 = const()[name = tensor<string, []>("op_3837_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3837_cast = slice_by_index(begin = var_3837_begin_0, end = var_3837_end_0, end_mask = var_3837_end_mask_0, x = q_23_cast);
            tensor<int32, [4]> var_3841_begin_0 = const()[name = tensor<string, []>("op_3841_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_3841_end_0 = const()[name = tensor<string, []>("op_3841_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_3841_end_mask_0 = const()[name = tensor<string, []>("op_3841_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3841_cast = slice_by_index(begin = var_3841_begin_0, end = var_3841_end_0, end_mask = var_3841_end_mask_0, x = q_23_cast);
            tensor<int32, [4]> var_3845_begin_0 = const()[name = tensor<string, []>("op_3845_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_3845_end_0 = const()[name = tensor<string, []>("op_3845_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_3845_end_mask_0 = const()[name = tensor<string, []>("op_3845_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3845_cast = slice_by_index(begin = var_3845_begin_0, end = var_3845_end_0, end_mask = var_3845_end_mask_0, x = q_23_cast);
            tensor<int32, [4]> var_3849_begin_0 = const()[name = tensor<string, []>("op_3849_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_3849_end_0 = const()[name = tensor<string, []>("op_3849_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_3849_end_mask_0 = const()[name = tensor<string, []>("op_3849_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3849_cast = slice_by_index(begin = var_3849_begin_0, end = var_3849_end_0, end_mask = var_3849_end_mask_0, x = q_23_cast);
            tensor<int32, [4]> var_3853_begin_0 = const()[name = tensor<string, []>("op_3853_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_3853_end_0 = const()[name = tensor<string, []>("op_3853_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_3853_end_mask_0 = const()[name = tensor<string, []>("op_3853_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3853_cast = slice_by_index(begin = var_3853_begin_0, end = var_3853_end_0, end_mask = var_3853_end_mask_0, x = q_23_cast);
            tensor<int32, [4]> var_3857_begin_0 = const()[name = tensor<string, []>("op_3857_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_3857_end_0 = const()[name = tensor<string, []>("op_3857_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_3857_end_mask_0 = const()[name = tensor<string, []>("op_3857_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3857_cast = slice_by_index(begin = var_3857_begin_0, end = var_3857_end_0, end_mask = var_3857_end_mask_0, x = q_23_cast);
            tensor<int32, [4]> var_3861_begin_0 = const()[name = tensor<string, []>("op_3861_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_3861_end_0 = const()[name = tensor<string, []>("op_3861_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_3861_end_mask_0 = const()[name = tensor<string, []>("op_3861_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3861_cast = slice_by_index(begin = var_3861_begin_0, end = var_3861_end_0, end_mask = var_3861_end_mask_0, x = q_23_cast);
            tensor<int32, [4]> var_3865_begin_0 = const()[name = tensor<string, []>("op_3865_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_3865_end_0 = const()[name = tensor<string, []>("op_3865_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_3865_end_mask_0 = const()[name = tensor<string, []>("op_3865_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3865_cast = slice_by_index(begin = var_3865_begin_0, end = var_3865_end_0, end_mask = var_3865_end_mask_0, x = q_23_cast);
            tensor<int32, [4]> var_3869_begin_0 = const()[name = tensor<string, []>("op_3869_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_3869_end_0 = const()[name = tensor<string, []>("op_3869_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_3869_end_mask_0 = const()[name = tensor<string, []>("op_3869_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3869_cast = slice_by_index(begin = var_3869_begin_0, end = var_3869_end_0, end_mask = var_3869_end_mask_0, x = q_23_cast);
            tensor<int32, [4]> var_3873_begin_0 = const()[name = tensor<string, []>("op_3873_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_3873_end_0 = const()[name = tensor<string, []>("op_3873_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_3873_end_mask_0 = const()[name = tensor<string, []>("op_3873_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_3873_cast = slice_by_index(begin = var_3873_begin_0, end = var_3873_end_0, end_mask = var_3873_end_mask_0, x = q_23_cast);
            tensor<int32, [4]> k_47_perm_0 = const()[name = tensor<string, []>("k_47_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_3880_begin_0 = const()[name = tensor<string, []>("op_3880_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3880_end_0 = const()[name = tensor<string, []>("op_3880_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_3880_end_mask_0 = const()[name = tensor<string, []>("op_3880_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_20 = transpose(perm = k_47_perm_0, x = k_45_cast);
            tensor<fp16, [2, 77, 1, 64]> var_3880_cast = slice_by_index(begin = var_3880_begin_0, end = var_3880_end_0, end_mask = var_3880_end_mask_0, x = transpose_20);
            tensor<int32, [4]> var_3884_begin_0 = const()[name = tensor<string, []>("op_3884_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_3884_end_0 = const()[name = tensor<string, []>("op_3884_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_3884_end_mask_0 = const()[name = tensor<string, []>("op_3884_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3884_cast = slice_by_index(begin = var_3884_begin_0, end = var_3884_end_0, end_mask = var_3884_end_mask_0, x = transpose_20);
            tensor<int32, [4]> var_3888_begin_0 = const()[name = tensor<string, []>("op_3888_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_3888_end_0 = const()[name = tensor<string, []>("op_3888_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_3888_end_mask_0 = const()[name = tensor<string, []>("op_3888_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3888_cast = slice_by_index(begin = var_3888_begin_0, end = var_3888_end_0, end_mask = var_3888_end_mask_0, x = transpose_20);
            tensor<int32, [4]> var_3892_begin_0 = const()[name = tensor<string, []>("op_3892_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_3892_end_0 = const()[name = tensor<string, []>("op_3892_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_3892_end_mask_0 = const()[name = tensor<string, []>("op_3892_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3892_cast = slice_by_index(begin = var_3892_begin_0, end = var_3892_end_0, end_mask = var_3892_end_mask_0, x = transpose_20);
            tensor<int32, [4]> var_3896_begin_0 = const()[name = tensor<string, []>("op_3896_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_3896_end_0 = const()[name = tensor<string, []>("op_3896_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_3896_end_mask_0 = const()[name = tensor<string, []>("op_3896_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3896_cast = slice_by_index(begin = var_3896_begin_0, end = var_3896_end_0, end_mask = var_3896_end_mask_0, x = transpose_20);
            tensor<int32, [4]> var_3900_begin_0 = const()[name = tensor<string, []>("op_3900_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_3900_end_0 = const()[name = tensor<string, []>("op_3900_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_3900_end_mask_0 = const()[name = tensor<string, []>("op_3900_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3900_cast = slice_by_index(begin = var_3900_begin_0, end = var_3900_end_0, end_mask = var_3900_end_mask_0, x = transpose_20);
            tensor<int32, [4]> var_3904_begin_0 = const()[name = tensor<string, []>("op_3904_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_3904_end_0 = const()[name = tensor<string, []>("op_3904_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_3904_end_mask_0 = const()[name = tensor<string, []>("op_3904_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3904_cast = slice_by_index(begin = var_3904_begin_0, end = var_3904_end_0, end_mask = var_3904_end_mask_0, x = transpose_20);
            tensor<int32, [4]> var_3908_begin_0 = const()[name = tensor<string, []>("op_3908_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_3908_end_0 = const()[name = tensor<string, []>("op_3908_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_3908_end_mask_0 = const()[name = tensor<string, []>("op_3908_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3908_cast = slice_by_index(begin = var_3908_begin_0, end = var_3908_end_0, end_mask = var_3908_end_mask_0, x = transpose_20);
            tensor<int32, [4]> var_3912_begin_0 = const()[name = tensor<string, []>("op_3912_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3912_end_0 = const()[name = tensor<string, []>("op_3912_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_3912_end_mask_0 = const()[name = tensor<string, []>("op_3912_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3912_cast = slice_by_index(begin = var_3912_begin_0, end = var_3912_end_0, end_mask = var_3912_end_mask_0, x = transpose_20);
            tensor<int32, [4]> var_3916_begin_0 = const()[name = tensor<string, []>("op_3916_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_3916_end_0 = const()[name = tensor<string, []>("op_3916_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_3916_end_mask_0 = const()[name = tensor<string, []>("op_3916_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3916_cast = slice_by_index(begin = var_3916_begin_0, end = var_3916_end_0, end_mask = var_3916_end_mask_0, x = transpose_20);
            tensor<int32, [4]> var_3920_begin_0 = const()[name = tensor<string, []>("op_3920_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_3920_end_0 = const()[name = tensor<string, []>("op_3920_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_3920_end_mask_0 = const()[name = tensor<string, []>("op_3920_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3920_cast = slice_by_index(begin = var_3920_begin_0, end = var_3920_end_0, end_mask = var_3920_end_mask_0, x = transpose_20);
            tensor<int32, [4]> var_3924_begin_0 = const()[name = tensor<string, []>("op_3924_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_3924_end_0 = const()[name = tensor<string, []>("op_3924_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_3924_end_mask_0 = const()[name = tensor<string, []>("op_3924_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3924_cast = slice_by_index(begin = var_3924_begin_0, end = var_3924_end_0, end_mask = var_3924_end_mask_0, x = transpose_20);
            tensor<int32, [4]> var_3928_begin_0 = const()[name = tensor<string, []>("op_3928_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_3928_end_0 = const()[name = tensor<string, []>("op_3928_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_3928_end_mask_0 = const()[name = tensor<string, []>("op_3928_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3928_cast = slice_by_index(begin = var_3928_begin_0, end = var_3928_end_0, end_mask = var_3928_end_mask_0, x = transpose_20);
            tensor<int32, [4]> var_3932_begin_0 = const()[name = tensor<string, []>("op_3932_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_3932_end_0 = const()[name = tensor<string, []>("op_3932_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_3932_end_mask_0 = const()[name = tensor<string, []>("op_3932_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3932_cast = slice_by_index(begin = var_3932_begin_0, end = var_3932_end_0, end_mask = var_3932_end_mask_0, x = transpose_20);
            tensor<int32, [4]> var_3936_begin_0 = const()[name = tensor<string, []>("op_3936_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_3936_end_0 = const()[name = tensor<string, []>("op_3936_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_3936_end_mask_0 = const()[name = tensor<string, []>("op_3936_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3936_cast = slice_by_index(begin = var_3936_begin_0, end = var_3936_end_0, end_mask = var_3936_end_mask_0, x = transpose_20);
            tensor<int32, [4]> var_3940_begin_0 = const()[name = tensor<string, []>("op_3940_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_3940_end_0 = const()[name = tensor<string, []>("op_3940_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_3940_end_mask_0 = const()[name = tensor<string, []>("op_3940_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3940_cast = slice_by_index(begin = var_3940_begin_0, end = var_3940_end_0, end_mask = var_3940_end_mask_0, x = transpose_20);
            tensor<int32, [4]> var_3944_begin_0 = const()[name = tensor<string, []>("op_3944_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_3944_end_0 = const()[name = tensor<string, []>("op_3944_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_3944_end_mask_0 = const()[name = tensor<string, []>("op_3944_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3944_cast = slice_by_index(begin = var_3944_begin_0, end = var_3944_end_0, end_mask = var_3944_end_mask_0, x = transpose_20);
            tensor<int32, [4]> var_3948_begin_0 = const()[name = tensor<string, []>("op_3948_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_3948_end_0 = const()[name = tensor<string, []>("op_3948_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_3948_end_mask_0 = const()[name = tensor<string, []>("op_3948_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3948_cast = slice_by_index(begin = var_3948_begin_0, end = var_3948_end_0, end_mask = var_3948_end_mask_0, x = transpose_20);
            tensor<int32, [4]> var_3952_begin_0 = const()[name = tensor<string, []>("op_3952_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_3952_end_0 = const()[name = tensor<string, []>("op_3952_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_3952_end_mask_0 = const()[name = tensor<string, []>("op_3952_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3952_cast = slice_by_index(begin = var_3952_begin_0, end = var_3952_end_0, end_mask = var_3952_end_mask_0, x = transpose_20);
            tensor<int32, [4]> var_3956_begin_0 = const()[name = tensor<string, []>("op_3956_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_3956_end_0 = const()[name = tensor<string, []>("op_3956_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_3956_end_mask_0 = const()[name = tensor<string, []>("op_3956_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3956_cast = slice_by_index(begin = var_3956_begin_0, end = var_3956_end_0, end_mask = var_3956_end_mask_0, x = transpose_20);
            tensor<int32, [4]> var_3958_begin_0 = const()[name = tensor<string, []>("op_3958_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3958_end_0 = const()[name = tensor<string, []>("op_3958_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_3958_end_mask_0 = const()[name = tensor<string, []>("op_3958_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3958_cast = slice_by_index(begin = var_3958_begin_0, end = var_3958_end_0, end_mask = var_3958_end_mask_0, x = v_23_cast);
            tensor<int32, [4]> var_3962_begin_0 = const()[name = tensor<string, []>("op_3962_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_3962_end_0 = const()[name = tensor<string, []>("op_3962_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_3962_end_mask_0 = const()[name = tensor<string, []>("op_3962_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3962_cast = slice_by_index(begin = var_3962_begin_0, end = var_3962_end_0, end_mask = var_3962_end_mask_0, x = v_23_cast);
            tensor<int32, [4]> var_3966_begin_0 = const()[name = tensor<string, []>("op_3966_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_3966_end_0 = const()[name = tensor<string, []>("op_3966_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_3966_end_mask_0 = const()[name = tensor<string, []>("op_3966_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3966_cast = slice_by_index(begin = var_3966_begin_0, end = var_3966_end_0, end_mask = var_3966_end_mask_0, x = v_23_cast);
            tensor<int32, [4]> var_3970_begin_0 = const()[name = tensor<string, []>("op_3970_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_3970_end_0 = const()[name = tensor<string, []>("op_3970_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_3970_end_mask_0 = const()[name = tensor<string, []>("op_3970_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3970_cast = slice_by_index(begin = var_3970_begin_0, end = var_3970_end_0, end_mask = var_3970_end_mask_0, x = v_23_cast);
            tensor<int32, [4]> var_3974_begin_0 = const()[name = tensor<string, []>("op_3974_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_3974_end_0 = const()[name = tensor<string, []>("op_3974_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_3974_end_mask_0 = const()[name = tensor<string, []>("op_3974_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3974_cast = slice_by_index(begin = var_3974_begin_0, end = var_3974_end_0, end_mask = var_3974_end_mask_0, x = v_23_cast);
            tensor<int32, [4]> var_3978_begin_0 = const()[name = tensor<string, []>("op_3978_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_3978_end_0 = const()[name = tensor<string, []>("op_3978_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_3978_end_mask_0 = const()[name = tensor<string, []>("op_3978_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3978_cast = slice_by_index(begin = var_3978_begin_0, end = var_3978_end_0, end_mask = var_3978_end_mask_0, x = v_23_cast);
            tensor<int32, [4]> var_3982_begin_0 = const()[name = tensor<string, []>("op_3982_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_3982_end_0 = const()[name = tensor<string, []>("op_3982_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_3982_end_mask_0 = const()[name = tensor<string, []>("op_3982_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3982_cast = slice_by_index(begin = var_3982_begin_0, end = var_3982_end_0, end_mask = var_3982_end_mask_0, x = v_23_cast);
            tensor<int32, [4]> var_3986_begin_0 = const()[name = tensor<string, []>("op_3986_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_3986_end_0 = const()[name = tensor<string, []>("op_3986_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_3986_end_mask_0 = const()[name = tensor<string, []>("op_3986_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3986_cast = slice_by_index(begin = var_3986_begin_0, end = var_3986_end_0, end_mask = var_3986_end_mask_0, x = v_23_cast);
            tensor<int32, [4]> var_3990_begin_0 = const()[name = tensor<string, []>("op_3990_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_3990_end_0 = const()[name = tensor<string, []>("op_3990_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_3990_end_mask_0 = const()[name = tensor<string, []>("op_3990_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3990_cast = slice_by_index(begin = var_3990_begin_0, end = var_3990_end_0, end_mask = var_3990_end_mask_0, x = v_23_cast);
            tensor<int32, [4]> var_3994_begin_0 = const()[name = tensor<string, []>("op_3994_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_3994_end_0 = const()[name = tensor<string, []>("op_3994_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_3994_end_mask_0 = const()[name = tensor<string, []>("op_3994_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3994_cast = slice_by_index(begin = var_3994_begin_0, end = var_3994_end_0, end_mask = var_3994_end_mask_0, x = v_23_cast);
            tensor<int32, [4]> var_3998_begin_0 = const()[name = tensor<string, []>("op_3998_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_3998_end_0 = const()[name = tensor<string, []>("op_3998_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_3998_end_mask_0 = const()[name = tensor<string, []>("op_3998_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3998_cast = slice_by_index(begin = var_3998_begin_0, end = var_3998_end_0, end_mask = var_3998_end_mask_0, x = v_23_cast);
            tensor<int32, [4]> var_4002_begin_0 = const()[name = tensor<string, []>("op_4002_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_4002_end_0 = const()[name = tensor<string, []>("op_4002_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_4002_end_mask_0 = const()[name = tensor<string, []>("op_4002_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4002_cast = slice_by_index(begin = var_4002_begin_0, end = var_4002_end_0, end_mask = var_4002_end_mask_0, x = v_23_cast);
            tensor<int32, [4]> var_4006_begin_0 = const()[name = tensor<string, []>("op_4006_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_4006_end_0 = const()[name = tensor<string, []>("op_4006_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_4006_end_mask_0 = const()[name = tensor<string, []>("op_4006_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4006_cast = slice_by_index(begin = var_4006_begin_0, end = var_4006_end_0, end_mask = var_4006_end_mask_0, x = v_23_cast);
            tensor<int32, [4]> var_4010_begin_0 = const()[name = tensor<string, []>("op_4010_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_4010_end_0 = const()[name = tensor<string, []>("op_4010_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_4010_end_mask_0 = const()[name = tensor<string, []>("op_4010_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4010_cast = slice_by_index(begin = var_4010_begin_0, end = var_4010_end_0, end_mask = var_4010_end_mask_0, x = v_23_cast);
            tensor<int32, [4]> var_4014_begin_0 = const()[name = tensor<string, []>("op_4014_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_4014_end_0 = const()[name = tensor<string, []>("op_4014_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_4014_end_mask_0 = const()[name = tensor<string, []>("op_4014_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4014_cast = slice_by_index(begin = var_4014_begin_0, end = var_4014_end_0, end_mask = var_4014_end_mask_0, x = v_23_cast);
            tensor<int32, [4]> var_4018_begin_0 = const()[name = tensor<string, []>("op_4018_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_4018_end_0 = const()[name = tensor<string, []>("op_4018_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_4018_end_mask_0 = const()[name = tensor<string, []>("op_4018_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4018_cast = slice_by_index(begin = var_4018_begin_0, end = var_4018_end_0, end_mask = var_4018_end_mask_0, x = v_23_cast);
            tensor<int32, [4]> var_4022_begin_0 = const()[name = tensor<string, []>("op_4022_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_4022_end_0 = const()[name = tensor<string, []>("op_4022_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_4022_end_mask_0 = const()[name = tensor<string, []>("op_4022_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4022_cast = slice_by_index(begin = var_4022_begin_0, end = var_4022_end_0, end_mask = var_4022_end_mask_0, x = v_23_cast);
            tensor<int32, [4]> var_4026_begin_0 = const()[name = tensor<string, []>("op_4026_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_4026_end_0 = const()[name = tensor<string, []>("op_4026_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_4026_end_mask_0 = const()[name = tensor<string, []>("op_4026_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4026_cast = slice_by_index(begin = var_4026_begin_0, end = var_4026_end_0, end_mask = var_4026_end_mask_0, x = v_23_cast);
            tensor<int32, [4]> var_4030_begin_0 = const()[name = tensor<string, []>("op_4030_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_4030_end_0 = const()[name = tensor<string, []>("op_4030_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_4030_end_mask_0 = const()[name = tensor<string, []>("op_4030_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4030_cast = slice_by_index(begin = var_4030_begin_0, end = var_4030_end_0, end_mask = var_4030_end_mask_0, x = v_23_cast);
            tensor<int32, [4]> var_4034_begin_0 = const()[name = tensor<string, []>("op_4034_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_4034_end_0 = const()[name = tensor<string, []>("op_4034_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_4034_end_mask_0 = const()[name = tensor<string, []>("op_4034_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4034_cast = slice_by_index(begin = var_4034_begin_0, end = var_4034_end_0, end_mask = var_4034_end_mask_0, x = v_23_cast);
            tensor<string, []> var_4038_equation_0 = const()[name = tensor<string, []>("op_4038_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4038_cast = einsum(equation = var_4038_equation_0, values = (var_3880_cast, var_3797_cast));
            tensor<fp16, []> var_4039_to_fp16 = const()[name = tensor<string, []>("op_4039_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_241_cast = mul(x = var_4038_cast, y = var_4039_to_fp16);
            tensor<string, []> var_4042_equation_0 = const()[name = tensor<string, []>("op_4042_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4042_cast = einsum(equation = var_4042_equation_0, values = (var_3884_cast, var_3801_cast));
            tensor<fp16, []> var_4043_to_fp16 = const()[name = tensor<string, []>("op_4043_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_243_cast = mul(x = var_4042_cast, y = var_4043_to_fp16);
            tensor<string, []> var_4046_equation_0 = const()[name = tensor<string, []>("op_4046_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4046_cast = einsum(equation = var_4046_equation_0, values = (var_3888_cast, var_3805_cast));
            tensor<fp16, []> var_4047_to_fp16 = const()[name = tensor<string, []>("op_4047_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_245_cast = mul(x = var_4046_cast, y = var_4047_to_fp16);
            tensor<string, []> var_4050_equation_0 = const()[name = tensor<string, []>("op_4050_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4050_cast = einsum(equation = var_4050_equation_0, values = (var_3892_cast, var_3809_cast));
            tensor<fp16, []> var_4051_to_fp16 = const()[name = tensor<string, []>("op_4051_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_247_cast = mul(x = var_4050_cast, y = var_4051_to_fp16);
            tensor<string, []> var_4054_equation_0 = const()[name = tensor<string, []>("op_4054_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4054_cast = einsum(equation = var_4054_equation_0, values = (var_3896_cast, var_3813_cast));
            tensor<fp16, []> var_4055_to_fp16 = const()[name = tensor<string, []>("op_4055_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_249_cast = mul(x = var_4054_cast, y = var_4055_to_fp16);
            tensor<string, []> var_4058_equation_0 = const()[name = tensor<string, []>("op_4058_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4058_cast = einsum(equation = var_4058_equation_0, values = (var_3900_cast, var_3817_cast));
            tensor<fp16, []> var_4059_to_fp16 = const()[name = tensor<string, []>("op_4059_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_251_cast = mul(x = var_4058_cast, y = var_4059_to_fp16);
            tensor<string, []> var_4062_equation_0 = const()[name = tensor<string, []>("op_4062_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4062_cast = einsum(equation = var_4062_equation_0, values = (var_3904_cast, var_3821_cast));
            tensor<fp16, []> var_4063_to_fp16 = const()[name = tensor<string, []>("op_4063_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_253_cast = mul(x = var_4062_cast, y = var_4063_to_fp16);
            tensor<string, []> var_4066_equation_0 = const()[name = tensor<string, []>("op_4066_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4066_cast = einsum(equation = var_4066_equation_0, values = (var_3908_cast, var_3825_cast));
            tensor<fp16, []> var_4067_to_fp16 = const()[name = tensor<string, []>("op_4067_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_255_cast = mul(x = var_4066_cast, y = var_4067_to_fp16);
            tensor<string, []> var_4070_equation_0 = const()[name = tensor<string, []>("op_4070_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4070_cast = einsum(equation = var_4070_equation_0, values = (var_3912_cast, var_3829_cast));
            tensor<fp16, []> var_4071_to_fp16 = const()[name = tensor<string, []>("op_4071_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_257_cast = mul(x = var_4070_cast, y = var_4071_to_fp16);
            tensor<string, []> var_4074_equation_0 = const()[name = tensor<string, []>("op_4074_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4074_cast = einsum(equation = var_4074_equation_0, values = (var_3916_cast, var_3833_cast));
            tensor<fp16, []> var_4075_to_fp16 = const()[name = tensor<string, []>("op_4075_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_259_cast = mul(x = var_4074_cast, y = var_4075_to_fp16);
            tensor<string, []> var_4078_equation_0 = const()[name = tensor<string, []>("op_4078_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4078_cast = einsum(equation = var_4078_equation_0, values = (var_3920_cast, var_3837_cast));
            tensor<fp16, []> var_4079_to_fp16 = const()[name = tensor<string, []>("op_4079_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_261_cast = mul(x = var_4078_cast, y = var_4079_to_fp16);
            tensor<string, []> var_4082_equation_0 = const()[name = tensor<string, []>("op_4082_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4082_cast = einsum(equation = var_4082_equation_0, values = (var_3924_cast, var_3841_cast));
            tensor<fp16, []> var_4083_to_fp16 = const()[name = tensor<string, []>("op_4083_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_263_cast = mul(x = var_4082_cast, y = var_4083_to_fp16);
            tensor<string, []> var_4086_equation_0 = const()[name = tensor<string, []>("op_4086_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4086_cast = einsum(equation = var_4086_equation_0, values = (var_3928_cast, var_3845_cast));
            tensor<fp16, []> var_4087_to_fp16 = const()[name = tensor<string, []>("op_4087_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_265_cast = mul(x = var_4086_cast, y = var_4087_to_fp16);
            tensor<string, []> var_4090_equation_0 = const()[name = tensor<string, []>("op_4090_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4090_cast = einsum(equation = var_4090_equation_0, values = (var_3932_cast, var_3849_cast));
            tensor<fp16, []> var_4091_to_fp16 = const()[name = tensor<string, []>("op_4091_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_267_cast = mul(x = var_4090_cast, y = var_4091_to_fp16);
            tensor<string, []> var_4094_equation_0 = const()[name = tensor<string, []>("op_4094_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4094_cast = einsum(equation = var_4094_equation_0, values = (var_3936_cast, var_3853_cast));
            tensor<fp16, []> var_4095_to_fp16 = const()[name = tensor<string, []>("op_4095_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_269_cast = mul(x = var_4094_cast, y = var_4095_to_fp16);
            tensor<string, []> var_4098_equation_0 = const()[name = tensor<string, []>("op_4098_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4098_cast = einsum(equation = var_4098_equation_0, values = (var_3940_cast, var_3857_cast));
            tensor<fp16, []> var_4099_to_fp16 = const()[name = tensor<string, []>("op_4099_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_271_cast = mul(x = var_4098_cast, y = var_4099_to_fp16);
            tensor<string, []> var_4102_equation_0 = const()[name = tensor<string, []>("op_4102_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4102_cast = einsum(equation = var_4102_equation_0, values = (var_3944_cast, var_3861_cast));
            tensor<fp16, []> var_4103_to_fp16 = const()[name = tensor<string, []>("op_4103_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_273_cast = mul(x = var_4102_cast, y = var_4103_to_fp16);
            tensor<string, []> var_4106_equation_0 = const()[name = tensor<string, []>("op_4106_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4106_cast = einsum(equation = var_4106_equation_0, values = (var_3948_cast, var_3865_cast));
            tensor<fp16, []> var_4107_to_fp16 = const()[name = tensor<string, []>("op_4107_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_275_cast = mul(x = var_4106_cast, y = var_4107_to_fp16);
            tensor<string, []> var_4110_equation_0 = const()[name = tensor<string, []>("op_4110_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4110_cast = einsum(equation = var_4110_equation_0, values = (var_3952_cast, var_3869_cast));
            tensor<fp16, []> var_4111_to_fp16 = const()[name = tensor<string, []>("op_4111_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_277_cast = mul(x = var_4110_cast, y = var_4111_to_fp16);
            tensor<string, []> var_4114_equation_0 = const()[name = tensor<string, []>("op_4114_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4114_cast = einsum(equation = var_4114_equation_0, values = (var_3956_cast, var_3873_cast));
            tensor<fp16, []> var_4115_to_fp16 = const()[name = tensor<string, []>("op_4115_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_279_cast = mul(x = var_4114_cast, y = var_4115_to_fp16);
            tensor<fp16, [2, 77, 1, 256]> var_4117_cast = softmax(axis = var_2254, x = aw_241_cast);
            tensor<fp16, [2, 77, 1, 256]> var_4118_cast = softmax(axis = var_2254, x = aw_243_cast);
            tensor<fp16, [2, 77, 1, 256]> var_4119_cast = softmax(axis = var_2254, x = aw_245_cast);
            tensor<fp16, [2, 77, 1, 256]> var_4120_cast = softmax(axis = var_2254, x = aw_247_cast);
            tensor<fp16, [2, 77, 1, 256]> var_4121_cast = softmax(axis = var_2254, x = aw_249_cast);
            tensor<fp16, [2, 77, 1, 256]> var_4122_cast = softmax(axis = var_2254, x = aw_251_cast);
            tensor<fp16, [2, 77, 1, 256]> var_4123_cast = softmax(axis = var_2254, x = aw_253_cast);
            tensor<fp16, [2, 77, 1, 256]> var_4124_cast = softmax(axis = var_2254, x = aw_255_cast);
            tensor<fp16, [2, 77, 1, 256]> var_4125_cast = softmax(axis = var_2254, x = aw_257_cast);
            tensor<fp16, [2, 77, 1, 256]> var_4126_cast = softmax(axis = var_2254, x = aw_259_cast);
            tensor<fp16, [2, 77, 1, 256]> var_4127_cast = softmax(axis = var_2254, x = aw_261_cast);
            tensor<fp16, [2, 77, 1, 256]> var_4128_cast = softmax(axis = var_2254, x = aw_263_cast);
            tensor<fp16, [2, 77, 1, 256]> var_4129_cast = softmax(axis = var_2254, x = aw_265_cast);
            tensor<fp16, [2, 77, 1, 256]> var_4130_cast = softmax(axis = var_2254, x = aw_267_cast);
            tensor<fp16, [2, 77, 1, 256]> var_4131_cast = softmax(axis = var_2254, x = aw_269_cast);
            tensor<fp16, [2, 77, 1, 256]> var_4132_cast = softmax(axis = var_2254, x = aw_271_cast);
            tensor<fp16, [2, 77, 1, 256]> var_4133_cast = softmax(axis = var_2254, x = aw_273_cast);
            tensor<fp16, [2, 77, 1, 256]> var_4134_cast = softmax(axis = var_2254, x = aw_275_cast);
            tensor<fp16, [2, 77, 1, 256]> var_4135_cast = softmax(axis = var_2254, x = aw_277_cast);
            tensor<fp16, [2, 77, 1, 256]> var_4136_cast = softmax(axis = var_2254, x = aw_279_cast);
            tensor<string, []> var_4138_equation_0 = const()[name = tensor<string, []>("op_4138_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4138_cast = einsum(equation = var_4138_equation_0, values = (var_3958_cast, var_4117_cast));
            tensor<string, []> var_4140_equation_0 = const()[name = tensor<string, []>("op_4140_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4140_cast = einsum(equation = var_4140_equation_0, values = (var_3962_cast, var_4118_cast));
            tensor<string, []> var_4142_equation_0 = const()[name = tensor<string, []>("op_4142_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4142_cast = einsum(equation = var_4142_equation_0, values = (var_3966_cast, var_4119_cast));
            tensor<string, []> var_4144_equation_0 = const()[name = tensor<string, []>("op_4144_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4144_cast = einsum(equation = var_4144_equation_0, values = (var_3970_cast, var_4120_cast));
            tensor<string, []> var_4146_equation_0 = const()[name = tensor<string, []>("op_4146_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4146_cast = einsum(equation = var_4146_equation_0, values = (var_3974_cast, var_4121_cast));
            tensor<string, []> var_4148_equation_0 = const()[name = tensor<string, []>("op_4148_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4148_cast = einsum(equation = var_4148_equation_0, values = (var_3978_cast, var_4122_cast));
            tensor<string, []> var_4150_equation_0 = const()[name = tensor<string, []>("op_4150_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4150_cast = einsum(equation = var_4150_equation_0, values = (var_3982_cast, var_4123_cast));
            tensor<string, []> var_4152_equation_0 = const()[name = tensor<string, []>("op_4152_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4152_cast = einsum(equation = var_4152_equation_0, values = (var_3986_cast, var_4124_cast));
            tensor<string, []> var_4154_equation_0 = const()[name = tensor<string, []>("op_4154_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4154_cast = einsum(equation = var_4154_equation_0, values = (var_3990_cast, var_4125_cast));
            tensor<string, []> var_4156_equation_0 = const()[name = tensor<string, []>("op_4156_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4156_cast = einsum(equation = var_4156_equation_0, values = (var_3994_cast, var_4126_cast));
            tensor<string, []> var_4158_equation_0 = const()[name = tensor<string, []>("op_4158_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4158_cast = einsum(equation = var_4158_equation_0, values = (var_3998_cast, var_4127_cast));
            tensor<string, []> var_4160_equation_0 = const()[name = tensor<string, []>("op_4160_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4160_cast = einsum(equation = var_4160_equation_0, values = (var_4002_cast, var_4128_cast));
            tensor<string, []> var_4162_equation_0 = const()[name = tensor<string, []>("op_4162_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4162_cast = einsum(equation = var_4162_equation_0, values = (var_4006_cast, var_4129_cast));
            tensor<string, []> var_4164_equation_0 = const()[name = tensor<string, []>("op_4164_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4164_cast = einsum(equation = var_4164_equation_0, values = (var_4010_cast, var_4130_cast));
            tensor<string, []> var_4166_equation_0 = const()[name = tensor<string, []>("op_4166_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4166_cast = einsum(equation = var_4166_equation_0, values = (var_4014_cast, var_4131_cast));
            tensor<string, []> var_4168_equation_0 = const()[name = tensor<string, []>("op_4168_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4168_cast = einsum(equation = var_4168_equation_0, values = (var_4018_cast, var_4132_cast));
            tensor<string, []> var_4170_equation_0 = const()[name = tensor<string, []>("op_4170_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4170_cast = einsum(equation = var_4170_equation_0, values = (var_4022_cast, var_4133_cast));
            tensor<string, []> var_4172_equation_0 = const()[name = tensor<string, []>("op_4172_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4172_cast = einsum(equation = var_4172_equation_0, values = (var_4026_cast, var_4134_cast));
            tensor<string, []> var_4174_equation_0 = const()[name = tensor<string, []>("op_4174_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4174_cast = einsum(equation = var_4174_equation_0, values = (var_4030_cast, var_4135_cast));
            tensor<string, []> var_4176_equation_0 = const()[name = tensor<string, []>("op_4176_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4176_cast = einsum(equation = var_4176_equation_0, values = (var_4034_cast, var_4136_cast));
            tensor<bool, []> input_161_interleave_0 = const()[name = tensor<string, []>("input_161_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 256]> input_161_cast = concat(axis = var_2254, interleave = input_161_interleave_0, values = (var_4138_cast, var_4140_cast, var_4142_cast, var_4144_cast, var_4146_cast, var_4148_cast, var_4150_cast, var_4152_cast, var_4154_cast, var_4156_cast, var_4158_cast, var_4160_cast, var_4162_cast, var_4164_cast, var_4166_cast, var_4168_cast, var_4170_cast, var_4172_cast, var_4174_cast, var_4176_cast));
            tensor<int32, [2]> var_4182 = const()[name = tensor<string, []>("op_4182"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4184 = const()[name = tensor<string, []>("op_4184"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_4186_pad_type_0 = const()[name = tensor<string, []>("op_4186_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_4186_pad_0 = const()[name = tensor<string, []>("op_4186_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(308047552)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(311324416)))];
            tensor<fp16, [2, 1280, 1, 256]> var_4186_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_4184, groups = var_2254, pad = var_4186_pad_0, pad_type = var_4186_pad_type_0, strides = var_4182, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_161_cast);
            tensor<fp16, [2, 1280, 1, 256]> inputs_35_cast = add(x = var_4186_cast, y = inputs_33_cast);
            tensor<int32, [1]> var_4190 = const()[name = tensor<string, []>("op_4190"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_35_cast = reduce_mean(axes = var_4190, keep_dims = var_2249, x = inputs_35_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_35_cast = sub(x = inputs_35_cast, y = channels_mean_35_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_sq_35_cast = mul(x = zero_mean_35_cast, y = zero_mean_35_cast);
            tensor<int32, [1]> var_4194 = const()[name = tensor<string, []>("op_4194"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_4195_cast = reduce_mean(axes = var_4194, keep_dims = var_2249, x = zero_mean_sq_35_cast);
            tensor<fp16, []> var_4196_to_fp16 = const()[name = tensor<string, []>("op_4196_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_4197_cast = add(x = var_4195_cast, y = var_4196_to_fp16);
            tensor<fp16, []> denom_35_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_35_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 256]> denom_35_cast = rsqrt(epsilon = denom_35_epsilon_0_to_fp16, x = var_4197_cast);
            tensor<fp16, [2, 1280, 1, 256]> out_35_cast = mul(x = zero_mean_35_cast, y = denom_35_cast);
            tensor<fp16, [1, 1280, 1, 1]> var_4201_to_fp16 = const()[name = tensor<string, []>("op_4201_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(311327040)))];
            tensor<fp16, [2, 1280, 1, 256]> var_4202_cast = add(x = out_35_cast, y = var_4201_to_fp16);
            tensor<fp16, [1, 1280, 1, 1]> var_4204_to_fp16 = const()[name = tensor<string, []>("op_4204_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(311329664)))];
            tensor<fp16, [2, 1280, 1, 256]> input_163_cast = mul(x = var_4202_cast, y = var_4204_to_fp16);
            tensor<int32, [2]> var_4212 = const()[name = tensor<string, []>("op_4212"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4214 = const()[name = tensor<string, []>("op_4214"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_4216_pad_type_0 = const()[name = tensor<string, []>("op_4216_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_4216_pad_0 = const()[name = tensor<string, []>("op_4216_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(311332288)))];
            tensor<fp16, [10240]> down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(337546752)))];
            tensor<fp16, [2, 10240, 1, 256]> var_4216_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_4214, groups = var_2254, pad = var_4216_pad_0, pad_type = var_4216_pad_type_0, strides = var_4212, weight = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_163_cast);
            tensor<int32, [2]> var_4217_split_sizes_0 = const()[name = tensor<string, []>("op_4217_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_4217_axis_0 = const()[name = tensor<string, []>("op_4217_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 256]> var_4217_cast_0, tensor<fp16, [2, 5120, 1, 256]> var_4217_cast_1 = split(axis = var_4217_axis_0, split_sizes = var_4217_split_sizes_0, x = var_4216_cast);
            tensor<string, []> var_4219_mode_0 = const()[name = tensor<string, []>("op_4219_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 256]> var_4219_cast = gelu(mode = var_4219_mode_0, x = var_4217_cast_1);
            tensor<fp16, [2, 5120, 1, 256]> input_165_cast = mul(x = var_4217_cast_0, y = var_4219_cast);
            tensor<int32, [2]> var_4223 = const()[name = tensor<string, []>("op_4223"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4225 = const()[name = tensor<string, []>("op_4225"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_4227_pad_type_0 = const()[name = tensor<string, []>("op_4227_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_4227_pad_0 = const()[name = tensor<string, []>("op_4227_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(337567296)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(350674560)))];
            tensor<fp16, [2, 1280, 1, 256]> var_4227_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_4225, groups = var_2254, pad = var_4227_pad_0, pad_type = var_4227_pad_type_0, strides = var_4223, weight = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_165_cast);
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_105_cast = add(x = var_4227_cast, y = inputs_35_cast);
            tensor<int32, [4]> var_4229 = const()[name = tensor<string, []>("op_4229"), val = tensor<int32, [4]>([2, 1280, 16, 16])];
            tensor<fp16, [2, 1280, 16, 16]> input_167_cast = reshape(shape = var_4229, x = hidden_states_105_cast);
            tensor<int32, [2]> var_4233 = const()[name = tensor<string, []>("op_4233"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4235 = const()[name = tensor<string, []>("op_4235"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_107_pad_type_0 = const()[name = tensor<string, []>("hidden_states_107_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_107_pad_0 = const()[name = tensor<string, []>("hidden_states_107_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_2_attentions_1_proj_out_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_proj_out_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(350677184)))];
            tensor<fp16, [1280]> down_blocks_2_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_attentions_1_proj_out_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(353954048)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_107_cast = conv(bias = down_blocks_2_attentions_1_proj_out_bias_to_fp16, dilations = var_4235, groups = var_2254, pad = hidden_states_107_pad_0, pad_type = hidden_states_107_pad_type_0, strides = var_4233, weight = down_blocks_2_attentions_1_proj_out_weight_to_fp16, x = input_167_cast);
            tensor<fp16, [2, 1280, 16, 16]> input_169_cast = add(x = hidden_states_107_cast, y = hidden_states_95_cast);
            tensor<int32, [2]> var_4242 = const()[name = tensor<string, []>("op_4242"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [2]> var_4244 = const()[name = tensor<string, []>("op_4244"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> input_171_pad_type_0 = const()[name = tensor<string, []>("input_171_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> input_171_pad_0 = const()[name = tensor<string, []>("input_171_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> down_blocks_2_downsamplers_0_conv_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_downsamplers_0_conv_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(353956672)))];
            tensor<fp16, [1280]> down_blocks_2_downsamplers_0_conv_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_2_downsamplers_0_conv_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(383447936)))];
            tensor<fp16, [2, 1280, 8, 8]> input_171_cast = conv(bias = down_blocks_2_downsamplers_0_conv_bias_to_fp16, dilations = var_4244, groups = var_2254, pad = input_171_pad_0, pad_type = input_171_pad_type_0, strides = var_4242, weight = down_blocks_2_downsamplers_0_conv_weight_to_fp16, x = input_169_cast);
            tensor<int32, []> var_4256 = const()[name = tensor<string, []>("op_4256"), val = tensor<int32, []>(1)];
            tensor<int32, [5]> reshape_72_shape_0 = const()[name = tensor<string, []>("reshape_72_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 8, 8])];
            tensor<fp16, [2, 32, 40, 8, 8]> reshape_72_cast = reshape(shape = reshape_72_shape_0, x = input_171_cast);
            tensor<int32, [3]> reduce_mean_54_axes_0 = const()[name = tensor<string, []>("reduce_mean_54_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_54_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_54_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_54_cast = reduce_mean(axes = reduce_mean_54_axes_0, keep_dims = reduce_mean_54_keep_dims_0, x = reshape_72_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> sub_36_cast = sub(x = reshape_72_cast, y = reduce_mean_54_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> square_18_cast = square(x = sub_36_cast);
            tensor<int32, [3]> reduce_mean_56_axes_0 = const()[name = tensor<string, []>("reduce_mean_56_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_56_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_56_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_56_cast = reduce_mean(axes = reduce_mean_56_axes_0, keep_dims = reduce_mean_56_keep_dims_0, x = square_18_cast);
            tensor<fp16, []> add_36_y_0_to_fp16 = const()[name = tensor<string, []>("add_36_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_36_cast = add(x = reduce_mean_56_cast, y = add_36_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_18_cast = sqrt(x = add_36_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> real_div_18_cast = real_div(x = sub_36_cast, y = sqrt_18_cast);
            tensor<int32, [4]> reshape_73_shape_0 = const()[name = tensor<string, []>("reshape_73_shape_0"), val = tensor<int32, [4]>([2, 1280, 8, 8])];
            tensor<fp16, [2, 1280, 8, 8]> reshape_73_cast = reshape(shape = reshape_73_shape_0, x = real_div_18_cast);
            tensor<fp16, [1280]> add_37_mean_0_to_fp16 = const()[name = tensor<string, []>("add_37_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(383450560)))];
            tensor<fp16, [1280]> add_37_variance_0_to_fp16 = const()[name = tensor<string, []>("add_37_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(383453184)))];
            tensor<fp16, [1280]> add_37_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_37_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(383455808)))];
            tensor<fp16, [1280]> add_37_beta_0_to_fp16 = const()[name = tensor<string, []>("add_37_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(383458432)))];
            tensor<fp16, []> add_37_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_37_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 8, 8]> add_37_cast = batch_norm(beta = add_37_beta_0_to_fp16, epsilon = add_37_epsilon_0_to_fp16, gamma = add_37_gamma_0_to_fp16, mean = add_37_mean_0_to_fp16, variance = add_37_variance_0_to_fp16, x = reshape_73_cast);
            tensor<fp16, [2, 1280, 8, 8]> input_175_cast = silu(x = add_37_cast);
            tensor<int32, [2]> var_4272 = const()[name = tensor<string, []>("op_4272"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4274 = const()[name = tensor<string, []>("op_4274"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_109_pad_type_0 = const()[name = tensor<string, []>("hidden_states_109_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_109_pad_0 = const()[name = tensor<string, []>("hidden_states_109_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> down_blocks_3_resnets_0_conv1_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_3_resnets_0_conv1_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(383461056)))];
            tensor<fp16, [1280]> down_blocks_3_resnets_0_conv1_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_3_resnets_0_conv1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(412952320)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_109_cast = conv(bias = down_blocks_3_resnets_0_conv1_bias_to_fp16, dilations = var_4274, groups = var_4256, pad = hidden_states_109_pad_0, pad_type = hidden_states_109_pad_type_0, strides = var_4272, weight = down_blocks_3_resnets_0_conv1_weight_to_fp16, x = input_175_cast);
            tensor<int32, [2]> var_4280 = const()[name = tensor<string, []>("op_4280"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4282 = const()[name = tensor<string, []>("op_4282"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_13_pad_type_0 = const()[name = tensor<string, []>("temb_13_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_13_pad_0 = const()[name = tensor<string, []>("temb_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_3_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_3_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(412954944)))];
            tensor<fp16, [1280]> down_blocks_3_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_3_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(416231808)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_13_cast = conv(bias = down_blocks_3_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_4282, groups = var_4256, pad = temb_13_pad_0, pad_type = temb_13_pad_type_0, strides = var_4280, weight = down_blocks_3_resnets_0_time_emb_proj_weight_to_fp16, x = input_15_cast);
            tensor<fp16, [2, 1280, 8, 8]> input_179_cast = add(x = hidden_states_109_cast, y = temb_13_cast);
            tensor<int32, [5]> reshape_76_shape_0 = const()[name = tensor<string, []>("reshape_76_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 8, 8])];
            tensor<fp16, [2, 32, 40, 8, 8]> reshape_76_cast = reshape(shape = reshape_76_shape_0, x = input_179_cast);
            tensor<int32, [3]> reduce_mean_57_axes_0 = const()[name = tensor<string, []>("reduce_mean_57_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_57_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_57_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_57_cast = reduce_mean(axes = reduce_mean_57_axes_0, keep_dims = reduce_mean_57_keep_dims_0, x = reshape_76_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> sub_38_cast = sub(x = reshape_76_cast, y = reduce_mean_57_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> square_19_cast = square(x = sub_38_cast);
            tensor<int32, [3]> reduce_mean_59_axes_0 = const()[name = tensor<string, []>("reduce_mean_59_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_59_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_59_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_59_cast = reduce_mean(axes = reduce_mean_59_axes_0, keep_dims = reduce_mean_59_keep_dims_0, x = square_19_cast);
            tensor<fp16, []> add_38_y_0_to_fp16 = const()[name = tensor<string, []>("add_38_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_38_cast = add(x = reduce_mean_59_cast, y = add_38_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_19_cast = sqrt(x = add_38_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> real_div_19_cast = real_div(x = sub_38_cast, y = sqrt_19_cast);
            tensor<int32, [4]> reshape_77_shape_0 = const()[name = tensor<string, []>("reshape_77_shape_0"), val = tensor<int32, [4]>([2, 1280, 8, 8])];
            tensor<fp16, [2, 1280, 8, 8]> reshape_77_cast = reshape(shape = reshape_77_shape_0, x = real_div_19_cast);
            tensor<fp16, [1280]> add_39_mean_0_to_fp16 = const()[name = tensor<string, []>("add_39_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(416234432)))];
            tensor<fp16, [1280]> add_39_variance_0_to_fp16 = const()[name = tensor<string, []>("add_39_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(416237056)))];
            tensor<fp16, [1280]> add_39_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_39_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(416239680)))];
            tensor<fp16, [1280]> add_39_beta_0_to_fp16 = const()[name = tensor<string, []>("add_39_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(416242304)))];
            tensor<fp16, []> add_39_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_39_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 8, 8]> add_39_cast = batch_norm(beta = add_39_beta_0_to_fp16, epsilon = add_39_epsilon_0_to_fp16, gamma = add_39_gamma_0_to_fp16, mean = add_39_mean_0_to_fp16, variance = add_39_variance_0_to_fp16, x = reshape_77_cast);
            tensor<fp16, [2, 1280, 8, 8]> input_183_cast = silu(x = add_39_cast);
            tensor<int32, [2]> var_4292 = const()[name = tensor<string, []>("op_4292"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4294 = const()[name = tensor<string, []>("op_4294"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_111_pad_type_0 = const()[name = tensor<string, []>("hidden_states_111_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_111_pad_0 = const()[name = tensor<string, []>("hidden_states_111_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> down_blocks_3_resnets_0_conv2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_3_resnets_0_conv2_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(416244928)))];
            tensor<fp16, [1280]> down_blocks_3_resnets_0_conv2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_3_resnets_0_conv2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(445736192)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_111_cast = conv(bias = down_blocks_3_resnets_0_conv2_bias_to_fp16, dilations = var_4294, groups = var_4256, pad = hidden_states_111_pad_0, pad_type = hidden_states_111_pad_type_0, strides = var_4292, weight = down_blocks_3_resnets_0_conv2_weight_to_fp16, x = input_183_cast);
            tensor<fp16, [2, 1280, 8, 8]> input_185_cast = add(x = input_171_cast, y = hidden_states_111_cast);
            tensor<int32, [5]> reshape_80_shape_0 = const()[name = tensor<string, []>("reshape_80_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 8, 8])];
            tensor<fp16, [2, 32, 40, 8, 8]> reshape_80_cast = reshape(shape = reshape_80_shape_0, x = input_185_cast);
            tensor<int32, [3]> reduce_mean_60_axes_0 = const()[name = tensor<string, []>("reduce_mean_60_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_60_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_60_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_60_cast = reduce_mean(axes = reduce_mean_60_axes_0, keep_dims = reduce_mean_60_keep_dims_0, x = reshape_80_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> sub_40_cast = sub(x = reshape_80_cast, y = reduce_mean_60_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> square_20_cast = square(x = sub_40_cast);
            tensor<int32, [3]> reduce_mean_62_axes_0 = const()[name = tensor<string, []>("reduce_mean_62_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_62_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_62_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_62_cast = reduce_mean(axes = reduce_mean_62_axes_0, keep_dims = reduce_mean_62_keep_dims_0, x = square_20_cast);
            tensor<fp16, []> add_40_y_0_to_fp16 = const()[name = tensor<string, []>("add_40_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_40_cast = add(x = reduce_mean_62_cast, y = add_40_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_20_cast = sqrt(x = add_40_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> real_div_20_cast = real_div(x = sub_40_cast, y = sqrt_20_cast);
            tensor<int32, [4]> reshape_81_shape_0 = const()[name = tensor<string, []>("reshape_81_shape_0"), val = tensor<int32, [4]>([2, 1280, 8, 8])];
            tensor<fp16, [2, 1280, 8, 8]> reshape_81_cast = reshape(shape = reshape_81_shape_0, x = real_div_20_cast);
            tensor<fp16, [1280]> add_41_mean_0_to_fp16 = const()[name = tensor<string, []>("add_41_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(445738816)))];
            tensor<fp16, [1280]> add_41_variance_0_to_fp16 = const()[name = tensor<string, []>("add_41_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(445741440)))];
            tensor<fp16, [1280]> add_41_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_41_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(445744064)))];
            tensor<fp16, [1280]> add_41_beta_0_to_fp16 = const()[name = tensor<string, []>("add_41_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(445746688)))];
            tensor<fp16, []> add_41_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_41_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 8, 8]> add_41_cast = batch_norm(beta = add_41_beta_0_to_fp16, epsilon = add_41_epsilon_0_to_fp16, gamma = add_41_gamma_0_to_fp16, mean = add_41_mean_0_to_fp16, variance = add_41_variance_0_to_fp16, x = reshape_81_cast);
            tensor<fp16, [2, 1280, 8, 8]> input_189_cast = silu(x = add_41_cast);
            tensor<int32, [2]> var_4309 = const()[name = tensor<string, []>("op_4309"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4311 = const()[name = tensor<string, []>("op_4311"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_113_pad_type_0 = const()[name = tensor<string, []>("hidden_states_113_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_113_pad_0 = const()[name = tensor<string, []>("hidden_states_113_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> down_blocks_3_resnets_1_conv1_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_3_resnets_1_conv1_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(445749312)))];
            tensor<fp16, [1280]> down_blocks_3_resnets_1_conv1_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_3_resnets_1_conv1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(475240576)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_113_cast = conv(bias = down_blocks_3_resnets_1_conv1_bias_to_fp16, dilations = var_4311, groups = var_4256, pad = hidden_states_113_pad_0, pad_type = hidden_states_113_pad_type_0, strides = var_4309, weight = down_blocks_3_resnets_1_conv1_weight_to_fp16, x = input_189_cast);
            tensor<int32, [2]> var_4317 = const()[name = tensor<string, []>("op_4317"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4319 = const()[name = tensor<string, []>("op_4319"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_15_pad_type_0 = const()[name = tensor<string, []>("temb_15_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_15_pad_0 = const()[name = tensor<string, []>("temb_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> down_blocks_3_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_3_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(475243200)))];
            tensor<fp16, [1280]> down_blocks_3_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_3_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(478520064)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_15_cast = conv(bias = down_blocks_3_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_4319, groups = var_4256, pad = temb_15_pad_0, pad_type = temb_15_pad_type_0, strides = var_4317, weight = down_blocks_3_resnets_1_time_emb_proj_weight_to_fp16, x = input_15_cast);
            tensor<fp16, [2, 1280, 8, 8]> input_193_cast = add(x = hidden_states_113_cast, y = temb_15_cast);
            tensor<int32, [5]> reshape_84_shape_0 = const()[name = tensor<string, []>("reshape_84_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 8, 8])];
            tensor<fp16, [2, 32, 40, 8, 8]> reshape_84_cast = reshape(shape = reshape_84_shape_0, x = input_193_cast);
            tensor<int32, [3]> reduce_mean_63_axes_0 = const()[name = tensor<string, []>("reduce_mean_63_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_63_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_63_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_63_cast = reduce_mean(axes = reduce_mean_63_axes_0, keep_dims = reduce_mean_63_keep_dims_0, x = reshape_84_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> sub_42_cast = sub(x = reshape_84_cast, y = reduce_mean_63_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> square_21_cast = square(x = sub_42_cast);
            tensor<int32, [3]> reduce_mean_65_axes_0 = const()[name = tensor<string, []>("reduce_mean_65_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_65_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_65_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_65_cast = reduce_mean(axes = reduce_mean_65_axes_0, keep_dims = reduce_mean_65_keep_dims_0, x = square_21_cast);
            tensor<fp16, []> add_42_y_0_to_fp16 = const()[name = tensor<string, []>("add_42_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_42_cast = add(x = reduce_mean_65_cast, y = add_42_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_21_cast = sqrt(x = add_42_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> real_div_21_cast = real_div(x = sub_42_cast, y = sqrt_21_cast);
            tensor<int32, [4]> reshape_85_shape_0 = const()[name = tensor<string, []>("reshape_85_shape_0"), val = tensor<int32, [4]>([2, 1280, 8, 8])];
            tensor<fp16, [2, 1280, 8, 8]> reshape_85_cast = reshape(shape = reshape_85_shape_0, x = real_div_21_cast);
            tensor<fp16, [1280]> add_43_mean_0_to_fp16 = const()[name = tensor<string, []>("add_43_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(478522688)))];
            tensor<fp16, [1280]> add_43_variance_0_to_fp16 = const()[name = tensor<string, []>("add_43_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(478525312)))];
            tensor<fp16, [1280]> add_43_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_43_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(478527936)))];
            tensor<fp16, [1280]> add_43_beta_0_to_fp16 = const()[name = tensor<string, []>("add_43_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(478530560)))];
            tensor<fp16, []> add_43_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_43_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 8, 8]> add_43_cast = batch_norm(beta = add_43_beta_0_to_fp16, epsilon = add_43_epsilon_0_to_fp16, gamma = add_43_gamma_0_to_fp16, mean = add_43_mean_0_to_fp16, variance = add_43_variance_0_to_fp16, x = reshape_85_cast);
            tensor<fp16, [2, 1280, 8, 8]> input_197_cast = silu(x = add_43_cast);
            tensor<int32, [2]> var_4329 = const()[name = tensor<string, []>("op_4329"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4331 = const()[name = tensor<string, []>("op_4331"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_115_pad_type_0 = const()[name = tensor<string, []>("hidden_states_115_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_115_pad_0 = const()[name = tensor<string, []>("hidden_states_115_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> down_blocks_3_resnets_1_conv2_weight_to_fp16 = const()[name = tensor<string, []>("down_blocks_3_resnets_1_conv2_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(478533184)))];
            tensor<fp16, [1280]> down_blocks_3_resnets_1_conv2_bias_to_fp16 = const()[name = tensor<string, []>("down_blocks_3_resnets_1_conv2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(508024448)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_115_cast = conv(bias = down_blocks_3_resnets_1_conv2_bias_to_fp16, dilations = var_4331, groups = var_4256, pad = hidden_states_115_pad_0, pad_type = hidden_states_115_pad_type_0, strides = var_4329, weight = down_blocks_3_resnets_1_conv2_weight_to_fp16, x = input_197_cast);
            tensor<fp16, [2, 1280, 8, 8]> input_199_cast = add(x = input_185_cast, y = hidden_states_115_cast);
            tensor<bool, []> var_4366 = const()[name = tensor<string, []>("op_4366"), val = tensor<bool, []>(true)];
            tensor<int32, []> var_4371 = const()[name = tensor<string, []>("op_4371"), val = tensor<int32, []>(1)];
            tensor<int32, [5]> reshape_88_shape_0 = const()[name = tensor<string, []>("reshape_88_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 8, 8])];
            tensor<fp16, [2, 32, 40, 8, 8]> reshape_88_cast = reshape(shape = reshape_88_shape_0, x = input_199_cast);
            tensor<int32, [3]> reduce_mean_66_axes_0 = const()[name = tensor<string, []>("reduce_mean_66_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_66_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_66_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_66_cast = reduce_mean(axes = reduce_mean_66_axes_0, keep_dims = reduce_mean_66_keep_dims_0, x = reshape_88_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> sub_44_cast = sub(x = reshape_88_cast, y = reduce_mean_66_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> square_22_cast = square(x = sub_44_cast);
            tensor<int32, [3]> reduce_mean_68_axes_0 = const()[name = tensor<string, []>("reduce_mean_68_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_68_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_68_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_68_cast = reduce_mean(axes = reduce_mean_68_axes_0, keep_dims = reduce_mean_68_keep_dims_0, x = square_22_cast);
            tensor<fp16, []> add_44_y_0_to_fp16 = const()[name = tensor<string, []>("add_44_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_44_cast = add(x = reduce_mean_68_cast, y = add_44_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_22_cast = sqrt(x = add_44_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> real_div_22_cast = real_div(x = sub_44_cast, y = sqrt_22_cast);
            tensor<int32, [4]> reshape_89_shape_0 = const()[name = tensor<string, []>("reshape_89_shape_0"), val = tensor<int32, [4]>([2, 1280, 8, 8])];
            tensor<fp16, [2, 1280, 8, 8]> reshape_89_cast = reshape(shape = reshape_89_shape_0, x = real_div_22_cast);
            tensor<fp16, [1280]> add_45_mean_0_to_fp16 = const()[name = tensor<string, []>("add_45_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(508027072)))];
            tensor<fp16, [1280]> add_45_variance_0_to_fp16 = const()[name = tensor<string, []>("add_45_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(508029696)))];
            tensor<fp16, [1280]> add_45_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_45_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(508032320)))];
            tensor<fp16, [1280]> add_45_beta_0_to_fp16 = const()[name = tensor<string, []>("add_45_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(508034944)))];
            tensor<fp16, []> add_45_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_45_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 8, 8]> add_45_cast = batch_norm(beta = add_45_beta_0_to_fp16, epsilon = add_45_epsilon_0_to_fp16, gamma = add_45_gamma_0_to_fp16, mean = add_45_mean_0_to_fp16, variance = add_45_variance_0_to_fp16, x = reshape_89_cast);
            tensor<fp16, [2, 1280, 8, 8]> input_203_cast = silu(x = add_45_cast);
            tensor<int32, [2]> var_4389 = const()[name = tensor<string, []>("op_4389"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4391 = const()[name = tensor<string, []>("op_4391"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_117_pad_type_0 = const()[name = tensor<string, []>("hidden_states_117_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_117_pad_0 = const()[name = tensor<string, []>("hidden_states_117_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> mid_block_resnets_0_conv1_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_resnets_0_conv1_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(508037568)))];
            tensor<fp16, [1280]> mid_block_resnets_0_conv1_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_resnets_0_conv1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(537528832)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_117_cast = conv(bias = mid_block_resnets_0_conv1_bias_to_fp16, dilations = var_4391, groups = var_4371, pad = hidden_states_117_pad_0, pad_type = hidden_states_117_pad_type_0, strides = var_4389, weight = mid_block_resnets_0_conv1_weight_to_fp16, x = input_203_cast);
            tensor<int32, [2]> var_4397 = const()[name = tensor<string, []>("op_4397"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4399 = const()[name = tensor<string, []>("op_4399"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_17_pad_type_0 = const()[name = tensor<string, []>("temb_17_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_17_pad_0 = const()[name = tensor<string, []>("temb_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(537531456)))];
            tensor<fp16, [1280]> mid_block_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(540808320)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_17_cast = conv(bias = mid_block_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_4399, groups = var_4371, pad = temb_17_pad_0, pad_type = temb_17_pad_type_0, strides = var_4397, weight = mid_block_resnets_0_time_emb_proj_weight_to_fp16, x = input_15_cast);
            tensor<fp16, [2, 1280, 8, 8]> input_207_cast = add(x = hidden_states_117_cast, y = temb_17_cast);
            tensor<int32, [5]> reshape_92_shape_0 = const()[name = tensor<string, []>("reshape_92_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 8, 8])];
            tensor<fp16, [2, 32, 40, 8, 8]> reshape_92_cast = reshape(shape = reshape_92_shape_0, x = input_207_cast);
            tensor<int32, [3]> reduce_mean_69_axes_0 = const()[name = tensor<string, []>("reduce_mean_69_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_69_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_69_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_69_cast = reduce_mean(axes = reduce_mean_69_axes_0, keep_dims = reduce_mean_69_keep_dims_0, x = reshape_92_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> sub_46_cast = sub(x = reshape_92_cast, y = reduce_mean_69_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> square_23_cast = square(x = sub_46_cast);
            tensor<int32, [3]> reduce_mean_71_axes_0 = const()[name = tensor<string, []>("reduce_mean_71_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_71_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_71_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_71_cast = reduce_mean(axes = reduce_mean_71_axes_0, keep_dims = reduce_mean_71_keep_dims_0, x = square_23_cast);
            tensor<fp16, []> add_46_y_0_to_fp16 = const()[name = tensor<string, []>("add_46_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_46_cast = add(x = reduce_mean_71_cast, y = add_46_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_23_cast = sqrt(x = add_46_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> real_div_23_cast = real_div(x = sub_46_cast, y = sqrt_23_cast);
            tensor<int32, [4]> reshape_93_shape_0 = const()[name = tensor<string, []>("reshape_93_shape_0"), val = tensor<int32, [4]>([2, 1280, 8, 8])];
            tensor<fp16, [2, 1280, 8, 8]> reshape_93_cast = reshape(shape = reshape_93_shape_0, x = real_div_23_cast);
            tensor<fp16, [1280]> add_47_mean_0_to_fp16 = const()[name = tensor<string, []>("add_47_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(540810944)))];
            tensor<fp16, [1280]> add_47_variance_0_to_fp16 = const()[name = tensor<string, []>("add_47_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(540813568)))];
            tensor<fp16, [1280]> add_47_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_47_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(540816192)))];
            tensor<fp16, [1280]> add_47_beta_0_to_fp16 = const()[name = tensor<string, []>("add_47_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(540818816)))];
            tensor<fp16, []> add_47_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_47_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 8, 8]> add_47_cast = batch_norm(beta = add_47_beta_0_to_fp16, epsilon = add_47_epsilon_0_to_fp16, gamma = add_47_gamma_0_to_fp16, mean = add_47_mean_0_to_fp16, variance = add_47_variance_0_to_fp16, x = reshape_93_cast);
            tensor<fp16, [2, 1280, 8, 8]> input_211_cast = silu(x = add_47_cast);
            tensor<int32, [2]> var_4409 = const()[name = tensor<string, []>("op_4409"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4411 = const()[name = tensor<string, []>("op_4411"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_119_pad_type_0 = const()[name = tensor<string, []>("hidden_states_119_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_119_pad_0 = const()[name = tensor<string, []>("hidden_states_119_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> mid_block_resnets_0_conv2_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_resnets_0_conv2_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(540821440)))];
            tensor<fp16, [1280]> mid_block_resnets_0_conv2_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_resnets_0_conv2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(570312704)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_119_cast = conv(bias = mid_block_resnets_0_conv2_bias_to_fp16, dilations = var_4411, groups = var_4371, pad = hidden_states_119_pad_0, pad_type = hidden_states_119_pad_type_0, strides = var_4409, weight = mid_block_resnets_0_conv2_weight_to_fp16, x = input_211_cast);
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_121_cast = add(x = input_199_cast, y = hidden_states_119_cast);
            tensor<int32, [5]> reshape_96_shape_0 = const()[name = tensor<string, []>("reshape_96_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 8, 8])];
            tensor<fp16, [2, 32, 40, 8, 8]> reshape_96_cast = reshape(shape = reshape_96_shape_0, x = hidden_states_121_cast);
            tensor<int32, [3]> reduce_mean_72_axes_0 = const()[name = tensor<string, []>("reduce_mean_72_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_72_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_72_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_72_cast = reduce_mean(axes = reduce_mean_72_axes_0, keep_dims = reduce_mean_72_keep_dims_0, x = reshape_96_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> sub_48_cast = sub(x = reshape_96_cast, y = reduce_mean_72_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> square_24_cast = square(x = sub_48_cast);
            tensor<int32, [3]> reduce_mean_74_axes_0 = const()[name = tensor<string, []>("reduce_mean_74_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_74_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_74_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_74_cast = reduce_mean(axes = reduce_mean_74_axes_0, keep_dims = reduce_mean_74_keep_dims_0, x = square_24_cast);
            tensor<fp16, []> add_48_y_0_to_fp16 = const()[name = tensor<string, []>("add_48_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_48_cast = add(x = reduce_mean_74_cast, y = add_48_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_24_cast = sqrt(x = add_48_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> real_div_24_cast = real_div(x = sub_48_cast, y = sqrt_24_cast);
            tensor<int32, [4]> reshape_97_shape_0 = const()[name = tensor<string, []>("reshape_97_shape_0"), val = tensor<int32, [4]>([2, 1280, 8, 8])];
            tensor<fp16, [2, 1280, 8, 8]> reshape_97_cast = reshape(shape = reshape_97_shape_0, x = real_div_24_cast);
            tensor<fp16, [1280]> add_49_mean_0_to_fp16 = const()[name = tensor<string, []>("add_49_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(570315328)))];
            tensor<fp16, [1280]> add_49_variance_0_to_fp16 = const()[name = tensor<string, []>("add_49_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(570317952)))];
            tensor<fp16, [1280]> add_49_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_49_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(570320576)))];
            tensor<fp16, [1280]> add_49_beta_0_to_fp16 = const()[name = tensor<string, []>("add_49_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(570323200)))];
            tensor<fp16, []> add_49_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_49_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 8, 8]> add_49_cast = batch_norm(beta = add_49_beta_0_to_fp16, epsilon = add_49_epsilon_0_to_fp16, gamma = add_49_gamma_0_to_fp16, mean = add_49_mean_0_to_fp16, variance = add_49_variance_0_to_fp16, x = reshape_97_cast);
            tensor<int32, [2]> var_4431 = const()[name = tensor<string, []>("op_4431"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4433 = const()[name = tensor<string, []>("op_4433"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_123_pad_type_0 = const()[name = tensor<string, []>("hidden_states_123_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_123_pad_0 = const()[name = tensor<string, []>("hidden_states_123_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_proj_in_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_proj_in_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(570325824)))];
            tensor<fp16, [1280]> mid_block_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_proj_in_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(573602688)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_123_cast = conv(bias = mid_block_attentions_0_proj_in_bias_to_fp16, dilations = var_4433, groups = var_4371, pad = hidden_states_123_pad_0, pad_type = hidden_states_123_pad_type_0, strides = var_4431, weight = mid_block_attentions_0_proj_in_weight_to_fp16, x = add_49_cast);
            tensor<int32, [4]> var_4438 = const()[name = tensor<string, []>("op_4438"), val = tensor<int32, [4]>([2, 1280, 1, 64])];
            tensor<fp16, [2, 1280, 1, 64]> inputs_37_cast = reshape(shape = var_4438, x = hidden_states_123_cast);
            tensor<int32, [1]> var_4448 = const()[name = tensor<string, []>("op_4448"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 64]> channels_mean_37_cast = reduce_mean(axes = var_4448, keep_dims = var_4366, x = inputs_37_cast);
            tensor<fp16, [2, 1280, 1, 64]> zero_mean_37_cast = sub(x = inputs_37_cast, y = channels_mean_37_cast);
            tensor<fp16, [2, 1280, 1, 64]> zero_mean_sq_37_cast = mul(x = zero_mean_37_cast, y = zero_mean_37_cast);
            tensor<int32, [1]> var_4452 = const()[name = tensor<string, []>("op_4452"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 64]> var_4453_cast = reduce_mean(axes = var_4452, keep_dims = var_4366, x = zero_mean_sq_37_cast);
            tensor<fp16, []> var_4454_to_fp16 = const()[name = tensor<string, []>("op_4454_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 64]> var_4455_cast = add(x = var_4453_cast, y = var_4454_to_fp16);
            tensor<fp16, []> denom_37_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_37_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 64]> denom_37_cast = rsqrt(epsilon = denom_37_epsilon_0_to_fp16, x = var_4455_cast);
            tensor<fp16, [2, 1280, 1, 64]> out_37_cast = mul(x = zero_mean_37_cast, y = denom_37_cast);
            tensor<fp16, [1, 1280, 1, 1]> var_4459_to_fp16 = const()[name = tensor<string, []>("op_4459_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(573605312)))];
            tensor<fp16, [2, 1280, 1, 64]> var_4460_cast = add(x = out_37_cast, y = var_4459_to_fp16);
            tensor<fp16, [1, 1280, 1, 1]> var_4462_to_fp16 = const()[name = tensor<string, []>("op_4462_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(573607936)))];
            tensor<fp16, [2, 1280, 1, 64]> hidden_states_125_cast = mul(x = var_4460_cast, y = var_4462_to_fp16);
            tensor<int32, [2]> var_4469 = const()[name = tensor<string, []>("op_4469"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4471 = const()[name = tensor<string, []>("op_4471"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_25_pad_type_0 = const()[name = tensor<string, []>("q_25_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_25_pad_0 = const()[name = tensor<string, []>("q_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(573610560)))];
            tensor<fp16, [2, 1280, 1, 64]> q_25_cast = conv(dilations = var_4471, groups = var_4371, pad = q_25_pad_0, pad_type = q_25_pad_type_0, strides = var_4469, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_125_cast);
            tensor<int32, [2]> var_4475 = const()[name = tensor<string, []>("op_4475"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4477 = const()[name = tensor<string, []>("op_4477"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_49_pad_type_0 = const()[name = tensor<string, []>("k_49_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_49_pad_0 = const()[name = tensor<string, []>("k_49_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(576887424)))];
            tensor<fp16, [2, 1280, 1, 64]> k_49_cast = conv(dilations = var_4477, groups = var_4371, pad = k_49_pad_0, pad_type = k_49_pad_type_0, strides = var_4475, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_125_cast);
            tensor<int32, [2]> var_4481 = const()[name = tensor<string, []>("op_4481"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4483 = const()[name = tensor<string, []>("op_4483"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_25_pad_type_0 = const()[name = tensor<string, []>("v_25_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_25_pad_0 = const()[name = tensor<string, []>("v_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(580164288)))];
            tensor<fp16, [2, 1280, 1, 64]> v_25_cast = conv(dilations = var_4483, groups = var_4371, pad = v_25_pad_0, pad_type = v_25_pad_type_0, strides = var_4481, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_125_cast);
            tensor<int32, [4]> var_4487_begin_0 = const()[name = tensor<string, []>("op_4487_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4487_end_0 = const()[name = tensor<string, []>("op_4487_end_0"), val = tensor<int32, [4]>([2, 64, 1, 64])];
            tensor<bool, [4]> var_4487_end_mask_0 = const()[name = tensor<string, []>("op_4487_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4487_cast = slice_by_index(begin = var_4487_begin_0, end = var_4487_end_0, end_mask = var_4487_end_mask_0, x = q_25_cast);
            tensor<int32, [4]> var_4491_begin_0 = const()[name = tensor<string, []>("op_4491_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_4491_end_0 = const()[name = tensor<string, []>("op_4491_end_0"), val = tensor<int32, [4]>([2, 128, 1, 64])];
            tensor<bool, [4]> var_4491_end_mask_0 = const()[name = tensor<string, []>("op_4491_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4491_cast = slice_by_index(begin = var_4491_begin_0, end = var_4491_end_0, end_mask = var_4491_end_mask_0, x = q_25_cast);
            tensor<int32, [4]> var_4495_begin_0 = const()[name = tensor<string, []>("op_4495_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_4495_end_0 = const()[name = tensor<string, []>("op_4495_end_0"), val = tensor<int32, [4]>([2, 192, 1, 64])];
            tensor<bool, [4]> var_4495_end_mask_0 = const()[name = tensor<string, []>("op_4495_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4495_cast = slice_by_index(begin = var_4495_begin_0, end = var_4495_end_0, end_mask = var_4495_end_mask_0, x = q_25_cast);
            tensor<int32, [4]> var_4499_begin_0 = const()[name = tensor<string, []>("op_4499_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_4499_end_0 = const()[name = tensor<string, []>("op_4499_end_0"), val = tensor<int32, [4]>([2, 256, 1, 64])];
            tensor<bool, [4]> var_4499_end_mask_0 = const()[name = tensor<string, []>("op_4499_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4499_cast = slice_by_index(begin = var_4499_begin_0, end = var_4499_end_0, end_mask = var_4499_end_mask_0, x = q_25_cast);
            tensor<int32, [4]> var_4503_begin_0 = const()[name = tensor<string, []>("op_4503_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_4503_end_0 = const()[name = tensor<string, []>("op_4503_end_0"), val = tensor<int32, [4]>([2, 320, 1, 64])];
            tensor<bool, [4]> var_4503_end_mask_0 = const()[name = tensor<string, []>("op_4503_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4503_cast = slice_by_index(begin = var_4503_begin_0, end = var_4503_end_0, end_mask = var_4503_end_mask_0, x = q_25_cast);
            tensor<int32, [4]> var_4507_begin_0 = const()[name = tensor<string, []>("op_4507_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_4507_end_0 = const()[name = tensor<string, []>("op_4507_end_0"), val = tensor<int32, [4]>([2, 384, 1, 64])];
            tensor<bool, [4]> var_4507_end_mask_0 = const()[name = tensor<string, []>("op_4507_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4507_cast = slice_by_index(begin = var_4507_begin_0, end = var_4507_end_0, end_mask = var_4507_end_mask_0, x = q_25_cast);
            tensor<int32, [4]> var_4511_begin_0 = const()[name = tensor<string, []>("op_4511_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_4511_end_0 = const()[name = tensor<string, []>("op_4511_end_0"), val = tensor<int32, [4]>([2, 448, 1, 64])];
            tensor<bool, [4]> var_4511_end_mask_0 = const()[name = tensor<string, []>("op_4511_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4511_cast = slice_by_index(begin = var_4511_begin_0, end = var_4511_end_0, end_mask = var_4511_end_mask_0, x = q_25_cast);
            tensor<int32, [4]> var_4515_begin_0 = const()[name = tensor<string, []>("op_4515_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_4515_end_0 = const()[name = tensor<string, []>("op_4515_end_0"), val = tensor<int32, [4]>([2, 512, 1, 64])];
            tensor<bool, [4]> var_4515_end_mask_0 = const()[name = tensor<string, []>("op_4515_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4515_cast = slice_by_index(begin = var_4515_begin_0, end = var_4515_end_0, end_mask = var_4515_end_mask_0, x = q_25_cast);
            tensor<int32, [4]> var_4519_begin_0 = const()[name = tensor<string, []>("op_4519_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_4519_end_0 = const()[name = tensor<string, []>("op_4519_end_0"), val = tensor<int32, [4]>([2, 576, 1, 64])];
            tensor<bool, [4]> var_4519_end_mask_0 = const()[name = tensor<string, []>("op_4519_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4519_cast = slice_by_index(begin = var_4519_begin_0, end = var_4519_end_0, end_mask = var_4519_end_mask_0, x = q_25_cast);
            tensor<int32, [4]> var_4523_begin_0 = const()[name = tensor<string, []>("op_4523_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_4523_end_0 = const()[name = tensor<string, []>("op_4523_end_0"), val = tensor<int32, [4]>([2, 640, 1, 64])];
            tensor<bool, [4]> var_4523_end_mask_0 = const()[name = tensor<string, []>("op_4523_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4523_cast = slice_by_index(begin = var_4523_begin_0, end = var_4523_end_0, end_mask = var_4523_end_mask_0, x = q_25_cast);
            tensor<int32, [4]> var_4527_begin_0 = const()[name = tensor<string, []>("op_4527_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_4527_end_0 = const()[name = tensor<string, []>("op_4527_end_0"), val = tensor<int32, [4]>([2, 704, 1, 64])];
            tensor<bool, [4]> var_4527_end_mask_0 = const()[name = tensor<string, []>("op_4527_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4527_cast = slice_by_index(begin = var_4527_begin_0, end = var_4527_end_0, end_mask = var_4527_end_mask_0, x = q_25_cast);
            tensor<int32, [4]> var_4531_begin_0 = const()[name = tensor<string, []>("op_4531_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_4531_end_0 = const()[name = tensor<string, []>("op_4531_end_0"), val = tensor<int32, [4]>([2, 768, 1, 64])];
            tensor<bool, [4]> var_4531_end_mask_0 = const()[name = tensor<string, []>("op_4531_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4531_cast = slice_by_index(begin = var_4531_begin_0, end = var_4531_end_0, end_mask = var_4531_end_mask_0, x = q_25_cast);
            tensor<int32, [4]> var_4535_begin_0 = const()[name = tensor<string, []>("op_4535_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_4535_end_0 = const()[name = tensor<string, []>("op_4535_end_0"), val = tensor<int32, [4]>([2, 832, 1, 64])];
            tensor<bool, [4]> var_4535_end_mask_0 = const()[name = tensor<string, []>("op_4535_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4535_cast = slice_by_index(begin = var_4535_begin_0, end = var_4535_end_0, end_mask = var_4535_end_mask_0, x = q_25_cast);
            tensor<int32, [4]> var_4539_begin_0 = const()[name = tensor<string, []>("op_4539_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_4539_end_0 = const()[name = tensor<string, []>("op_4539_end_0"), val = tensor<int32, [4]>([2, 896, 1, 64])];
            tensor<bool, [4]> var_4539_end_mask_0 = const()[name = tensor<string, []>("op_4539_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4539_cast = slice_by_index(begin = var_4539_begin_0, end = var_4539_end_0, end_mask = var_4539_end_mask_0, x = q_25_cast);
            tensor<int32, [4]> var_4543_begin_0 = const()[name = tensor<string, []>("op_4543_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_4543_end_0 = const()[name = tensor<string, []>("op_4543_end_0"), val = tensor<int32, [4]>([2, 960, 1, 64])];
            tensor<bool, [4]> var_4543_end_mask_0 = const()[name = tensor<string, []>("op_4543_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4543_cast = slice_by_index(begin = var_4543_begin_0, end = var_4543_end_0, end_mask = var_4543_end_mask_0, x = q_25_cast);
            tensor<int32, [4]> var_4547_begin_0 = const()[name = tensor<string, []>("op_4547_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_4547_end_0 = const()[name = tensor<string, []>("op_4547_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_4547_end_mask_0 = const()[name = tensor<string, []>("op_4547_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4547_cast = slice_by_index(begin = var_4547_begin_0, end = var_4547_end_0, end_mask = var_4547_end_mask_0, x = q_25_cast);
            tensor<int32, [4]> var_4551_begin_0 = const()[name = tensor<string, []>("op_4551_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_4551_end_0 = const()[name = tensor<string, []>("op_4551_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 64])];
            tensor<bool, [4]> var_4551_end_mask_0 = const()[name = tensor<string, []>("op_4551_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4551_cast = slice_by_index(begin = var_4551_begin_0, end = var_4551_end_0, end_mask = var_4551_end_mask_0, x = q_25_cast);
            tensor<int32, [4]> var_4555_begin_0 = const()[name = tensor<string, []>("op_4555_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_4555_end_0 = const()[name = tensor<string, []>("op_4555_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 64])];
            tensor<bool, [4]> var_4555_end_mask_0 = const()[name = tensor<string, []>("op_4555_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4555_cast = slice_by_index(begin = var_4555_begin_0, end = var_4555_end_0, end_mask = var_4555_end_mask_0, x = q_25_cast);
            tensor<int32, [4]> var_4559_begin_0 = const()[name = tensor<string, []>("op_4559_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_4559_end_0 = const()[name = tensor<string, []>("op_4559_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 64])];
            tensor<bool, [4]> var_4559_end_mask_0 = const()[name = tensor<string, []>("op_4559_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4559_cast = slice_by_index(begin = var_4559_begin_0, end = var_4559_end_0, end_mask = var_4559_end_mask_0, x = q_25_cast);
            tensor<int32, [4]> var_4563_begin_0 = const()[name = tensor<string, []>("op_4563_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_4563_end_0 = const()[name = tensor<string, []>("op_4563_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 64])];
            tensor<bool, [4]> var_4563_end_mask_0 = const()[name = tensor<string, []>("op_4563_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4563_cast = slice_by_index(begin = var_4563_begin_0, end = var_4563_end_0, end_mask = var_4563_end_mask_0, x = q_25_cast);
            tensor<int32, [4]> k_51_perm_0 = const()[name = tensor<string, []>("k_51_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_4570_begin_0 = const()[name = tensor<string, []>("op_4570_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4570_end_0 = const()[name = tensor<string, []>("op_4570_end_0"), val = tensor<int32, [4]>([2, 64, 1, 64])];
            tensor<bool, [4]> var_4570_end_mask_0 = const()[name = tensor<string, []>("op_4570_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 1280]> transpose_19 = transpose(perm = k_51_perm_0, x = k_49_cast);
            tensor<fp16, [2, 64, 1, 64]> var_4570_cast = slice_by_index(begin = var_4570_begin_0, end = var_4570_end_0, end_mask = var_4570_end_mask_0, x = transpose_19);
            tensor<int32, [4]> var_4574_begin_0 = const()[name = tensor<string, []>("op_4574_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_4574_end_0 = const()[name = tensor<string, []>("op_4574_end_0"), val = tensor<int32, [4]>([2, 64, 1, 128])];
            tensor<bool, [4]> var_4574_end_mask_0 = const()[name = tensor<string, []>("op_4574_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_4574_cast = slice_by_index(begin = var_4574_begin_0, end = var_4574_end_0, end_mask = var_4574_end_mask_0, x = transpose_19);
            tensor<int32, [4]> var_4578_begin_0 = const()[name = tensor<string, []>("op_4578_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_4578_end_0 = const()[name = tensor<string, []>("op_4578_end_0"), val = tensor<int32, [4]>([2, 64, 1, 192])];
            tensor<bool, [4]> var_4578_end_mask_0 = const()[name = tensor<string, []>("op_4578_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_4578_cast = slice_by_index(begin = var_4578_begin_0, end = var_4578_end_0, end_mask = var_4578_end_mask_0, x = transpose_19);
            tensor<int32, [4]> var_4582_begin_0 = const()[name = tensor<string, []>("op_4582_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_4582_end_0 = const()[name = tensor<string, []>("op_4582_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_4582_end_mask_0 = const()[name = tensor<string, []>("op_4582_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_4582_cast = slice_by_index(begin = var_4582_begin_0, end = var_4582_end_0, end_mask = var_4582_end_mask_0, x = transpose_19);
            tensor<int32, [4]> var_4586_begin_0 = const()[name = tensor<string, []>("op_4586_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_4586_end_0 = const()[name = tensor<string, []>("op_4586_end_0"), val = tensor<int32, [4]>([2, 64, 1, 320])];
            tensor<bool, [4]> var_4586_end_mask_0 = const()[name = tensor<string, []>("op_4586_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_4586_cast = slice_by_index(begin = var_4586_begin_0, end = var_4586_end_0, end_mask = var_4586_end_mask_0, x = transpose_19);
            tensor<int32, [4]> var_4590_begin_0 = const()[name = tensor<string, []>("op_4590_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_4590_end_0 = const()[name = tensor<string, []>("op_4590_end_0"), val = tensor<int32, [4]>([2, 64, 1, 384])];
            tensor<bool, [4]> var_4590_end_mask_0 = const()[name = tensor<string, []>("op_4590_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_4590_cast = slice_by_index(begin = var_4590_begin_0, end = var_4590_end_0, end_mask = var_4590_end_mask_0, x = transpose_19);
            tensor<int32, [4]> var_4594_begin_0 = const()[name = tensor<string, []>("op_4594_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_4594_end_0 = const()[name = tensor<string, []>("op_4594_end_0"), val = tensor<int32, [4]>([2, 64, 1, 448])];
            tensor<bool, [4]> var_4594_end_mask_0 = const()[name = tensor<string, []>("op_4594_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_4594_cast = slice_by_index(begin = var_4594_begin_0, end = var_4594_end_0, end_mask = var_4594_end_mask_0, x = transpose_19);
            tensor<int32, [4]> var_4598_begin_0 = const()[name = tensor<string, []>("op_4598_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_4598_end_0 = const()[name = tensor<string, []>("op_4598_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_4598_end_mask_0 = const()[name = tensor<string, []>("op_4598_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_4598_cast = slice_by_index(begin = var_4598_begin_0, end = var_4598_end_0, end_mask = var_4598_end_mask_0, x = transpose_19);
            tensor<int32, [4]> var_4602_begin_0 = const()[name = tensor<string, []>("op_4602_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_4602_end_0 = const()[name = tensor<string, []>("op_4602_end_0"), val = tensor<int32, [4]>([2, 64, 1, 576])];
            tensor<bool, [4]> var_4602_end_mask_0 = const()[name = tensor<string, []>("op_4602_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_4602_cast = slice_by_index(begin = var_4602_begin_0, end = var_4602_end_0, end_mask = var_4602_end_mask_0, x = transpose_19);
            tensor<int32, [4]> var_4606_begin_0 = const()[name = tensor<string, []>("op_4606_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_4606_end_0 = const()[name = tensor<string, []>("op_4606_end_0"), val = tensor<int32, [4]>([2, 64, 1, 640])];
            tensor<bool, [4]> var_4606_end_mask_0 = const()[name = tensor<string, []>("op_4606_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_4606_cast = slice_by_index(begin = var_4606_begin_0, end = var_4606_end_0, end_mask = var_4606_end_mask_0, x = transpose_19);
            tensor<int32, [4]> var_4610_begin_0 = const()[name = tensor<string, []>("op_4610_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_4610_end_0 = const()[name = tensor<string, []>("op_4610_end_0"), val = tensor<int32, [4]>([2, 64, 1, 704])];
            tensor<bool, [4]> var_4610_end_mask_0 = const()[name = tensor<string, []>("op_4610_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_4610_cast = slice_by_index(begin = var_4610_begin_0, end = var_4610_end_0, end_mask = var_4610_end_mask_0, x = transpose_19);
            tensor<int32, [4]> var_4614_begin_0 = const()[name = tensor<string, []>("op_4614_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_4614_end_0 = const()[name = tensor<string, []>("op_4614_end_0"), val = tensor<int32, [4]>([2, 64, 1, 768])];
            tensor<bool, [4]> var_4614_end_mask_0 = const()[name = tensor<string, []>("op_4614_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_4614_cast = slice_by_index(begin = var_4614_begin_0, end = var_4614_end_0, end_mask = var_4614_end_mask_0, x = transpose_19);
            tensor<int32, [4]> var_4618_begin_0 = const()[name = tensor<string, []>("op_4618_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_4618_end_0 = const()[name = tensor<string, []>("op_4618_end_0"), val = tensor<int32, [4]>([2, 64, 1, 832])];
            tensor<bool, [4]> var_4618_end_mask_0 = const()[name = tensor<string, []>("op_4618_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_4618_cast = slice_by_index(begin = var_4618_begin_0, end = var_4618_end_0, end_mask = var_4618_end_mask_0, x = transpose_19);
            tensor<int32, [4]> var_4622_begin_0 = const()[name = tensor<string, []>("op_4622_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_4622_end_0 = const()[name = tensor<string, []>("op_4622_end_0"), val = tensor<int32, [4]>([2, 64, 1, 896])];
            tensor<bool, [4]> var_4622_end_mask_0 = const()[name = tensor<string, []>("op_4622_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_4622_cast = slice_by_index(begin = var_4622_begin_0, end = var_4622_end_0, end_mask = var_4622_end_mask_0, x = transpose_19);
            tensor<int32, [4]> var_4626_begin_0 = const()[name = tensor<string, []>("op_4626_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_4626_end_0 = const()[name = tensor<string, []>("op_4626_end_0"), val = tensor<int32, [4]>([2, 64, 1, 960])];
            tensor<bool, [4]> var_4626_end_mask_0 = const()[name = tensor<string, []>("op_4626_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_4626_cast = slice_by_index(begin = var_4626_begin_0, end = var_4626_end_0, end_mask = var_4626_end_mask_0, x = transpose_19);
            tensor<int32, [4]> var_4630_begin_0 = const()[name = tensor<string, []>("op_4630_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_4630_end_0 = const()[name = tensor<string, []>("op_4630_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_4630_end_mask_0 = const()[name = tensor<string, []>("op_4630_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_4630_cast = slice_by_index(begin = var_4630_begin_0, end = var_4630_end_0, end_mask = var_4630_end_mask_0, x = transpose_19);
            tensor<int32, [4]> var_4634_begin_0 = const()[name = tensor<string, []>("op_4634_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_4634_end_0 = const()[name = tensor<string, []>("op_4634_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1088])];
            tensor<bool, [4]> var_4634_end_mask_0 = const()[name = tensor<string, []>("op_4634_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_4634_cast = slice_by_index(begin = var_4634_begin_0, end = var_4634_end_0, end_mask = var_4634_end_mask_0, x = transpose_19);
            tensor<int32, [4]> var_4638_begin_0 = const()[name = tensor<string, []>("op_4638_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_4638_end_0 = const()[name = tensor<string, []>("op_4638_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1152])];
            tensor<bool, [4]> var_4638_end_mask_0 = const()[name = tensor<string, []>("op_4638_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_4638_cast = slice_by_index(begin = var_4638_begin_0, end = var_4638_end_0, end_mask = var_4638_end_mask_0, x = transpose_19);
            tensor<int32, [4]> var_4642_begin_0 = const()[name = tensor<string, []>("op_4642_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_4642_end_0 = const()[name = tensor<string, []>("op_4642_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1216])];
            tensor<bool, [4]> var_4642_end_mask_0 = const()[name = tensor<string, []>("op_4642_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_4642_cast = slice_by_index(begin = var_4642_begin_0, end = var_4642_end_0, end_mask = var_4642_end_mask_0, x = transpose_19);
            tensor<int32, [4]> var_4646_begin_0 = const()[name = tensor<string, []>("op_4646_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_4646_end_0 = const()[name = tensor<string, []>("op_4646_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1280])];
            tensor<bool, [4]> var_4646_end_mask_0 = const()[name = tensor<string, []>("op_4646_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_4646_cast = slice_by_index(begin = var_4646_begin_0, end = var_4646_end_0, end_mask = var_4646_end_mask_0, x = transpose_19);
            tensor<int32, [4]> var_4648_begin_0 = const()[name = tensor<string, []>("op_4648_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4648_end_0 = const()[name = tensor<string, []>("op_4648_end_0"), val = tensor<int32, [4]>([2, 64, 1, 64])];
            tensor<bool, [4]> var_4648_end_mask_0 = const()[name = tensor<string, []>("op_4648_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4648_cast = slice_by_index(begin = var_4648_begin_0, end = var_4648_end_0, end_mask = var_4648_end_mask_0, x = v_25_cast);
            tensor<int32, [4]> var_4652_begin_0 = const()[name = tensor<string, []>("op_4652_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_4652_end_0 = const()[name = tensor<string, []>("op_4652_end_0"), val = tensor<int32, [4]>([2, 128, 1, 64])];
            tensor<bool, [4]> var_4652_end_mask_0 = const()[name = tensor<string, []>("op_4652_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4652_cast = slice_by_index(begin = var_4652_begin_0, end = var_4652_end_0, end_mask = var_4652_end_mask_0, x = v_25_cast);
            tensor<int32, [4]> var_4656_begin_0 = const()[name = tensor<string, []>("op_4656_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_4656_end_0 = const()[name = tensor<string, []>("op_4656_end_0"), val = tensor<int32, [4]>([2, 192, 1, 64])];
            tensor<bool, [4]> var_4656_end_mask_0 = const()[name = tensor<string, []>("op_4656_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4656_cast = slice_by_index(begin = var_4656_begin_0, end = var_4656_end_0, end_mask = var_4656_end_mask_0, x = v_25_cast);
            tensor<int32, [4]> var_4660_begin_0 = const()[name = tensor<string, []>("op_4660_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_4660_end_0 = const()[name = tensor<string, []>("op_4660_end_0"), val = tensor<int32, [4]>([2, 256, 1, 64])];
            tensor<bool, [4]> var_4660_end_mask_0 = const()[name = tensor<string, []>("op_4660_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4660_cast = slice_by_index(begin = var_4660_begin_0, end = var_4660_end_0, end_mask = var_4660_end_mask_0, x = v_25_cast);
            tensor<int32, [4]> var_4664_begin_0 = const()[name = tensor<string, []>("op_4664_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_4664_end_0 = const()[name = tensor<string, []>("op_4664_end_0"), val = tensor<int32, [4]>([2, 320, 1, 64])];
            tensor<bool, [4]> var_4664_end_mask_0 = const()[name = tensor<string, []>("op_4664_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4664_cast = slice_by_index(begin = var_4664_begin_0, end = var_4664_end_0, end_mask = var_4664_end_mask_0, x = v_25_cast);
            tensor<int32, [4]> var_4668_begin_0 = const()[name = tensor<string, []>("op_4668_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_4668_end_0 = const()[name = tensor<string, []>("op_4668_end_0"), val = tensor<int32, [4]>([2, 384, 1, 64])];
            tensor<bool, [4]> var_4668_end_mask_0 = const()[name = tensor<string, []>("op_4668_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4668_cast = slice_by_index(begin = var_4668_begin_0, end = var_4668_end_0, end_mask = var_4668_end_mask_0, x = v_25_cast);
            tensor<int32, [4]> var_4672_begin_0 = const()[name = tensor<string, []>("op_4672_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_4672_end_0 = const()[name = tensor<string, []>("op_4672_end_0"), val = tensor<int32, [4]>([2, 448, 1, 64])];
            tensor<bool, [4]> var_4672_end_mask_0 = const()[name = tensor<string, []>("op_4672_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4672_cast = slice_by_index(begin = var_4672_begin_0, end = var_4672_end_0, end_mask = var_4672_end_mask_0, x = v_25_cast);
            tensor<int32, [4]> var_4676_begin_0 = const()[name = tensor<string, []>("op_4676_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_4676_end_0 = const()[name = tensor<string, []>("op_4676_end_0"), val = tensor<int32, [4]>([2, 512, 1, 64])];
            tensor<bool, [4]> var_4676_end_mask_0 = const()[name = tensor<string, []>("op_4676_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4676_cast = slice_by_index(begin = var_4676_begin_0, end = var_4676_end_0, end_mask = var_4676_end_mask_0, x = v_25_cast);
            tensor<int32, [4]> var_4680_begin_0 = const()[name = tensor<string, []>("op_4680_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_4680_end_0 = const()[name = tensor<string, []>("op_4680_end_0"), val = tensor<int32, [4]>([2, 576, 1, 64])];
            tensor<bool, [4]> var_4680_end_mask_0 = const()[name = tensor<string, []>("op_4680_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4680_cast = slice_by_index(begin = var_4680_begin_0, end = var_4680_end_0, end_mask = var_4680_end_mask_0, x = v_25_cast);
            tensor<int32, [4]> var_4684_begin_0 = const()[name = tensor<string, []>("op_4684_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_4684_end_0 = const()[name = tensor<string, []>("op_4684_end_0"), val = tensor<int32, [4]>([2, 640, 1, 64])];
            tensor<bool, [4]> var_4684_end_mask_0 = const()[name = tensor<string, []>("op_4684_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4684_cast = slice_by_index(begin = var_4684_begin_0, end = var_4684_end_0, end_mask = var_4684_end_mask_0, x = v_25_cast);
            tensor<int32, [4]> var_4688_begin_0 = const()[name = tensor<string, []>("op_4688_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_4688_end_0 = const()[name = tensor<string, []>("op_4688_end_0"), val = tensor<int32, [4]>([2, 704, 1, 64])];
            tensor<bool, [4]> var_4688_end_mask_0 = const()[name = tensor<string, []>("op_4688_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4688_cast = slice_by_index(begin = var_4688_begin_0, end = var_4688_end_0, end_mask = var_4688_end_mask_0, x = v_25_cast);
            tensor<int32, [4]> var_4692_begin_0 = const()[name = tensor<string, []>("op_4692_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_4692_end_0 = const()[name = tensor<string, []>("op_4692_end_0"), val = tensor<int32, [4]>([2, 768, 1, 64])];
            tensor<bool, [4]> var_4692_end_mask_0 = const()[name = tensor<string, []>("op_4692_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4692_cast = slice_by_index(begin = var_4692_begin_0, end = var_4692_end_0, end_mask = var_4692_end_mask_0, x = v_25_cast);
            tensor<int32, [4]> var_4696_begin_0 = const()[name = tensor<string, []>("op_4696_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_4696_end_0 = const()[name = tensor<string, []>("op_4696_end_0"), val = tensor<int32, [4]>([2, 832, 1, 64])];
            tensor<bool, [4]> var_4696_end_mask_0 = const()[name = tensor<string, []>("op_4696_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4696_cast = slice_by_index(begin = var_4696_begin_0, end = var_4696_end_0, end_mask = var_4696_end_mask_0, x = v_25_cast);
            tensor<int32, [4]> var_4700_begin_0 = const()[name = tensor<string, []>("op_4700_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_4700_end_0 = const()[name = tensor<string, []>("op_4700_end_0"), val = tensor<int32, [4]>([2, 896, 1, 64])];
            tensor<bool, [4]> var_4700_end_mask_0 = const()[name = tensor<string, []>("op_4700_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4700_cast = slice_by_index(begin = var_4700_begin_0, end = var_4700_end_0, end_mask = var_4700_end_mask_0, x = v_25_cast);
            tensor<int32, [4]> var_4704_begin_0 = const()[name = tensor<string, []>("op_4704_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_4704_end_0 = const()[name = tensor<string, []>("op_4704_end_0"), val = tensor<int32, [4]>([2, 960, 1, 64])];
            tensor<bool, [4]> var_4704_end_mask_0 = const()[name = tensor<string, []>("op_4704_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4704_cast = slice_by_index(begin = var_4704_begin_0, end = var_4704_end_0, end_mask = var_4704_end_mask_0, x = v_25_cast);
            tensor<int32, [4]> var_4708_begin_0 = const()[name = tensor<string, []>("op_4708_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_4708_end_0 = const()[name = tensor<string, []>("op_4708_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_4708_end_mask_0 = const()[name = tensor<string, []>("op_4708_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4708_cast = slice_by_index(begin = var_4708_begin_0, end = var_4708_end_0, end_mask = var_4708_end_mask_0, x = v_25_cast);
            tensor<int32, [4]> var_4712_begin_0 = const()[name = tensor<string, []>("op_4712_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_4712_end_0 = const()[name = tensor<string, []>("op_4712_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 64])];
            tensor<bool, [4]> var_4712_end_mask_0 = const()[name = tensor<string, []>("op_4712_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4712_cast = slice_by_index(begin = var_4712_begin_0, end = var_4712_end_0, end_mask = var_4712_end_mask_0, x = v_25_cast);
            tensor<int32, [4]> var_4716_begin_0 = const()[name = tensor<string, []>("op_4716_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_4716_end_0 = const()[name = tensor<string, []>("op_4716_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 64])];
            tensor<bool, [4]> var_4716_end_mask_0 = const()[name = tensor<string, []>("op_4716_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4716_cast = slice_by_index(begin = var_4716_begin_0, end = var_4716_end_0, end_mask = var_4716_end_mask_0, x = v_25_cast);
            tensor<int32, [4]> var_4720_begin_0 = const()[name = tensor<string, []>("op_4720_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_4720_end_0 = const()[name = tensor<string, []>("op_4720_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 64])];
            tensor<bool, [4]> var_4720_end_mask_0 = const()[name = tensor<string, []>("op_4720_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4720_cast = slice_by_index(begin = var_4720_begin_0, end = var_4720_end_0, end_mask = var_4720_end_mask_0, x = v_25_cast);
            tensor<int32, [4]> var_4724_begin_0 = const()[name = tensor<string, []>("op_4724_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_4724_end_0 = const()[name = tensor<string, []>("op_4724_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 64])];
            tensor<bool, [4]> var_4724_end_mask_0 = const()[name = tensor<string, []>("op_4724_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4724_cast = slice_by_index(begin = var_4724_begin_0, end = var_4724_end_0, end_mask = var_4724_end_mask_0, x = v_25_cast);
            tensor<string, []> var_4728_equation_0 = const()[name = tensor<string, []>("op_4728_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_4728_cast = einsum(equation = var_4728_equation_0, values = (var_4570_cast, var_4487_cast));
            tensor<fp16, []> var_4729_to_fp16 = const()[name = tensor<string, []>("op_4729_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_281_cast = mul(x = var_4728_cast, y = var_4729_to_fp16);
            tensor<string, []> var_4732_equation_0 = const()[name = tensor<string, []>("op_4732_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_4732_cast = einsum(equation = var_4732_equation_0, values = (var_4574_cast, var_4491_cast));
            tensor<fp16, []> var_4733_to_fp16 = const()[name = tensor<string, []>("op_4733_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_283_cast = mul(x = var_4732_cast, y = var_4733_to_fp16);
            tensor<string, []> var_4736_equation_0 = const()[name = tensor<string, []>("op_4736_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_4736_cast = einsum(equation = var_4736_equation_0, values = (var_4578_cast, var_4495_cast));
            tensor<fp16, []> var_4737_to_fp16 = const()[name = tensor<string, []>("op_4737_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_285_cast = mul(x = var_4736_cast, y = var_4737_to_fp16);
            tensor<string, []> var_4740_equation_0 = const()[name = tensor<string, []>("op_4740_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_4740_cast = einsum(equation = var_4740_equation_0, values = (var_4582_cast, var_4499_cast));
            tensor<fp16, []> var_4741_to_fp16 = const()[name = tensor<string, []>("op_4741_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_287_cast = mul(x = var_4740_cast, y = var_4741_to_fp16);
            tensor<string, []> var_4744_equation_0 = const()[name = tensor<string, []>("op_4744_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_4744_cast = einsum(equation = var_4744_equation_0, values = (var_4586_cast, var_4503_cast));
            tensor<fp16, []> var_4745_to_fp16 = const()[name = tensor<string, []>("op_4745_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_289_cast = mul(x = var_4744_cast, y = var_4745_to_fp16);
            tensor<string, []> var_4748_equation_0 = const()[name = tensor<string, []>("op_4748_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_4748_cast = einsum(equation = var_4748_equation_0, values = (var_4590_cast, var_4507_cast));
            tensor<fp16, []> var_4749_to_fp16 = const()[name = tensor<string, []>("op_4749_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_291_cast = mul(x = var_4748_cast, y = var_4749_to_fp16);
            tensor<string, []> var_4752_equation_0 = const()[name = tensor<string, []>("op_4752_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_4752_cast = einsum(equation = var_4752_equation_0, values = (var_4594_cast, var_4511_cast));
            tensor<fp16, []> var_4753_to_fp16 = const()[name = tensor<string, []>("op_4753_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_293_cast = mul(x = var_4752_cast, y = var_4753_to_fp16);
            tensor<string, []> var_4756_equation_0 = const()[name = tensor<string, []>("op_4756_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_4756_cast = einsum(equation = var_4756_equation_0, values = (var_4598_cast, var_4515_cast));
            tensor<fp16, []> var_4757_to_fp16 = const()[name = tensor<string, []>("op_4757_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_295_cast = mul(x = var_4756_cast, y = var_4757_to_fp16);
            tensor<string, []> var_4760_equation_0 = const()[name = tensor<string, []>("op_4760_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_4760_cast = einsum(equation = var_4760_equation_0, values = (var_4602_cast, var_4519_cast));
            tensor<fp16, []> var_4761_to_fp16 = const()[name = tensor<string, []>("op_4761_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_297_cast = mul(x = var_4760_cast, y = var_4761_to_fp16);
            tensor<string, []> var_4764_equation_0 = const()[name = tensor<string, []>("op_4764_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_4764_cast = einsum(equation = var_4764_equation_0, values = (var_4606_cast, var_4523_cast));
            tensor<fp16, []> var_4765_to_fp16 = const()[name = tensor<string, []>("op_4765_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_299_cast = mul(x = var_4764_cast, y = var_4765_to_fp16);
            tensor<string, []> var_4768_equation_0 = const()[name = tensor<string, []>("op_4768_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_4768_cast = einsum(equation = var_4768_equation_0, values = (var_4610_cast, var_4527_cast));
            tensor<fp16, []> var_4769_to_fp16 = const()[name = tensor<string, []>("op_4769_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_301_cast = mul(x = var_4768_cast, y = var_4769_to_fp16);
            tensor<string, []> var_4772_equation_0 = const()[name = tensor<string, []>("op_4772_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_4772_cast = einsum(equation = var_4772_equation_0, values = (var_4614_cast, var_4531_cast));
            tensor<fp16, []> var_4773_to_fp16 = const()[name = tensor<string, []>("op_4773_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_303_cast = mul(x = var_4772_cast, y = var_4773_to_fp16);
            tensor<string, []> var_4776_equation_0 = const()[name = tensor<string, []>("op_4776_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_4776_cast = einsum(equation = var_4776_equation_0, values = (var_4618_cast, var_4535_cast));
            tensor<fp16, []> var_4777_to_fp16 = const()[name = tensor<string, []>("op_4777_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_305_cast = mul(x = var_4776_cast, y = var_4777_to_fp16);
            tensor<string, []> var_4780_equation_0 = const()[name = tensor<string, []>("op_4780_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_4780_cast = einsum(equation = var_4780_equation_0, values = (var_4622_cast, var_4539_cast));
            tensor<fp16, []> var_4781_to_fp16 = const()[name = tensor<string, []>("op_4781_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_307_cast = mul(x = var_4780_cast, y = var_4781_to_fp16);
            tensor<string, []> var_4784_equation_0 = const()[name = tensor<string, []>("op_4784_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_4784_cast = einsum(equation = var_4784_equation_0, values = (var_4626_cast, var_4543_cast));
            tensor<fp16, []> var_4785_to_fp16 = const()[name = tensor<string, []>("op_4785_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_309_cast = mul(x = var_4784_cast, y = var_4785_to_fp16);
            tensor<string, []> var_4788_equation_0 = const()[name = tensor<string, []>("op_4788_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_4788_cast = einsum(equation = var_4788_equation_0, values = (var_4630_cast, var_4547_cast));
            tensor<fp16, []> var_4789_to_fp16 = const()[name = tensor<string, []>("op_4789_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_311_cast = mul(x = var_4788_cast, y = var_4789_to_fp16);
            tensor<string, []> var_4792_equation_0 = const()[name = tensor<string, []>("op_4792_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_4792_cast = einsum(equation = var_4792_equation_0, values = (var_4634_cast, var_4551_cast));
            tensor<fp16, []> var_4793_to_fp16 = const()[name = tensor<string, []>("op_4793_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_313_cast = mul(x = var_4792_cast, y = var_4793_to_fp16);
            tensor<string, []> var_4796_equation_0 = const()[name = tensor<string, []>("op_4796_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_4796_cast = einsum(equation = var_4796_equation_0, values = (var_4638_cast, var_4555_cast));
            tensor<fp16, []> var_4797_to_fp16 = const()[name = tensor<string, []>("op_4797_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_315_cast = mul(x = var_4796_cast, y = var_4797_to_fp16);
            tensor<string, []> var_4800_equation_0 = const()[name = tensor<string, []>("op_4800_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_4800_cast = einsum(equation = var_4800_equation_0, values = (var_4642_cast, var_4559_cast));
            tensor<fp16, []> var_4801_to_fp16 = const()[name = tensor<string, []>("op_4801_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_317_cast = mul(x = var_4800_cast, y = var_4801_to_fp16);
            tensor<string, []> var_4804_equation_0 = const()[name = tensor<string, []>("op_4804_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_4804_cast = einsum(equation = var_4804_equation_0, values = (var_4646_cast, var_4563_cast));
            tensor<fp16, []> var_4805_to_fp16 = const()[name = tensor<string, []>("op_4805_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_319_cast = mul(x = var_4804_cast, y = var_4805_to_fp16);
            tensor<fp16, [2, 64, 1, 64]> var_4807_cast = softmax(axis = var_4371, x = aw_281_cast);
            tensor<fp16, [2, 64, 1, 64]> var_4808_cast = softmax(axis = var_4371, x = aw_283_cast);
            tensor<fp16, [2, 64, 1, 64]> var_4809_cast = softmax(axis = var_4371, x = aw_285_cast);
            tensor<fp16, [2, 64, 1, 64]> var_4810_cast = softmax(axis = var_4371, x = aw_287_cast);
            tensor<fp16, [2, 64, 1, 64]> var_4811_cast = softmax(axis = var_4371, x = aw_289_cast);
            tensor<fp16, [2, 64, 1, 64]> var_4812_cast = softmax(axis = var_4371, x = aw_291_cast);
            tensor<fp16, [2, 64, 1, 64]> var_4813_cast = softmax(axis = var_4371, x = aw_293_cast);
            tensor<fp16, [2, 64, 1, 64]> var_4814_cast = softmax(axis = var_4371, x = aw_295_cast);
            tensor<fp16, [2, 64, 1, 64]> var_4815_cast = softmax(axis = var_4371, x = aw_297_cast);
            tensor<fp16, [2, 64, 1, 64]> var_4816_cast = softmax(axis = var_4371, x = aw_299_cast);
            tensor<fp16, [2, 64, 1, 64]> var_4817_cast = softmax(axis = var_4371, x = aw_301_cast);
            tensor<fp16, [2, 64, 1, 64]> var_4818_cast = softmax(axis = var_4371, x = aw_303_cast);
            tensor<fp16, [2, 64, 1, 64]> var_4819_cast = softmax(axis = var_4371, x = aw_305_cast);
            tensor<fp16, [2, 64, 1, 64]> var_4820_cast = softmax(axis = var_4371, x = aw_307_cast);
            tensor<fp16, [2, 64, 1, 64]> var_4821_cast = softmax(axis = var_4371, x = aw_309_cast);
            tensor<fp16, [2, 64, 1, 64]> var_4822_cast = softmax(axis = var_4371, x = aw_311_cast);
            tensor<fp16, [2, 64, 1, 64]> var_4823_cast = softmax(axis = var_4371, x = aw_313_cast);
            tensor<fp16, [2, 64, 1, 64]> var_4824_cast = softmax(axis = var_4371, x = aw_315_cast);
            tensor<fp16, [2, 64, 1, 64]> var_4825_cast = softmax(axis = var_4371, x = aw_317_cast);
            tensor<fp16, [2, 64, 1, 64]> var_4826_cast = softmax(axis = var_4371, x = aw_319_cast);
            tensor<string, []> var_4828_equation_0 = const()[name = tensor<string, []>("op_4828_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_4828_cast = einsum(equation = var_4828_equation_0, values = (var_4648_cast, var_4807_cast));
            tensor<string, []> var_4830_equation_0 = const()[name = tensor<string, []>("op_4830_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_4830_cast = einsum(equation = var_4830_equation_0, values = (var_4652_cast, var_4808_cast));
            tensor<string, []> var_4832_equation_0 = const()[name = tensor<string, []>("op_4832_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_4832_cast = einsum(equation = var_4832_equation_0, values = (var_4656_cast, var_4809_cast));
            tensor<string, []> var_4834_equation_0 = const()[name = tensor<string, []>("op_4834_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_4834_cast = einsum(equation = var_4834_equation_0, values = (var_4660_cast, var_4810_cast));
            tensor<string, []> var_4836_equation_0 = const()[name = tensor<string, []>("op_4836_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_4836_cast = einsum(equation = var_4836_equation_0, values = (var_4664_cast, var_4811_cast));
            tensor<string, []> var_4838_equation_0 = const()[name = tensor<string, []>("op_4838_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_4838_cast = einsum(equation = var_4838_equation_0, values = (var_4668_cast, var_4812_cast));
            tensor<string, []> var_4840_equation_0 = const()[name = tensor<string, []>("op_4840_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_4840_cast = einsum(equation = var_4840_equation_0, values = (var_4672_cast, var_4813_cast));
            tensor<string, []> var_4842_equation_0 = const()[name = tensor<string, []>("op_4842_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_4842_cast = einsum(equation = var_4842_equation_0, values = (var_4676_cast, var_4814_cast));
            tensor<string, []> var_4844_equation_0 = const()[name = tensor<string, []>("op_4844_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_4844_cast = einsum(equation = var_4844_equation_0, values = (var_4680_cast, var_4815_cast));
            tensor<string, []> var_4846_equation_0 = const()[name = tensor<string, []>("op_4846_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_4846_cast = einsum(equation = var_4846_equation_0, values = (var_4684_cast, var_4816_cast));
            tensor<string, []> var_4848_equation_0 = const()[name = tensor<string, []>("op_4848_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_4848_cast = einsum(equation = var_4848_equation_0, values = (var_4688_cast, var_4817_cast));
            tensor<string, []> var_4850_equation_0 = const()[name = tensor<string, []>("op_4850_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_4850_cast = einsum(equation = var_4850_equation_0, values = (var_4692_cast, var_4818_cast));
            tensor<string, []> var_4852_equation_0 = const()[name = tensor<string, []>("op_4852_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_4852_cast = einsum(equation = var_4852_equation_0, values = (var_4696_cast, var_4819_cast));
            tensor<string, []> var_4854_equation_0 = const()[name = tensor<string, []>("op_4854_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_4854_cast = einsum(equation = var_4854_equation_0, values = (var_4700_cast, var_4820_cast));
            tensor<string, []> var_4856_equation_0 = const()[name = tensor<string, []>("op_4856_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_4856_cast = einsum(equation = var_4856_equation_0, values = (var_4704_cast, var_4821_cast));
            tensor<string, []> var_4858_equation_0 = const()[name = tensor<string, []>("op_4858_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_4858_cast = einsum(equation = var_4858_equation_0, values = (var_4708_cast, var_4822_cast));
            tensor<string, []> var_4860_equation_0 = const()[name = tensor<string, []>("op_4860_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_4860_cast = einsum(equation = var_4860_equation_0, values = (var_4712_cast, var_4823_cast));
            tensor<string, []> var_4862_equation_0 = const()[name = tensor<string, []>("op_4862_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_4862_cast = einsum(equation = var_4862_equation_0, values = (var_4716_cast, var_4824_cast));
            tensor<string, []> var_4864_equation_0 = const()[name = tensor<string, []>("op_4864_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_4864_cast = einsum(equation = var_4864_equation_0, values = (var_4720_cast, var_4825_cast));
            tensor<string, []> var_4866_equation_0 = const()[name = tensor<string, []>("op_4866_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_4866_cast = einsum(equation = var_4866_equation_0, values = (var_4724_cast, var_4826_cast));
            tensor<bool, []> input_215_interleave_0 = const()[name = tensor<string, []>("input_215_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 64]> input_215_cast = concat(axis = var_4371, interleave = input_215_interleave_0, values = (var_4828_cast, var_4830_cast, var_4832_cast, var_4834_cast, var_4836_cast, var_4838_cast, var_4840_cast, var_4842_cast, var_4844_cast, var_4846_cast, var_4848_cast, var_4850_cast, var_4852_cast, var_4854_cast, var_4856_cast, var_4858_cast, var_4860_cast, var_4862_cast, var_4864_cast, var_4866_cast));
            tensor<int32, [2]> var_4872 = const()[name = tensor<string, []>("op_4872"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4874 = const()[name = tensor<string, []>("op_4874"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_4876_pad_type_0 = const()[name = tensor<string, []>("op_4876_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_4876_pad_0 = const()[name = tensor<string, []>("op_4876_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(583441152)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(586718016)))];
            tensor<fp16, [2, 1280, 1, 64]> var_4876_cast = conv(bias = mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_4874, groups = var_4371, pad = var_4876_pad_0, pad_type = var_4876_pad_type_0, strides = var_4872, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_215_cast);
            tensor<fp16, [2, 1280, 1, 64]> inputs_39_cast = add(x = var_4876_cast, y = inputs_37_cast);
            tensor<int32, [1]> var_4880 = const()[name = tensor<string, []>("op_4880"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 64]> channels_mean_39_cast = reduce_mean(axes = var_4880, keep_dims = var_4366, x = inputs_39_cast);
            tensor<fp16, [2, 1280, 1, 64]> zero_mean_39_cast = sub(x = inputs_39_cast, y = channels_mean_39_cast);
            tensor<fp16, [2, 1280, 1, 64]> zero_mean_sq_39_cast = mul(x = zero_mean_39_cast, y = zero_mean_39_cast);
            tensor<int32, [1]> var_4884 = const()[name = tensor<string, []>("op_4884"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 64]> var_4885_cast = reduce_mean(axes = var_4884, keep_dims = var_4366, x = zero_mean_sq_39_cast);
            tensor<fp16, []> var_4886_to_fp16 = const()[name = tensor<string, []>("op_4886_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 64]> var_4887_cast = add(x = var_4885_cast, y = var_4886_to_fp16);
            tensor<fp16, []> denom_39_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_39_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 64]> denom_39_cast = rsqrt(epsilon = denom_39_epsilon_0_to_fp16, x = var_4887_cast);
            tensor<fp16, [2, 1280, 1, 64]> out_39_cast = mul(x = zero_mean_39_cast, y = denom_39_cast);
            tensor<fp16, [1, 1280, 1, 1]> var_4891_to_fp16 = const()[name = tensor<string, []>("op_4891_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(586720640)))];
            tensor<fp16, [2, 1280, 1, 64]> var_4892_cast = add(x = out_39_cast, y = var_4891_to_fp16);
            tensor<fp16, [1, 1280, 1, 1]> var_4894_to_fp16 = const()[name = tensor<string, []>("op_4894_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(586723264)))];
            tensor<fp16, [2, 1280, 1, 64]> hidden_states_127_cast = mul(x = var_4892_cast, y = var_4894_to_fp16);
            tensor<int32, [2]> var_4901 = const()[name = tensor<string, []>("op_4901"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4903 = const()[name = tensor<string, []>("op_4903"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_27_pad_type_0 = const()[name = tensor<string, []>("q_27_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_27_pad_0 = const()[name = tensor<string, []>("q_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(586725888)))];
            tensor<fp16, [2, 1280, 1, 64]> q_27_cast = conv(dilations = var_4903, groups = var_4371, pad = q_27_pad_0, pad_type = q_27_pad_type_0, strides = var_4901, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_127_cast);
            tensor<int32, [2]> var_4907 = const()[name = tensor<string, []>("op_4907"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4909 = const()[name = tensor<string, []>("op_4909"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_53_pad_type_0 = const()[name = tensor<string, []>("k_53_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_53_pad_0 = const()[name = tensor<string, []>("k_53_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1024, 1, 1]> mid_block_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(590002752)))];
            tensor<fp16, [2, 1280, 1, 77]> k_53_cast = conv(dilations = var_4909, groups = var_4371, pad = k_53_pad_0, pad_type = k_53_pad_type_0, strides = var_4907, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [2]> var_4913 = const()[name = tensor<string, []>("op_4913"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4915 = const()[name = tensor<string, []>("op_4915"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_27_pad_type_0 = const()[name = tensor<string, []>("v_27_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_27_pad_0 = const()[name = tensor<string, []>("v_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1024, 1, 1]> mid_block_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(592624256)))];
            tensor<fp16, [2, 1280, 1, 77]> v_27_cast = conv(dilations = var_4915, groups = var_4371, pad = v_27_pad_0, pad_type = v_27_pad_type_0, strides = var_4913, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [4]> var_4919_begin_0 = const()[name = tensor<string, []>("op_4919_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4919_end_0 = const()[name = tensor<string, []>("op_4919_end_0"), val = tensor<int32, [4]>([2, 64, 1, 64])];
            tensor<bool, [4]> var_4919_end_mask_0 = const()[name = tensor<string, []>("op_4919_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4919_cast = slice_by_index(begin = var_4919_begin_0, end = var_4919_end_0, end_mask = var_4919_end_mask_0, x = q_27_cast);
            tensor<int32, [4]> var_4923_begin_0 = const()[name = tensor<string, []>("op_4923_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_4923_end_0 = const()[name = tensor<string, []>("op_4923_end_0"), val = tensor<int32, [4]>([2, 128, 1, 64])];
            tensor<bool, [4]> var_4923_end_mask_0 = const()[name = tensor<string, []>("op_4923_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4923_cast = slice_by_index(begin = var_4923_begin_0, end = var_4923_end_0, end_mask = var_4923_end_mask_0, x = q_27_cast);
            tensor<int32, [4]> var_4927_begin_0 = const()[name = tensor<string, []>("op_4927_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_4927_end_0 = const()[name = tensor<string, []>("op_4927_end_0"), val = tensor<int32, [4]>([2, 192, 1, 64])];
            tensor<bool, [4]> var_4927_end_mask_0 = const()[name = tensor<string, []>("op_4927_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4927_cast = slice_by_index(begin = var_4927_begin_0, end = var_4927_end_0, end_mask = var_4927_end_mask_0, x = q_27_cast);
            tensor<int32, [4]> var_4931_begin_0 = const()[name = tensor<string, []>("op_4931_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_4931_end_0 = const()[name = tensor<string, []>("op_4931_end_0"), val = tensor<int32, [4]>([2, 256, 1, 64])];
            tensor<bool, [4]> var_4931_end_mask_0 = const()[name = tensor<string, []>("op_4931_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4931_cast = slice_by_index(begin = var_4931_begin_0, end = var_4931_end_0, end_mask = var_4931_end_mask_0, x = q_27_cast);
            tensor<int32, [4]> var_4935_begin_0 = const()[name = tensor<string, []>("op_4935_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_4935_end_0 = const()[name = tensor<string, []>("op_4935_end_0"), val = tensor<int32, [4]>([2, 320, 1, 64])];
            tensor<bool, [4]> var_4935_end_mask_0 = const()[name = tensor<string, []>("op_4935_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4935_cast = slice_by_index(begin = var_4935_begin_0, end = var_4935_end_0, end_mask = var_4935_end_mask_0, x = q_27_cast);
            tensor<int32, [4]> var_4939_begin_0 = const()[name = tensor<string, []>("op_4939_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_4939_end_0 = const()[name = tensor<string, []>("op_4939_end_0"), val = tensor<int32, [4]>([2, 384, 1, 64])];
            tensor<bool, [4]> var_4939_end_mask_0 = const()[name = tensor<string, []>("op_4939_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4939_cast = slice_by_index(begin = var_4939_begin_0, end = var_4939_end_0, end_mask = var_4939_end_mask_0, x = q_27_cast);
            tensor<int32, [4]> var_4943_begin_0 = const()[name = tensor<string, []>("op_4943_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_4943_end_0 = const()[name = tensor<string, []>("op_4943_end_0"), val = tensor<int32, [4]>([2, 448, 1, 64])];
            tensor<bool, [4]> var_4943_end_mask_0 = const()[name = tensor<string, []>("op_4943_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4943_cast = slice_by_index(begin = var_4943_begin_0, end = var_4943_end_0, end_mask = var_4943_end_mask_0, x = q_27_cast);
            tensor<int32, [4]> var_4947_begin_0 = const()[name = tensor<string, []>("op_4947_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_4947_end_0 = const()[name = tensor<string, []>("op_4947_end_0"), val = tensor<int32, [4]>([2, 512, 1, 64])];
            tensor<bool, [4]> var_4947_end_mask_0 = const()[name = tensor<string, []>("op_4947_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4947_cast = slice_by_index(begin = var_4947_begin_0, end = var_4947_end_0, end_mask = var_4947_end_mask_0, x = q_27_cast);
            tensor<int32, [4]> var_4951_begin_0 = const()[name = tensor<string, []>("op_4951_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_4951_end_0 = const()[name = tensor<string, []>("op_4951_end_0"), val = tensor<int32, [4]>([2, 576, 1, 64])];
            tensor<bool, [4]> var_4951_end_mask_0 = const()[name = tensor<string, []>("op_4951_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4951_cast = slice_by_index(begin = var_4951_begin_0, end = var_4951_end_0, end_mask = var_4951_end_mask_0, x = q_27_cast);
            tensor<int32, [4]> var_4955_begin_0 = const()[name = tensor<string, []>("op_4955_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_4955_end_0 = const()[name = tensor<string, []>("op_4955_end_0"), val = tensor<int32, [4]>([2, 640, 1, 64])];
            tensor<bool, [4]> var_4955_end_mask_0 = const()[name = tensor<string, []>("op_4955_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4955_cast = slice_by_index(begin = var_4955_begin_0, end = var_4955_end_0, end_mask = var_4955_end_mask_0, x = q_27_cast);
            tensor<int32, [4]> var_4959_begin_0 = const()[name = tensor<string, []>("op_4959_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_4959_end_0 = const()[name = tensor<string, []>("op_4959_end_0"), val = tensor<int32, [4]>([2, 704, 1, 64])];
            tensor<bool, [4]> var_4959_end_mask_0 = const()[name = tensor<string, []>("op_4959_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4959_cast = slice_by_index(begin = var_4959_begin_0, end = var_4959_end_0, end_mask = var_4959_end_mask_0, x = q_27_cast);
            tensor<int32, [4]> var_4963_begin_0 = const()[name = tensor<string, []>("op_4963_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_4963_end_0 = const()[name = tensor<string, []>("op_4963_end_0"), val = tensor<int32, [4]>([2, 768, 1, 64])];
            tensor<bool, [4]> var_4963_end_mask_0 = const()[name = tensor<string, []>("op_4963_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4963_cast = slice_by_index(begin = var_4963_begin_0, end = var_4963_end_0, end_mask = var_4963_end_mask_0, x = q_27_cast);
            tensor<int32, [4]> var_4967_begin_0 = const()[name = tensor<string, []>("op_4967_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_4967_end_0 = const()[name = tensor<string, []>("op_4967_end_0"), val = tensor<int32, [4]>([2, 832, 1, 64])];
            tensor<bool, [4]> var_4967_end_mask_0 = const()[name = tensor<string, []>("op_4967_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4967_cast = slice_by_index(begin = var_4967_begin_0, end = var_4967_end_0, end_mask = var_4967_end_mask_0, x = q_27_cast);
            tensor<int32, [4]> var_4971_begin_0 = const()[name = tensor<string, []>("op_4971_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_4971_end_0 = const()[name = tensor<string, []>("op_4971_end_0"), val = tensor<int32, [4]>([2, 896, 1, 64])];
            tensor<bool, [4]> var_4971_end_mask_0 = const()[name = tensor<string, []>("op_4971_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4971_cast = slice_by_index(begin = var_4971_begin_0, end = var_4971_end_0, end_mask = var_4971_end_mask_0, x = q_27_cast);
            tensor<int32, [4]> var_4975_begin_0 = const()[name = tensor<string, []>("op_4975_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_4975_end_0 = const()[name = tensor<string, []>("op_4975_end_0"), val = tensor<int32, [4]>([2, 960, 1, 64])];
            tensor<bool, [4]> var_4975_end_mask_0 = const()[name = tensor<string, []>("op_4975_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4975_cast = slice_by_index(begin = var_4975_begin_0, end = var_4975_end_0, end_mask = var_4975_end_mask_0, x = q_27_cast);
            tensor<int32, [4]> var_4979_begin_0 = const()[name = tensor<string, []>("op_4979_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_4979_end_0 = const()[name = tensor<string, []>("op_4979_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_4979_end_mask_0 = const()[name = tensor<string, []>("op_4979_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4979_cast = slice_by_index(begin = var_4979_begin_0, end = var_4979_end_0, end_mask = var_4979_end_mask_0, x = q_27_cast);
            tensor<int32, [4]> var_4983_begin_0 = const()[name = tensor<string, []>("op_4983_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_4983_end_0 = const()[name = tensor<string, []>("op_4983_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 64])];
            tensor<bool, [4]> var_4983_end_mask_0 = const()[name = tensor<string, []>("op_4983_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4983_cast = slice_by_index(begin = var_4983_begin_0, end = var_4983_end_0, end_mask = var_4983_end_mask_0, x = q_27_cast);
            tensor<int32, [4]> var_4987_begin_0 = const()[name = tensor<string, []>("op_4987_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_4987_end_0 = const()[name = tensor<string, []>("op_4987_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 64])];
            tensor<bool, [4]> var_4987_end_mask_0 = const()[name = tensor<string, []>("op_4987_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4987_cast = slice_by_index(begin = var_4987_begin_0, end = var_4987_end_0, end_mask = var_4987_end_mask_0, x = q_27_cast);
            tensor<int32, [4]> var_4991_begin_0 = const()[name = tensor<string, []>("op_4991_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_4991_end_0 = const()[name = tensor<string, []>("op_4991_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 64])];
            tensor<bool, [4]> var_4991_end_mask_0 = const()[name = tensor<string, []>("op_4991_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4991_cast = slice_by_index(begin = var_4991_begin_0, end = var_4991_end_0, end_mask = var_4991_end_mask_0, x = q_27_cast);
            tensor<int32, [4]> var_4995_begin_0 = const()[name = tensor<string, []>("op_4995_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_4995_end_0 = const()[name = tensor<string, []>("op_4995_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 64])];
            tensor<bool, [4]> var_4995_end_mask_0 = const()[name = tensor<string, []>("op_4995_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_4995_cast = slice_by_index(begin = var_4995_begin_0, end = var_4995_end_0, end_mask = var_4995_end_mask_0, x = q_27_cast);
            tensor<int32, [4]> k_55_perm_0 = const()[name = tensor<string, []>("k_55_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_5002_begin_0 = const()[name = tensor<string, []>("op_5002_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5002_end_0 = const()[name = tensor<string, []>("op_5002_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_5002_end_mask_0 = const()[name = tensor<string, []>("op_5002_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_18 = transpose(perm = k_55_perm_0, x = k_53_cast);
            tensor<fp16, [2, 77, 1, 64]> var_5002_cast = slice_by_index(begin = var_5002_begin_0, end = var_5002_end_0, end_mask = var_5002_end_mask_0, x = transpose_18);
            tensor<int32, [4]> var_5006_begin_0 = const()[name = tensor<string, []>("op_5006_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_5006_end_0 = const()[name = tensor<string, []>("op_5006_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_5006_end_mask_0 = const()[name = tensor<string, []>("op_5006_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5006_cast = slice_by_index(begin = var_5006_begin_0, end = var_5006_end_0, end_mask = var_5006_end_mask_0, x = transpose_18);
            tensor<int32, [4]> var_5010_begin_0 = const()[name = tensor<string, []>("op_5010_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_5010_end_0 = const()[name = tensor<string, []>("op_5010_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_5010_end_mask_0 = const()[name = tensor<string, []>("op_5010_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5010_cast = slice_by_index(begin = var_5010_begin_0, end = var_5010_end_0, end_mask = var_5010_end_mask_0, x = transpose_18);
            tensor<int32, [4]> var_5014_begin_0 = const()[name = tensor<string, []>("op_5014_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_5014_end_0 = const()[name = tensor<string, []>("op_5014_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_5014_end_mask_0 = const()[name = tensor<string, []>("op_5014_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5014_cast = slice_by_index(begin = var_5014_begin_0, end = var_5014_end_0, end_mask = var_5014_end_mask_0, x = transpose_18);
            tensor<int32, [4]> var_5018_begin_0 = const()[name = tensor<string, []>("op_5018_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_5018_end_0 = const()[name = tensor<string, []>("op_5018_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_5018_end_mask_0 = const()[name = tensor<string, []>("op_5018_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5018_cast = slice_by_index(begin = var_5018_begin_0, end = var_5018_end_0, end_mask = var_5018_end_mask_0, x = transpose_18);
            tensor<int32, [4]> var_5022_begin_0 = const()[name = tensor<string, []>("op_5022_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_5022_end_0 = const()[name = tensor<string, []>("op_5022_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_5022_end_mask_0 = const()[name = tensor<string, []>("op_5022_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5022_cast = slice_by_index(begin = var_5022_begin_0, end = var_5022_end_0, end_mask = var_5022_end_mask_0, x = transpose_18);
            tensor<int32, [4]> var_5026_begin_0 = const()[name = tensor<string, []>("op_5026_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_5026_end_0 = const()[name = tensor<string, []>("op_5026_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_5026_end_mask_0 = const()[name = tensor<string, []>("op_5026_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5026_cast = slice_by_index(begin = var_5026_begin_0, end = var_5026_end_0, end_mask = var_5026_end_mask_0, x = transpose_18);
            tensor<int32, [4]> var_5030_begin_0 = const()[name = tensor<string, []>("op_5030_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_5030_end_0 = const()[name = tensor<string, []>("op_5030_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_5030_end_mask_0 = const()[name = tensor<string, []>("op_5030_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5030_cast = slice_by_index(begin = var_5030_begin_0, end = var_5030_end_0, end_mask = var_5030_end_mask_0, x = transpose_18);
            tensor<int32, [4]> var_5034_begin_0 = const()[name = tensor<string, []>("op_5034_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_5034_end_0 = const()[name = tensor<string, []>("op_5034_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_5034_end_mask_0 = const()[name = tensor<string, []>("op_5034_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5034_cast = slice_by_index(begin = var_5034_begin_0, end = var_5034_end_0, end_mask = var_5034_end_mask_0, x = transpose_18);
            tensor<int32, [4]> var_5038_begin_0 = const()[name = tensor<string, []>("op_5038_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_5038_end_0 = const()[name = tensor<string, []>("op_5038_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_5038_end_mask_0 = const()[name = tensor<string, []>("op_5038_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5038_cast = slice_by_index(begin = var_5038_begin_0, end = var_5038_end_0, end_mask = var_5038_end_mask_0, x = transpose_18);
            tensor<int32, [4]> var_5042_begin_0 = const()[name = tensor<string, []>("op_5042_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_5042_end_0 = const()[name = tensor<string, []>("op_5042_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_5042_end_mask_0 = const()[name = tensor<string, []>("op_5042_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5042_cast = slice_by_index(begin = var_5042_begin_0, end = var_5042_end_0, end_mask = var_5042_end_mask_0, x = transpose_18);
            tensor<int32, [4]> var_5046_begin_0 = const()[name = tensor<string, []>("op_5046_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_5046_end_0 = const()[name = tensor<string, []>("op_5046_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_5046_end_mask_0 = const()[name = tensor<string, []>("op_5046_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5046_cast = slice_by_index(begin = var_5046_begin_0, end = var_5046_end_0, end_mask = var_5046_end_mask_0, x = transpose_18);
            tensor<int32, [4]> var_5050_begin_0 = const()[name = tensor<string, []>("op_5050_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_5050_end_0 = const()[name = tensor<string, []>("op_5050_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_5050_end_mask_0 = const()[name = tensor<string, []>("op_5050_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5050_cast = slice_by_index(begin = var_5050_begin_0, end = var_5050_end_0, end_mask = var_5050_end_mask_0, x = transpose_18);
            tensor<int32, [4]> var_5054_begin_0 = const()[name = tensor<string, []>("op_5054_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_5054_end_0 = const()[name = tensor<string, []>("op_5054_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_5054_end_mask_0 = const()[name = tensor<string, []>("op_5054_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5054_cast = slice_by_index(begin = var_5054_begin_0, end = var_5054_end_0, end_mask = var_5054_end_mask_0, x = transpose_18);
            tensor<int32, [4]> var_5058_begin_0 = const()[name = tensor<string, []>("op_5058_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_5058_end_0 = const()[name = tensor<string, []>("op_5058_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_5058_end_mask_0 = const()[name = tensor<string, []>("op_5058_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5058_cast = slice_by_index(begin = var_5058_begin_0, end = var_5058_end_0, end_mask = var_5058_end_mask_0, x = transpose_18);
            tensor<int32, [4]> var_5062_begin_0 = const()[name = tensor<string, []>("op_5062_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_5062_end_0 = const()[name = tensor<string, []>("op_5062_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_5062_end_mask_0 = const()[name = tensor<string, []>("op_5062_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5062_cast = slice_by_index(begin = var_5062_begin_0, end = var_5062_end_0, end_mask = var_5062_end_mask_0, x = transpose_18);
            tensor<int32, [4]> var_5066_begin_0 = const()[name = tensor<string, []>("op_5066_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_5066_end_0 = const()[name = tensor<string, []>("op_5066_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_5066_end_mask_0 = const()[name = tensor<string, []>("op_5066_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5066_cast = slice_by_index(begin = var_5066_begin_0, end = var_5066_end_0, end_mask = var_5066_end_mask_0, x = transpose_18);
            tensor<int32, [4]> var_5070_begin_0 = const()[name = tensor<string, []>("op_5070_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_5070_end_0 = const()[name = tensor<string, []>("op_5070_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_5070_end_mask_0 = const()[name = tensor<string, []>("op_5070_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5070_cast = slice_by_index(begin = var_5070_begin_0, end = var_5070_end_0, end_mask = var_5070_end_mask_0, x = transpose_18);
            tensor<int32, [4]> var_5074_begin_0 = const()[name = tensor<string, []>("op_5074_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_5074_end_0 = const()[name = tensor<string, []>("op_5074_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_5074_end_mask_0 = const()[name = tensor<string, []>("op_5074_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5074_cast = slice_by_index(begin = var_5074_begin_0, end = var_5074_end_0, end_mask = var_5074_end_mask_0, x = transpose_18);
            tensor<int32, [4]> var_5078_begin_0 = const()[name = tensor<string, []>("op_5078_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_5078_end_0 = const()[name = tensor<string, []>("op_5078_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_5078_end_mask_0 = const()[name = tensor<string, []>("op_5078_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5078_cast = slice_by_index(begin = var_5078_begin_0, end = var_5078_end_0, end_mask = var_5078_end_mask_0, x = transpose_18);
            tensor<int32, [4]> var_5080_begin_0 = const()[name = tensor<string, []>("op_5080_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5080_end_0 = const()[name = tensor<string, []>("op_5080_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_5080_end_mask_0 = const()[name = tensor<string, []>("op_5080_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5080_cast = slice_by_index(begin = var_5080_begin_0, end = var_5080_end_0, end_mask = var_5080_end_mask_0, x = v_27_cast);
            tensor<int32, [4]> var_5084_begin_0 = const()[name = tensor<string, []>("op_5084_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_5084_end_0 = const()[name = tensor<string, []>("op_5084_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_5084_end_mask_0 = const()[name = tensor<string, []>("op_5084_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5084_cast = slice_by_index(begin = var_5084_begin_0, end = var_5084_end_0, end_mask = var_5084_end_mask_0, x = v_27_cast);
            tensor<int32, [4]> var_5088_begin_0 = const()[name = tensor<string, []>("op_5088_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_5088_end_0 = const()[name = tensor<string, []>("op_5088_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_5088_end_mask_0 = const()[name = tensor<string, []>("op_5088_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5088_cast = slice_by_index(begin = var_5088_begin_0, end = var_5088_end_0, end_mask = var_5088_end_mask_0, x = v_27_cast);
            tensor<int32, [4]> var_5092_begin_0 = const()[name = tensor<string, []>("op_5092_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_5092_end_0 = const()[name = tensor<string, []>("op_5092_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_5092_end_mask_0 = const()[name = tensor<string, []>("op_5092_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5092_cast = slice_by_index(begin = var_5092_begin_0, end = var_5092_end_0, end_mask = var_5092_end_mask_0, x = v_27_cast);
            tensor<int32, [4]> var_5096_begin_0 = const()[name = tensor<string, []>("op_5096_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_5096_end_0 = const()[name = tensor<string, []>("op_5096_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_5096_end_mask_0 = const()[name = tensor<string, []>("op_5096_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5096_cast = slice_by_index(begin = var_5096_begin_0, end = var_5096_end_0, end_mask = var_5096_end_mask_0, x = v_27_cast);
            tensor<int32, [4]> var_5100_begin_0 = const()[name = tensor<string, []>("op_5100_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_5100_end_0 = const()[name = tensor<string, []>("op_5100_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_5100_end_mask_0 = const()[name = tensor<string, []>("op_5100_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5100_cast = slice_by_index(begin = var_5100_begin_0, end = var_5100_end_0, end_mask = var_5100_end_mask_0, x = v_27_cast);
            tensor<int32, [4]> var_5104_begin_0 = const()[name = tensor<string, []>("op_5104_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_5104_end_0 = const()[name = tensor<string, []>("op_5104_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_5104_end_mask_0 = const()[name = tensor<string, []>("op_5104_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5104_cast = slice_by_index(begin = var_5104_begin_0, end = var_5104_end_0, end_mask = var_5104_end_mask_0, x = v_27_cast);
            tensor<int32, [4]> var_5108_begin_0 = const()[name = tensor<string, []>("op_5108_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_5108_end_0 = const()[name = tensor<string, []>("op_5108_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_5108_end_mask_0 = const()[name = tensor<string, []>("op_5108_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5108_cast = slice_by_index(begin = var_5108_begin_0, end = var_5108_end_0, end_mask = var_5108_end_mask_0, x = v_27_cast);
            tensor<int32, [4]> var_5112_begin_0 = const()[name = tensor<string, []>("op_5112_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_5112_end_0 = const()[name = tensor<string, []>("op_5112_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_5112_end_mask_0 = const()[name = tensor<string, []>("op_5112_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5112_cast = slice_by_index(begin = var_5112_begin_0, end = var_5112_end_0, end_mask = var_5112_end_mask_0, x = v_27_cast);
            tensor<int32, [4]> var_5116_begin_0 = const()[name = tensor<string, []>("op_5116_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_5116_end_0 = const()[name = tensor<string, []>("op_5116_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_5116_end_mask_0 = const()[name = tensor<string, []>("op_5116_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5116_cast = slice_by_index(begin = var_5116_begin_0, end = var_5116_end_0, end_mask = var_5116_end_mask_0, x = v_27_cast);
            tensor<int32, [4]> var_5120_begin_0 = const()[name = tensor<string, []>("op_5120_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_5120_end_0 = const()[name = tensor<string, []>("op_5120_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_5120_end_mask_0 = const()[name = tensor<string, []>("op_5120_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5120_cast = slice_by_index(begin = var_5120_begin_0, end = var_5120_end_0, end_mask = var_5120_end_mask_0, x = v_27_cast);
            tensor<int32, [4]> var_5124_begin_0 = const()[name = tensor<string, []>("op_5124_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_5124_end_0 = const()[name = tensor<string, []>("op_5124_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_5124_end_mask_0 = const()[name = tensor<string, []>("op_5124_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5124_cast = slice_by_index(begin = var_5124_begin_0, end = var_5124_end_0, end_mask = var_5124_end_mask_0, x = v_27_cast);
            tensor<int32, [4]> var_5128_begin_0 = const()[name = tensor<string, []>("op_5128_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_5128_end_0 = const()[name = tensor<string, []>("op_5128_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_5128_end_mask_0 = const()[name = tensor<string, []>("op_5128_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5128_cast = slice_by_index(begin = var_5128_begin_0, end = var_5128_end_0, end_mask = var_5128_end_mask_0, x = v_27_cast);
            tensor<int32, [4]> var_5132_begin_0 = const()[name = tensor<string, []>("op_5132_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_5132_end_0 = const()[name = tensor<string, []>("op_5132_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_5132_end_mask_0 = const()[name = tensor<string, []>("op_5132_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5132_cast = slice_by_index(begin = var_5132_begin_0, end = var_5132_end_0, end_mask = var_5132_end_mask_0, x = v_27_cast);
            tensor<int32, [4]> var_5136_begin_0 = const()[name = tensor<string, []>("op_5136_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_5136_end_0 = const()[name = tensor<string, []>("op_5136_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_5136_end_mask_0 = const()[name = tensor<string, []>("op_5136_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5136_cast = slice_by_index(begin = var_5136_begin_0, end = var_5136_end_0, end_mask = var_5136_end_mask_0, x = v_27_cast);
            tensor<int32, [4]> var_5140_begin_0 = const()[name = tensor<string, []>("op_5140_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_5140_end_0 = const()[name = tensor<string, []>("op_5140_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_5140_end_mask_0 = const()[name = tensor<string, []>("op_5140_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5140_cast = slice_by_index(begin = var_5140_begin_0, end = var_5140_end_0, end_mask = var_5140_end_mask_0, x = v_27_cast);
            tensor<int32, [4]> var_5144_begin_0 = const()[name = tensor<string, []>("op_5144_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_5144_end_0 = const()[name = tensor<string, []>("op_5144_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_5144_end_mask_0 = const()[name = tensor<string, []>("op_5144_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5144_cast = slice_by_index(begin = var_5144_begin_0, end = var_5144_end_0, end_mask = var_5144_end_mask_0, x = v_27_cast);
            tensor<int32, [4]> var_5148_begin_0 = const()[name = tensor<string, []>("op_5148_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_5148_end_0 = const()[name = tensor<string, []>("op_5148_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_5148_end_mask_0 = const()[name = tensor<string, []>("op_5148_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5148_cast = slice_by_index(begin = var_5148_begin_0, end = var_5148_end_0, end_mask = var_5148_end_mask_0, x = v_27_cast);
            tensor<int32, [4]> var_5152_begin_0 = const()[name = tensor<string, []>("op_5152_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_5152_end_0 = const()[name = tensor<string, []>("op_5152_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_5152_end_mask_0 = const()[name = tensor<string, []>("op_5152_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5152_cast = slice_by_index(begin = var_5152_begin_0, end = var_5152_end_0, end_mask = var_5152_end_mask_0, x = v_27_cast);
            tensor<int32, [4]> var_5156_begin_0 = const()[name = tensor<string, []>("op_5156_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_5156_end_0 = const()[name = tensor<string, []>("op_5156_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_5156_end_mask_0 = const()[name = tensor<string, []>("op_5156_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5156_cast = slice_by_index(begin = var_5156_begin_0, end = var_5156_end_0, end_mask = var_5156_end_mask_0, x = v_27_cast);
            tensor<string, []> var_5160_equation_0 = const()[name = tensor<string, []>("op_5160_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_5160_cast = einsum(equation = var_5160_equation_0, values = (var_5002_cast, var_4919_cast));
            tensor<fp16, []> var_5161_to_fp16 = const()[name = tensor<string, []>("op_5161_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_321_cast = mul(x = var_5160_cast, y = var_5161_to_fp16);
            tensor<string, []> var_5164_equation_0 = const()[name = tensor<string, []>("op_5164_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_5164_cast = einsum(equation = var_5164_equation_0, values = (var_5006_cast, var_4923_cast));
            tensor<fp16, []> var_5165_to_fp16 = const()[name = tensor<string, []>("op_5165_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_323_cast = mul(x = var_5164_cast, y = var_5165_to_fp16);
            tensor<string, []> var_5168_equation_0 = const()[name = tensor<string, []>("op_5168_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_5168_cast = einsum(equation = var_5168_equation_0, values = (var_5010_cast, var_4927_cast));
            tensor<fp16, []> var_5169_to_fp16 = const()[name = tensor<string, []>("op_5169_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_325_cast = mul(x = var_5168_cast, y = var_5169_to_fp16);
            tensor<string, []> var_5172_equation_0 = const()[name = tensor<string, []>("op_5172_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_5172_cast = einsum(equation = var_5172_equation_0, values = (var_5014_cast, var_4931_cast));
            tensor<fp16, []> var_5173_to_fp16 = const()[name = tensor<string, []>("op_5173_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_327_cast = mul(x = var_5172_cast, y = var_5173_to_fp16);
            tensor<string, []> var_5176_equation_0 = const()[name = tensor<string, []>("op_5176_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_5176_cast = einsum(equation = var_5176_equation_0, values = (var_5018_cast, var_4935_cast));
            tensor<fp16, []> var_5177_to_fp16 = const()[name = tensor<string, []>("op_5177_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_329_cast = mul(x = var_5176_cast, y = var_5177_to_fp16);
            tensor<string, []> var_5180_equation_0 = const()[name = tensor<string, []>("op_5180_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_5180_cast = einsum(equation = var_5180_equation_0, values = (var_5022_cast, var_4939_cast));
            tensor<fp16, []> var_5181_to_fp16 = const()[name = tensor<string, []>("op_5181_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_331_cast = mul(x = var_5180_cast, y = var_5181_to_fp16);
            tensor<string, []> var_5184_equation_0 = const()[name = tensor<string, []>("op_5184_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_5184_cast = einsum(equation = var_5184_equation_0, values = (var_5026_cast, var_4943_cast));
            tensor<fp16, []> var_5185_to_fp16 = const()[name = tensor<string, []>("op_5185_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_333_cast = mul(x = var_5184_cast, y = var_5185_to_fp16);
            tensor<string, []> var_5188_equation_0 = const()[name = tensor<string, []>("op_5188_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_5188_cast = einsum(equation = var_5188_equation_0, values = (var_5030_cast, var_4947_cast));
            tensor<fp16, []> var_5189_to_fp16 = const()[name = tensor<string, []>("op_5189_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_335_cast = mul(x = var_5188_cast, y = var_5189_to_fp16);
            tensor<string, []> var_5192_equation_0 = const()[name = tensor<string, []>("op_5192_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_5192_cast = einsum(equation = var_5192_equation_0, values = (var_5034_cast, var_4951_cast));
            tensor<fp16, []> var_5193_to_fp16 = const()[name = tensor<string, []>("op_5193_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_337_cast = mul(x = var_5192_cast, y = var_5193_to_fp16);
            tensor<string, []> var_5196_equation_0 = const()[name = tensor<string, []>("op_5196_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_5196_cast = einsum(equation = var_5196_equation_0, values = (var_5038_cast, var_4955_cast));
            tensor<fp16, []> var_5197_to_fp16 = const()[name = tensor<string, []>("op_5197_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_339_cast = mul(x = var_5196_cast, y = var_5197_to_fp16);
            tensor<string, []> var_5200_equation_0 = const()[name = tensor<string, []>("op_5200_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_5200_cast = einsum(equation = var_5200_equation_0, values = (var_5042_cast, var_4959_cast));
            tensor<fp16, []> var_5201_to_fp16 = const()[name = tensor<string, []>("op_5201_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_341_cast = mul(x = var_5200_cast, y = var_5201_to_fp16);
            tensor<string, []> var_5204_equation_0 = const()[name = tensor<string, []>("op_5204_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_5204_cast = einsum(equation = var_5204_equation_0, values = (var_5046_cast, var_4963_cast));
            tensor<fp16, []> var_5205_to_fp16 = const()[name = tensor<string, []>("op_5205_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_343_cast = mul(x = var_5204_cast, y = var_5205_to_fp16);
            tensor<string, []> var_5208_equation_0 = const()[name = tensor<string, []>("op_5208_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_5208_cast = einsum(equation = var_5208_equation_0, values = (var_5050_cast, var_4967_cast));
            tensor<fp16, []> var_5209_to_fp16 = const()[name = tensor<string, []>("op_5209_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_345_cast = mul(x = var_5208_cast, y = var_5209_to_fp16);
            tensor<string, []> var_5212_equation_0 = const()[name = tensor<string, []>("op_5212_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_5212_cast = einsum(equation = var_5212_equation_0, values = (var_5054_cast, var_4971_cast));
            tensor<fp16, []> var_5213_to_fp16 = const()[name = tensor<string, []>("op_5213_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_347_cast = mul(x = var_5212_cast, y = var_5213_to_fp16);
            tensor<string, []> var_5216_equation_0 = const()[name = tensor<string, []>("op_5216_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_5216_cast = einsum(equation = var_5216_equation_0, values = (var_5058_cast, var_4975_cast));
            tensor<fp16, []> var_5217_to_fp16 = const()[name = tensor<string, []>("op_5217_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_349_cast = mul(x = var_5216_cast, y = var_5217_to_fp16);
            tensor<string, []> var_5220_equation_0 = const()[name = tensor<string, []>("op_5220_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_5220_cast = einsum(equation = var_5220_equation_0, values = (var_5062_cast, var_4979_cast));
            tensor<fp16, []> var_5221_to_fp16 = const()[name = tensor<string, []>("op_5221_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_351_cast = mul(x = var_5220_cast, y = var_5221_to_fp16);
            tensor<string, []> var_5224_equation_0 = const()[name = tensor<string, []>("op_5224_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_5224_cast = einsum(equation = var_5224_equation_0, values = (var_5066_cast, var_4983_cast));
            tensor<fp16, []> var_5225_to_fp16 = const()[name = tensor<string, []>("op_5225_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_353_cast = mul(x = var_5224_cast, y = var_5225_to_fp16);
            tensor<string, []> var_5228_equation_0 = const()[name = tensor<string, []>("op_5228_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_5228_cast = einsum(equation = var_5228_equation_0, values = (var_5070_cast, var_4987_cast));
            tensor<fp16, []> var_5229_to_fp16 = const()[name = tensor<string, []>("op_5229_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_355_cast = mul(x = var_5228_cast, y = var_5229_to_fp16);
            tensor<string, []> var_5232_equation_0 = const()[name = tensor<string, []>("op_5232_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_5232_cast = einsum(equation = var_5232_equation_0, values = (var_5074_cast, var_4991_cast));
            tensor<fp16, []> var_5233_to_fp16 = const()[name = tensor<string, []>("op_5233_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_357_cast = mul(x = var_5232_cast, y = var_5233_to_fp16);
            tensor<string, []> var_5236_equation_0 = const()[name = tensor<string, []>("op_5236_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_5236_cast = einsum(equation = var_5236_equation_0, values = (var_5078_cast, var_4995_cast));
            tensor<fp16, []> var_5237_to_fp16 = const()[name = tensor<string, []>("op_5237_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_359_cast = mul(x = var_5236_cast, y = var_5237_to_fp16);
            tensor<fp16, [2, 77, 1, 64]> var_5239_cast = softmax(axis = var_4371, x = aw_321_cast);
            tensor<fp16, [2, 77, 1, 64]> var_5240_cast = softmax(axis = var_4371, x = aw_323_cast);
            tensor<fp16, [2, 77, 1, 64]> var_5241_cast = softmax(axis = var_4371, x = aw_325_cast);
            tensor<fp16, [2, 77, 1, 64]> var_5242_cast = softmax(axis = var_4371, x = aw_327_cast);
            tensor<fp16, [2, 77, 1, 64]> var_5243_cast = softmax(axis = var_4371, x = aw_329_cast);
            tensor<fp16, [2, 77, 1, 64]> var_5244_cast = softmax(axis = var_4371, x = aw_331_cast);
            tensor<fp16, [2, 77, 1, 64]> var_5245_cast = softmax(axis = var_4371, x = aw_333_cast);
            tensor<fp16, [2, 77, 1, 64]> var_5246_cast = softmax(axis = var_4371, x = aw_335_cast);
            tensor<fp16, [2, 77, 1, 64]> var_5247_cast = softmax(axis = var_4371, x = aw_337_cast);
            tensor<fp16, [2, 77, 1, 64]> var_5248_cast = softmax(axis = var_4371, x = aw_339_cast);
            tensor<fp16, [2, 77, 1, 64]> var_5249_cast = softmax(axis = var_4371, x = aw_341_cast);
            tensor<fp16, [2, 77, 1, 64]> var_5250_cast = softmax(axis = var_4371, x = aw_343_cast);
            tensor<fp16, [2, 77, 1, 64]> var_5251_cast = softmax(axis = var_4371, x = aw_345_cast);
            tensor<fp16, [2, 77, 1, 64]> var_5252_cast = softmax(axis = var_4371, x = aw_347_cast);
            tensor<fp16, [2, 77, 1, 64]> var_5253_cast = softmax(axis = var_4371, x = aw_349_cast);
            tensor<fp16, [2, 77, 1, 64]> var_5254_cast = softmax(axis = var_4371, x = aw_351_cast);
            tensor<fp16, [2, 77, 1, 64]> var_5255_cast = softmax(axis = var_4371, x = aw_353_cast);
            tensor<fp16, [2, 77, 1, 64]> var_5256_cast = softmax(axis = var_4371, x = aw_355_cast);
            tensor<fp16, [2, 77, 1, 64]> var_5257_cast = softmax(axis = var_4371, x = aw_357_cast);
            tensor<fp16, [2, 77, 1, 64]> var_5258_cast = softmax(axis = var_4371, x = aw_359_cast);
            tensor<string, []> var_5260_equation_0 = const()[name = tensor<string, []>("op_5260_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_5260_cast = einsum(equation = var_5260_equation_0, values = (var_5080_cast, var_5239_cast));
            tensor<string, []> var_5262_equation_0 = const()[name = tensor<string, []>("op_5262_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_5262_cast = einsum(equation = var_5262_equation_0, values = (var_5084_cast, var_5240_cast));
            tensor<string, []> var_5264_equation_0 = const()[name = tensor<string, []>("op_5264_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_5264_cast = einsum(equation = var_5264_equation_0, values = (var_5088_cast, var_5241_cast));
            tensor<string, []> var_5266_equation_0 = const()[name = tensor<string, []>("op_5266_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_5266_cast = einsum(equation = var_5266_equation_0, values = (var_5092_cast, var_5242_cast));
            tensor<string, []> var_5268_equation_0 = const()[name = tensor<string, []>("op_5268_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_5268_cast = einsum(equation = var_5268_equation_0, values = (var_5096_cast, var_5243_cast));
            tensor<string, []> var_5270_equation_0 = const()[name = tensor<string, []>("op_5270_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_5270_cast = einsum(equation = var_5270_equation_0, values = (var_5100_cast, var_5244_cast));
            tensor<string, []> var_5272_equation_0 = const()[name = tensor<string, []>("op_5272_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_5272_cast = einsum(equation = var_5272_equation_0, values = (var_5104_cast, var_5245_cast));
            tensor<string, []> var_5274_equation_0 = const()[name = tensor<string, []>("op_5274_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_5274_cast = einsum(equation = var_5274_equation_0, values = (var_5108_cast, var_5246_cast));
            tensor<string, []> var_5276_equation_0 = const()[name = tensor<string, []>("op_5276_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_5276_cast = einsum(equation = var_5276_equation_0, values = (var_5112_cast, var_5247_cast));
            tensor<string, []> var_5278_equation_0 = const()[name = tensor<string, []>("op_5278_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_5278_cast = einsum(equation = var_5278_equation_0, values = (var_5116_cast, var_5248_cast));
            tensor<string, []> var_5280_equation_0 = const()[name = tensor<string, []>("op_5280_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_5280_cast = einsum(equation = var_5280_equation_0, values = (var_5120_cast, var_5249_cast));
            tensor<string, []> var_5282_equation_0 = const()[name = tensor<string, []>("op_5282_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_5282_cast = einsum(equation = var_5282_equation_0, values = (var_5124_cast, var_5250_cast));
            tensor<string, []> var_5284_equation_0 = const()[name = tensor<string, []>("op_5284_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_5284_cast = einsum(equation = var_5284_equation_0, values = (var_5128_cast, var_5251_cast));
            tensor<string, []> var_5286_equation_0 = const()[name = tensor<string, []>("op_5286_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_5286_cast = einsum(equation = var_5286_equation_0, values = (var_5132_cast, var_5252_cast));
            tensor<string, []> var_5288_equation_0 = const()[name = tensor<string, []>("op_5288_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_5288_cast = einsum(equation = var_5288_equation_0, values = (var_5136_cast, var_5253_cast));
            tensor<string, []> var_5290_equation_0 = const()[name = tensor<string, []>("op_5290_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_5290_cast = einsum(equation = var_5290_equation_0, values = (var_5140_cast, var_5254_cast));
            tensor<string, []> var_5292_equation_0 = const()[name = tensor<string, []>("op_5292_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_5292_cast = einsum(equation = var_5292_equation_0, values = (var_5144_cast, var_5255_cast));
            tensor<string, []> var_5294_equation_0 = const()[name = tensor<string, []>("op_5294_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_5294_cast = einsum(equation = var_5294_equation_0, values = (var_5148_cast, var_5256_cast));
            tensor<string, []> var_5296_equation_0 = const()[name = tensor<string, []>("op_5296_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_5296_cast = einsum(equation = var_5296_equation_0, values = (var_5152_cast, var_5257_cast));
            tensor<string, []> var_5298_equation_0 = const()[name = tensor<string, []>("op_5298_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_5298_cast = einsum(equation = var_5298_equation_0, values = (var_5156_cast, var_5258_cast));
            tensor<bool, []> input_217_interleave_0 = const()[name = tensor<string, []>("input_217_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 64]> input_217_cast = concat(axis = var_4371, interleave = input_217_interleave_0, values = (var_5260_cast, var_5262_cast, var_5264_cast, var_5266_cast, var_5268_cast, var_5270_cast, var_5272_cast, var_5274_cast, var_5276_cast, var_5278_cast, var_5280_cast, var_5282_cast, var_5284_cast, var_5286_cast, var_5288_cast, var_5290_cast, var_5292_cast, var_5294_cast, var_5296_cast, var_5298_cast));
            tensor<int32, [2]> var_5304 = const()[name = tensor<string, []>("op_5304"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5306 = const()[name = tensor<string, []>("op_5306"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_5308_pad_type_0 = const()[name = tensor<string, []>("op_5308_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_5308_pad_0 = const()[name = tensor<string, []>("op_5308_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(595245760)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(598522624)))];
            tensor<fp16, [2, 1280, 1, 64]> var_5308_cast = conv(bias = mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_5306, groups = var_4371, pad = var_5308_pad_0, pad_type = var_5308_pad_type_0, strides = var_5304, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_217_cast);
            tensor<fp16, [2, 1280, 1, 64]> inputs_41_cast = add(x = var_5308_cast, y = inputs_39_cast);
            tensor<int32, [1]> var_5312 = const()[name = tensor<string, []>("op_5312"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 64]> channels_mean_41_cast = reduce_mean(axes = var_5312, keep_dims = var_4366, x = inputs_41_cast);
            tensor<fp16, [2, 1280, 1, 64]> zero_mean_41_cast = sub(x = inputs_41_cast, y = channels_mean_41_cast);
            tensor<fp16, [2, 1280, 1, 64]> zero_mean_sq_41_cast = mul(x = zero_mean_41_cast, y = zero_mean_41_cast);
            tensor<int32, [1]> var_5316 = const()[name = tensor<string, []>("op_5316"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 64]> var_5317_cast = reduce_mean(axes = var_5316, keep_dims = var_4366, x = zero_mean_sq_41_cast);
            tensor<fp16, []> var_5318_to_fp16 = const()[name = tensor<string, []>("op_5318_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 64]> var_5319_cast = add(x = var_5317_cast, y = var_5318_to_fp16);
            tensor<fp16, []> denom_41_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_41_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 64]> denom_41_cast = rsqrt(epsilon = denom_41_epsilon_0_to_fp16, x = var_5319_cast);
            tensor<fp16, [2, 1280, 1, 64]> out_41_cast = mul(x = zero_mean_41_cast, y = denom_41_cast);
            tensor<fp16, [1, 1280, 1, 1]> var_5323_to_fp16 = const()[name = tensor<string, []>("op_5323_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(598525248)))];
            tensor<fp16, [2, 1280, 1, 64]> var_5324_cast = add(x = out_41_cast, y = var_5323_to_fp16);
            tensor<fp16, [1, 1280, 1, 1]> var_5326_to_fp16 = const()[name = tensor<string, []>("op_5326_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(598527872)))];
            tensor<fp16, [2, 1280, 1, 64]> input_219_cast = mul(x = var_5324_cast, y = var_5326_to_fp16);
            tensor<int32, [2]> var_5334 = const()[name = tensor<string, []>("op_5334"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5336 = const()[name = tensor<string, []>("op_5336"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_5338_pad_type_0 = const()[name = tensor<string, []>("op_5338_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_5338_pad_0 = const()[name = tensor<string, []>("op_5338_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(598530496)))];
            tensor<fp16, [10240]> mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(624744960)))];
            tensor<fp16, [2, 10240, 1, 64]> var_5338_cast = conv(bias = mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_5336, groups = var_4371, pad = var_5338_pad_0, pad_type = var_5338_pad_type_0, strides = var_5334, weight = mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_219_cast);
            tensor<int32, [2]> var_5339_split_sizes_0 = const()[name = tensor<string, []>("op_5339_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_5339_axis_0 = const()[name = tensor<string, []>("op_5339_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 64]> var_5339_cast_0, tensor<fp16, [2, 5120, 1, 64]> var_5339_cast_1 = split(axis = var_5339_axis_0, split_sizes = var_5339_split_sizes_0, x = var_5338_cast);
            tensor<string, []> var_5341_mode_0 = const()[name = tensor<string, []>("op_5341_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 64]> var_5341_cast = gelu(mode = var_5341_mode_0, x = var_5339_cast_1);
            tensor<fp16, [2, 5120, 1, 64]> input_221_cast = mul(x = var_5339_cast_0, y = var_5341_cast);
            tensor<int32, [2]> var_5345 = const()[name = tensor<string, []>("op_5345"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5347 = const()[name = tensor<string, []>("op_5347"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_5349_pad_type_0 = const()[name = tensor<string, []>("op_5349_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_5349_pad_0 = const()[name = tensor<string, []>("op_5349_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> mid_block_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(624765504)))];
            tensor<fp16, [1280]> mid_block_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(637872768)))];
            tensor<fp16, [2, 1280, 1, 64]> var_5349_cast = conv(bias = mid_block_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_5347, groups = var_4371, pad = var_5349_pad_0, pad_type = var_5349_pad_type_0, strides = var_5345, weight = mid_block_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_221_cast);
            tensor<fp16, [2, 1280, 1, 64]> hidden_states_131_cast = add(x = var_5349_cast, y = inputs_41_cast);
            tensor<int32, [4]> var_5351 = const()[name = tensor<string, []>("op_5351"), val = tensor<int32, [4]>([2, 1280, 8, 8])];
            tensor<fp16, [2, 1280, 8, 8]> input_223_cast = reshape(shape = var_5351, x = hidden_states_131_cast);
            tensor<int32, [2]> var_5355 = const()[name = tensor<string, []>("op_5355"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5357 = const()[name = tensor<string, []>("op_5357"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_133_pad_type_0 = const()[name = tensor<string, []>("hidden_states_133_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_133_pad_0 = const()[name = tensor<string, []>("hidden_states_133_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_attentions_0_proj_out_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_proj_out_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(637875392)))];
            tensor<fp16, [1280]> mid_block_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_attentions_0_proj_out_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(641152256)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_133_cast = conv(bias = mid_block_attentions_0_proj_out_bias_to_fp16, dilations = var_5357, groups = var_4371, pad = hidden_states_133_pad_0, pad_type = hidden_states_133_pad_type_0, strides = var_5355, weight = mid_block_attentions_0_proj_out_weight_to_fp16, x = input_223_cast);
            tensor<fp16, [2, 1280, 8, 8]> input_225_cast = add(x = hidden_states_133_cast, y = hidden_states_121_cast);
            tensor<int32, [5]> reshape_100_shape_0 = const()[name = tensor<string, []>("reshape_100_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 8, 8])];
            tensor<fp16, [2, 32, 40, 8, 8]> reshape_100_cast = reshape(shape = reshape_100_shape_0, x = input_225_cast);
            tensor<int32, [3]> reduce_mean_75_axes_0 = const()[name = tensor<string, []>("reduce_mean_75_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_75_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_75_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_75_cast = reduce_mean(axes = reduce_mean_75_axes_0, keep_dims = reduce_mean_75_keep_dims_0, x = reshape_100_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> sub_50_cast = sub(x = reshape_100_cast, y = reduce_mean_75_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> square_25_cast = square(x = sub_50_cast);
            tensor<int32, [3]> reduce_mean_77_axes_0 = const()[name = tensor<string, []>("reduce_mean_77_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_77_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_77_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_77_cast = reduce_mean(axes = reduce_mean_77_axes_0, keep_dims = reduce_mean_77_keep_dims_0, x = square_25_cast);
            tensor<fp16, []> add_50_y_0_to_fp16 = const()[name = tensor<string, []>("add_50_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_50_cast = add(x = reduce_mean_77_cast, y = add_50_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_25_cast = sqrt(x = add_50_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> real_div_25_cast = real_div(x = sub_50_cast, y = sqrt_25_cast);
            tensor<int32, [4]> reshape_101_shape_0 = const()[name = tensor<string, []>("reshape_101_shape_0"), val = tensor<int32, [4]>([2, 1280, 8, 8])];
            tensor<fp16, [2, 1280, 8, 8]> reshape_101_cast = reshape(shape = reshape_101_shape_0, x = real_div_25_cast);
            tensor<fp16, [1280]> add_51_mean_0_to_fp16 = const()[name = tensor<string, []>("add_51_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(641154880)))];
            tensor<fp16, [1280]> add_51_variance_0_to_fp16 = const()[name = tensor<string, []>("add_51_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(641157504)))];
            tensor<fp16, [1280]> add_51_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_51_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(641160128)))];
            tensor<fp16, [1280]> add_51_beta_0_to_fp16 = const()[name = tensor<string, []>("add_51_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(641162752)))];
            tensor<fp16, []> add_51_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_51_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 8, 8]> add_51_cast = batch_norm(beta = add_51_beta_0_to_fp16, epsilon = add_51_epsilon_0_to_fp16, gamma = add_51_gamma_0_to_fp16, mean = add_51_mean_0_to_fp16, variance = add_51_variance_0_to_fp16, x = reshape_101_cast);
            tensor<fp16, [2, 1280, 8, 8]> input_229_cast = silu(x = add_51_cast);
            tensor<int32, [2]> var_5372 = const()[name = tensor<string, []>("op_5372"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5374 = const()[name = tensor<string, []>("op_5374"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_135_pad_type_0 = const()[name = tensor<string, []>("hidden_states_135_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_135_pad_0 = const()[name = tensor<string, []>("hidden_states_135_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> mid_block_resnets_1_conv1_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_resnets_1_conv1_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(641165376)))];
            tensor<fp16, [1280]> mid_block_resnets_1_conv1_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_resnets_1_conv1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(670656640)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_135_cast = conv(bias = mid_block_resnets_1_conv1_bias_to_fp16, dilations = var_5374, groups = var_4371, pad = hidden_states_135_pad_0, pad_type = hidden_states_135_pad_type_0, strides = var_5372, weight = mid_block_resnets_1_conv1_weight_to_fp16, x = input_229_cast);
            tensor<int32, [2]> var_5380 = const()[name = tensor<string, []>("op_5380"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5382 = const()[name = tensor<string, []>("op_5382"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_19_pad_type_0 = const()[name = tensor<string, []>("temb_19_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_19_pad_0 = const()[name = tensor<string, []>("temb_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> mid_block_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(670659264)))];
            tensor<fp16, [1280]> mid_block_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(673936128)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_19_cast = conv(bias = mid_block_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_5382, groups = var_4371, pad = temb_19_pad_0, pad_type = temb_19_pad_type_0, strides = var_5380, weight = mid_block_resnets_1_time_emb_proj_weight_to_fp16, x = input_15_cast);
            tensor<fp16, [2, 1280, 8, 8]> input_233_cast = add(x = hidden_states_135_cast, y = temb_19_cast);
            tensor<int32, [5]> reshape_104_shape_0 = const()[name = tensor<string, []>("reshape_104_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 8, 8])];
            tensor<fp16, [2, 32, 40, 8, 8]> reshape_104_cast = reshape(shape = reshape_104_shape_0, x = input_233_cast);
            tensor<int32, [3]> reduce_mean_78_axes_0 = const()[name = tensor<string, []>("reduce_mean_78_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_78_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_78_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_78_cast = reduce_mean(axes = reduce_mean_78_axes_0, keep_dims = reduce_mean_78_keep_dims_0, x = reshape_104_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> sub_52_cast = sub(x = reshape_104_cast, y = reduce_mean_78_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> square_26_cast = square(x = sub_52_cast);
            tensor<int32, [3]> reduce_mean_80_axes_0 = const()[name = tensor<string, []>("reduce_mean_80_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_80_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_80_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_80_cast = reduce_mean(axes = reduce_mean_80_axes_0, keep_dims = reduce_mean_80_keep_dims_0, x = square_26_cast);
            tensor<fp16, []> add_52_y_0_to_fp16 = const()[name = tensor<string, []>("add_52_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_52_cast = add(x = reduce_mean_80_cast, y = add_52_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_26_cast = sqrt(x = add_52_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> real_div_26_cast = real_div(x = sub_52_cast, y = sqrt_26_cast);
            tensor<int32, [4]> reshape_105_shape_0 = const()[name = tensor<string, []>("reshape_105_shape_0"), val = tensor<int32, [4]>([2, 1280, 8, 8])];
            tensor<fp16, [2, 1280, 8, 8]> reshape_105_cast = reshape(shape = reshape_105_shape_0, x = real_div_26_cast);
            tensor<fp16, [1280]> add_53_mean_0_to_fp16 = const()[name = tensor<string, []>("add_53_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(673938752)))];
            tensor<fp16, [1280]> add_53_variance_0_to_fp16 = const()[name = tensor<string, []>("add_53_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(673941376)))];
            tensor<fp16, [1280]> add_53_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_53_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(673944000)))];
            tensor<fp16, [1280]> add_53_beta_0_to_fp16 = const()[name = tensor<string, []>("add_53_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(673946624)))];
            tensor<fp16, []> add_53_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_53_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 8, 8]> add_53_cast = batch_norm(beta = add_53_beta_0_to_fp16, epsilon = add_53_epsilon_0_to_fp16, gamma = add_53_gamma_0_to_fp16, mean = add_53_mean_0_to_fp16, variance = add_53_variance_0_to_fp16, x = reshape_105_cast);
            tensor<fp16, [2, 1280, 8, 8]> input_237_cast = silu(x = add_53_cast);
            tensor<int32, [2]> var_5392 = const()[name = tensor<string, []>("op_5392"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5394 = const()[name = tensor<string, []>("op_5394"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_137_pad_type_0 = const()[name = tensor<string, []>("hidden_states_137_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_137_pad_0 = const()[name = tensor<string, []>("hidden_states_137_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> mid_block_resnets_1_conv2_weight_to_fp16 = const()[name = tensor<string, []>("mid_block_resnets_1_conv2_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(673949248)))];
            tensor<fp16, [1280]> mid_block_resnets_1_conv2_bias_to_fp16 = const()[name = tensor<string, []>("mid_block_resnets_1_conv2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(703440512)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_137_cast = conv(bias = mid_block_resnets_1_conv2_bias_to_fp16, dilations = var_5394, groups = var_4371, pad = hidden_states_137_pad_0, pad_type = hidden_states_137_pad_type_0, strides = var_5392, weight = mid_block_resnets_1_conv2_weight_to_fp16, x = input_237_cast);
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_139_cast = add(x = input_225_cast, y = hidden_states_137_cast);
            tensor<int32, []> var_5405 = const()[name = tensor<string, []>("op_5405"), val = tensor<int32, []>(1)];
            tensor<bool, []> input_239_interleave_0 = const()[name = tensor<string, []>("input_239_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 2560, 8, 8]> input_239_cast = concat(axis = var_5405, interleave = input_239_interleave_0, values = (hidden_states_139_cast, input_199_cast));
            tensor<int32, [5]> reshape_108_shape_0 = const()[name = tensor<string, []>("reshape_108_shape_0"), val = tensor<int32, [5]>([2, 32, 80, 8, 8])];
            tensor<fp16, [2, 32, 80, 8, 8]> reshape_108_cast = reshape(shape = reshape_108_shape_0, x = input_239_cast);
            tensor<int32, [3]> reduce_mean_81_axes_0 = const()[name = tensor<string, []>("reduce_mean_81_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_81_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_81_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_81_cast = reduce_mean(axes = reduce_mean_81_axes_0, keep_dims = reduce_mean_81_keep_dims_0, x = reshape_108_cast);
            tensor<fp16, [2, 32, 80, 8, 8]> sub_54_cast = sub(x = reshape_108_cast, y = reduce_mean_81_cast);
            tensor<fp16, [2, 32, 80, 8, 8]> square_27_cast = square(x = sub_54_cast);
            tensor<int32, [3]> reduce_mean_83_axes_0 = const()[name = tensor<string, []>("reduce_mean_83_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_83_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_83_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_83_cast = reduce_mean(axes = reduce_mean_83_axes_0, keep_dims = reduce_mean_83_keep_dims_0, x = square_27_cast);
            tensor<fp16, []> add_54_y_0_to_fp16 = const()[name = tensor<string, []>("add_54_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_54_cast = add(x = reduce_mean_83_cast, y = add_54_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_27_cast = sqrt(x = add_54_cast);
            tensor<fp16, [2, 32, 80, 8, 8]> real_div_27_cast = real_div(x = sub_54_cast, y = sqrt_27_cast);
            tensor<int32, [4]> reshape_109_shape_0 = const()[name = tensor<string, []>("reshape_109_shape_0"), val = tensor<int32, [4]>([2, 2560, 8, 8])];
            tensor<fp16, [2, 2560, 8, 8]> reshape_109_cast = reshape(shape = reshape_109_shape_0, x = real_div_27_cast);
            tensor<fp16, [2560]> add_55_mean_0_to_fp16 = const()[name = tensor<string, []>("add_55_mean_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(703443136)))];
            tensor<fp16, [2560]> add_55_variance_0_to_fp16 = const()[name = tensor<string, []>("add_55_variance_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(703448320)))];
            tensor<fp16, [2560]> add_55_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_55_gamma_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(703453504)))];
            tensor<fp16, [2560]> add_55_beta_0_to_fp16 = const()[name = tensor<string, []>("add_55_beta_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(703458688)))];
            tensor<fp16, []> add_55_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_55_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 2560, 8, 8]> add_55_cast = batch_norm(beta = add_55_beta_0_to_fp16, epsilon = add_55_epsilon_0_to_fp16, gamma = add_55_gamma_0_to_fp16, mean = add_55_mean_0_to_fp16, variance = add_55_variance_0_to_fp16, x = reshape_109_cast);
            tensor<fp16, [2, 2560, 8, 8]> input_243_cast = silu(x = add_55_cast);
            tensor<int32, [2]> var_5428 = const()[name = tensor<string, []>("op_5428"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5430 = const()[name = tensor<string, []>("op_5430"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_141_pad_type_0 = const()[name = tensor<string, []>("hidden_states_141_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_141_pad_0 = const()[name = tensor<string, []>("hidden_states_141_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 2560, 3, 3]> up_blocks_0_resnets_0_conv1_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_0_conv1_weight_to_fp16"), val = tensor<fp16, [1280, 2560, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(703463872)))];
            tensor<fp16, [1280]> up_blocks_0_resnets_0_conv1_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_0_conv1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(762446336)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_141_cast = conv(bias = up_blocks_0_resnets_0_conv1_bias_to_fp16, dilations = var_5430, groups = var_5405, pad = hidden_states_141_pad_0, pad_type = hidden_states_141_pad_type_0, strides = var_5428, weight = up_blocks_0_resnets_0_conv1_weight_to_fp16, x = input_243_cast);
            tensor<int32, [2]> var_5436 = const()[name = tensor<string, []>("op_5436"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5438 = const()[name = tensor<string, []>("op_5438"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_21_pad_type_0 = const()[name = tensor<string, []>("temb_21_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_21_pad_0 = const()[name = tensor<string, []>("temb_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(762448960)))];
            tensor<fp16, [1280]> up_blocks_0_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(765725824)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_21_cast = conv(bias = up_blocks_0_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_5438, groups = var_5405, pad = temb_21_pad_0, pad_type = temb_21_pad_type_0, strides = var_5436, weight = up_blocks_0_resnets_0_time_emb_proj_weight_to_fp16, x = input_15_cast);
            tensor<fp16, [2, 1280, 8, 8]> input_247_cast = add(x = hidden_states_141_cast, y = temb_21_cast);
            tensor<int32, [5]> reshape_112_shape_0 = const()[name = tensor<string, []>("reshape_112_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 8, 8])];
            tensor<fp16, [2, 32, 40, 8, 8]> reshape_112_cast = reshape(shape = reshape_112_shape_0, x = input_247_cast);
            tensor<int32, [3]> reduce_mean_84_axes_0 = const()[name = tensor<string, []>("reduce_mean_84_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_84_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_84_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_84_cast = reduce_mean(axes = reduce_mean_84_axes_0, keep_dims = reduce_mean_84_keep_dims_0, x = reshape_112_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> sub_56_cast = sub(x = reshape_112_cast, y = reduce_mean_84_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> square_28_cast = square(x = sub_56_cast);
            tensor<int32, [3]> reduce_mean_86_axes_0 = const()[name = tensor<string, []>("reduce_mean_86_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_86_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_86_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_86_cast = reduce_mean(axes = reduce_mean_86_axes_0, keep_dims = reduce_mean_86_keep_dims_0, x = square_28_cast);
            tensor<fp16, []> add_56_y_0_to_fp16 = const()[name = tensor<string, []>("add_56_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_56_cast = add(x = reduce_mean_86_cast, y = add_56_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_28_cast = sqrt(x = add_56_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> real_div_28_cast = real_div(x = sub_56_cast, y = sqrt_28_cast);
            tensor<int32, [4]> reshape_113_shape_0 = const()[name = tensor<string, []>("reshape_113_shape_0"), val = tensor<int32, [4]>([2, 1280, 8, 8])];
            tensor<fp16, [2, 1280, 8, 8]> reshape_113_cast = reshape(shape = reshape_113_shape_0, x = real_div_28_cast);
            tensor<fp16, [1280]> add_57_mean_0_to_fp16 = const()[name = tensor<string, []>("add_57_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(765728448)))];
            tensor<fp16, [1280]> add_57_variance_0_to_fp16 = const()[name = tensor<string, []>("add_57_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(765731072)))];
            tensor<fp16, [1280]> add_57_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_57_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(765733696)))];
            tensor<fp16, [1280]> add_57_beta_0_to_fp16 = const()[name = tensor<string, []>("add_57_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(765736320)))];
            tensor<fp16, []> add_57_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_57_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 8, 8]> add_57_cast = batch_norm(beta = add_57_beta_0_to_fp16, epsilon = add_57_epsilon_0_to_fp16, gamma = add_57_gamma_0_to_fp16, mean = add_57_mean_0_to_fp16, variance = add_57_variance_0_to_fp16, x = reshape_113_cast);
            tensor<fp16, [2, 1280, 8, 8]> input_251_cast = silu(x = add_57_cast);
            tensor<int32, [2]> var_5448 = const()[name = tensor<string, []>("op_5448"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5450 = const()[name = tensor<string, []>("op_5450"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_143_pad_type_0 = const()[name = tensor<string, []>("hidden_states_143_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_143_pad_0 = const()[name = tensor<string, []>("hidden_states_143_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> up_blocks_0_resnets_0_conv2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_0_conv2_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(765738944)))];
            tensor<fp16, [1280]> up_blocks_0_resnets_0_conv2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_0_conv2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(795230208)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_143_cast = conv(bias = up_blocks_0_resnets_0_conv2_bias_to_fp16, dilations = var_5450, groups = var_5405, pad = hidden_states_143_pad_0, pad_type = hidden_states_143_pad_type_0, strides = var_5448, weight = up_blocks_0_resnets_0_conv2_weight_to_fp16, x = input_251_cast);
            tensor<int32, [2]> var_5455 = const()[name = tensor<string, []>("op_5455"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5457 = const()[name = tensor<string, []>("op_5457"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> x_5_pad_type_0 = const()[name = tensor<string, []>("x_5_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> x_5_pad_0 = const()[name = tensor<string, []>("x_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2560, 1, 1]> up_blocks_0_resnets_0_conv_shortcut_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_0_conv_shortcut_weight_to_fp16"), val = tensor<fp16, [1280, 2560, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(795232832)))];
            tensor<fp16, [1280]> up_blocks_0_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(801786496)))];
            tensor<fp16, [2, 1280, 8, 8]> x_5_cast = conv(bias = up_blocks_0_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_5457, groups = var_5405, pad = x_5_pad_0, pad_type = x_5_pad_type_0, strides = var_5455, weight = up_blocks_0_resnets_0_conv_shortcut_weight_to_fp16, x = input_239_cast);
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_145_cast = add(x = x_5_cast, y = hidden_states_143_cast);
            tensor<bool, []> input_253_interleave_0 = const()[name = tensor<string, []>("input_253_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 2560, 8, 8]> input_253_cast = concat(axis = var_5405, interleave = input_253_interleave_0, values = (hidden_states_145_cast, input_185_cast));
            tensor<int32, [5]> reshape_116_shape_0 = const()[name = tensor<string, []>("reshape_116_shape_0"), val = tensor<int32, [5]>([2, 32, 80, 8, 8])];
            tensor<fp16, [2, 32, 80, 8, 8]> reshape_116_cast = reshape(shape = reshape_116_shape_0, x = input_253_cast);
            tensor<int32, [3]> reduce_mean_87_axes_0 = const()[name = tensor<string, []>("reduce_mean_87_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_87_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_87_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_87_cast = reduce_mean(axes = reduce_mean_87_axes_0, keep_dims = reduce_mean_87_keep_dims_0, x = reshape_116_cast);
            tensor<fp16, [2, 32, 80, 8, 8]> sub_58_cast = sub(x = reshape_116_cast, y = reduce_mean_87_cast);
            tensor<fp16, [2, 32, 80, 8, 8]> square_29_cast = square(x = sub_58_cast);
            tensor<int32, [3]> reduce_mean_89_axes_0 = const()[name = tensor<string, []>("reduce_mean_89_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_89_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_89_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_89_cast = reduce_mean(axes = reduce_mean_89_axes_0, keep_dims = reduce_mean_89_keep_dims_0, x = square_29_cast);
            tensor<fp16, []> add_58_y_0_to_fp16 = const()[name = tensor<string, []>("add_58_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_58_cast = add(x = reduce_mean_89_cast, y = add_58_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_29_cast = sqrt(x = add_58_cast);
            tensor<fp16, [2, 32, 80, 8, 8]> real_div_29_cast = real_div(x = sub_58_cast, y = sqrt_29_cast);
            tensor<int32, [4]> reshape_117_shape_0 = const()[name = tensor<string, []>("reshape_117_shape_0"), val = tensor<int32, [4]>([2, 2560, 8, 8])];
            tensor<fp16, [2, 2560, 8, 8]> reshape_117_cast = reshape(shape = reshape_117_shape_0, x = real_div_29_cast);
            tensor<fp16, [2560]> add_59_mean_0_to_fp16 = const()[name = tensor<string, []>("add_59_mean_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(801789120)))];
            tensor<fp16, [2560]> add_59_variance_0_to_fp16 = const()[name = tensor<string, []>("add_59_variance_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(801794304)))];
            tensor<fp16, [2560]> add_59_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_59_gamma_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(801799488)))];
            tensor<fp16, [2560]> add_59_beta_0_to_fp16 = const()[name = tensor<string, []>("add_59_beta_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(801804672)))];
            tensor<fp16, []> add_59_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_59_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 2560, 8, 8]> add_59_cast = batch_norm(beta = add_59_beta_0_to_fp16, epsilon = add_59_epsilon_0_to_fp16, gamma = add_59_gamma_0_to_fp16, mean = add_59_mean_0_to_fp16, variance = add_59_variance_0_to_fp16, x = reshape_117_cast);
            tensor<fp16, [2, 2560, 8, 8]> input_257_cast = silu(x = add_59_cast);
            tensor<int32, [2]> var_5475 = const()[name = tensor<string, []>("op_5475"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5477 = const()[name = tensor<string, []>("op_5477"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_147_pad_type_0 = const()[name = tensor<string, []>("hidden_states_147_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_147_pad_0 = const()[name = tensor<string, []>("hidden_states_147_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 2560, 3, 3]> up_blocks_0_resnets_1_conv1_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_1_conv1_weight_to_fp16"), val = tensor<fp16, [1280, 2560, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(801809856)))];
            tensor<fp16, [1280]> up_blocks_0_resnets_1_conv1_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_1_conv1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(860792320)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_147_cast = conv(bias = up_blocks_0_resnets_1_conv1_bias_to_fp16, dilations = var_5477, groups = var_5405, pad = hidden_states_147_pad_0, pad_type = hidden_states_147_pad_type_0, strides = var_5475, weight = up_blocks_0_resnets_1_conv1_weight_to_fp16, x = input_257_cast);
            tensor<int32, [2]> var_5483 = const()[name = tensor<string, []>("op_5483"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5485 = const()[name = tensor<string, []>("op_5485"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_23_pad_type_0 = const()[name = tensor<string, []>("temb_23_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_23_pad_0 = const()[name = tensor<string, []>("temb_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(860794944)))];
            tensor<fp16, [1280]> up_blocks_0_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(864071808)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_23_cast = conv(bias = up_blocks_0_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_5485, groups = var_5405, pad = temb_23_pad_0, pad_type = temb_23_pad_type_0, strides = var_5483, weight = up_blocks_0_resnets_1_time_emb_proj_weight_to_fp16, x = input_15_cast);
            tensor<fp16, [2, 1280, 8, 8]> input_261_cast = add(x = hidden_states_147_cast, y = temb_23_cast);
            tensor<int32, [5]> reshape_120_shape_0 = const()[name = tensor<string, []>("reshape_120_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 8, 8])];
            tensor<fp16, [2, 32, 40, 8, 8]> reshape_120_cast = reshape(shape = reshape_120_shape_0, x = input_261_cast);
            tensor<int32, [3]> reduce_mean_90_axes_0 = const()[name = tensor<string, []>("reduce_mean_90_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_90_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_90_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_90_cast = reduce_mean(axes = reduce_mean_90_axes_0, keep_dims = reduce_mean_90_keep_dims_0, x = reshape_120_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> sub_60_cast = sub(x = reshape_120_cast, y = reduce_mean_90_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> square_30_cast = square(x = sub_60_cast);
            tensor<int32, [3]> reduce_mean_92_axes_0 = const()[name = tensor<string, []>("reduce_mean_92_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_92_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_92_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_92_cast = reduce_mean(axes = reduce_mean_92_axes_0, keep_dims = reduce_mean_92_keep_dims_0, x = square_30_cast);
            tensor<fp16, []> add_60_y_0_to_fp16 = const()[name = tensor<string, []>("add_60_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_60_cast = add(x = reduce_mean_92_cast, y = add_60_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_30_cast = sqrt(x = add_60_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> real_div_30_cast = real_div(x = sub_60_cast, y = sqrt_30_cast);
            tensor<int32, [4]> reshape_121_shape_0 = const()[name = tensor<string, []>("reshape_121_shape_0"), val = tensor<int32, [4]>([2, 1280, 8, 8])];
            tensor<fp16, [2, 1280, 8, 8]> reshape_121_cast = reshape(shape = reshape_121_shape_0, x = real_div_30_cast);
            tensor<fp16, [1280]> add_61_mean_0_to_fp16 = const()[name = tensor<string, []>("add_61_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(864074432)))];
            tensor<fp16, [1280]> add_61_variance_0_to_fp16 = const()[name = tensor<string, []>("add_61_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(864077056)))];
            tensor<fp16, [1280]> add_61_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_61_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(864079680)))];
            tensor<fp16, [1280]> add_61_beta_0_to_fp16 = const()[name = tensor<string, []>("add_61_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(864082304)))];
            tensor<fp16, []> add_61_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_61_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 8, 8]> add_61_cast = batch_norm(beta = add_61_beta_0_to_fp16, epsilon = add_61_epsilon_0_to_fp16, gamma = add_61_gamma_0_to_fp16, mean = add_61_mean_0_to_fp16, variance = add_61_variance_0_to_fp16, x = reshape_121_cast);
            tensor<fp16, [2, 1280, 8, 8]> input_265_cast = silu(x = add_61_cast);
            tensor<int32, [2]> var_5495 = const()[name = tensor<string, []>("op_5495"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5497 = const()[name = tensor<string, []>("op_5497"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_149_pad_type_0 = const()[name = tensor<string, []>("hidden_states_149_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_149_pad_0 = const()[name = tensor<string, []>("hidden_states_149_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> up_blocks_0_resnets_1_conv2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_1_conv2_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(864084928)))];
            tensor<fp16, [1280]> up_blocks_0_resnets_1_conv2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_1_conv2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(893576192)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_149_cast = conv(bias = up_blocks_0_resnets_1_conv2_bias_to_fp16, dilations = var_5497, groups = var_5405, pad = hidden_states_149_pad_0, pad_type = hidden_states_149_pad_type_0, strides = var_5495, weight = up_blocks_0_resnets_1_conv2_weight_to_fp16, x = input_265_cast);
            tensor<int32, [2]> var_5502 = const()[name = tensor<string, []>("op_5502"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5504 = const()[name = tensor<string, []>("op_5504"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> x_7_pad_type_0 = const()[name = tensor<string, []>("x_7_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> x_7_pad_0 = const()[name = tensor<string, []>("x_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2560, 1, 1]> up_blocks_0_resnets_1_conv_shortcut_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_1_conv_shortcut_weight_to_fp16"), val = tensor<fp16, [1280, 2560, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(893578816)))];
            tensor<fp16, [1280]> up_blocks_0_resnets_1_conv_shortcut_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_1_conv_shortcut_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(900132480)))];
            tensor<fp16, [2, 1280, 8, 8]> x_7_cast = conv(bias = up_blocks_0_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_5504, groups = var_5405, pad = x_7_pad_0, pad_type = x_7_pad_type_0, strides = var_5502, weight = up_blocks_0_resnets_1_conv_shortcut_weight_to_fp16, x = input_253_cast);
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_151_cast = add(x = x_7_cast, y = hidden_states_149_cast);
            tensor<bool, []> input_267_interleave_0 = const()[name = tensor<string, []>("input_267_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 2560, 8, 8]> input_267_cast = concat(axis = var_5405, interleave = input_267_interleave_0, values = (hidden_states_151_cast, input_171_cast));
            tensor<int32, [5]> reshape_124_shape_0 = const()[name = tensor<string, []>("reshape_124_shape_0"), val = tensor<int32, [5]>([2, 32, 80, 8, 8])];
            tensor<fp16, [2, 32, 80, 8, 8]> reshape_124_cast = reshape(shape = reshape_124_shape_0, x = input_267_cast);
            tensor<int32, [3]> reduce_mean_93_axes_0 = const()[name = tensor<string, []>("reduce_mean_93_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_93_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_93_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_93_cast = reduce_mean(axes = reduce_mean_93_axes_0, keep_dims = reduce_mean_93_keep_dims_0, x = reshape_124_cast);
            tensor<fp16, [2, 32, 80, 8, 8]> sub_62_cast = sub(x = reshape_124_cast, y = reduce_mean_93_cast);
            tensor<fp16, [2, 32, 80, 8, 8]> square_31_cast = square(x = sub_62_cast);
            tensor<int32, [3]> reduce_mean_95_axes_0 = const()[name = tensor<string, []>("reduce_mean_95_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_95_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_95_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_95_cast = reduce_mean(axes = reduce_mean_95_axes_0, keep_dims = reduce_mean_95_keep_dims_0, x = square_31_cast);
            tensor<fp16, []> add_62_y_0_to_fp16 = const()[name = tensor<string, []>("add_62_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_62_cast = add(x = reduce_mean_95_cast, y = add_62_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_31_cast = sqrt(x = add_62_cast);
            tensor<fp16, [2, 32, 80, 8, 8]> real_div_31_cast = real_div(x = sub_62_cast, y = sqrt_31_cast);
            tensor<int32, [4]> reshape_125_shape_0 = const()[name = tensor<string, []>("reshape_125_shape_0"), val = tensor<int32, [4]>([2, 2560, 8, 8])];
            tensor<fp16, [2, 2560, 8, 8]> reshape_125_cast = reshape(shape = reshape_125_shape_0, x = real_div_31_cast);
            tensor<fp16, [2560]> add_63_mean_0_to_fp16 = const()[name = tensor<string, []>("add_63_mean_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(900135104)))];
            tensor<fp16, [2560]> add_63_variance_0_to_fp16 = const()[name = tensor<string, []>("add_63_variance_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(900140288)))];
            tensor<fp16, [2560]> add_63_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_63_gamma_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(900145472)))];
            tensor<fp16, [2560]> add_63_beta_0_to_fp16 = const()[name = tensor<string, []>("add_63_beta_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(900150656)))];
            tensor<fp16, []> add_63_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_63_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 2560, 8, 8]> add_63_cast = batch_norm(beta = add_63_beta_0_to_fp16, epsilon = add_63_epsilon_0_to_fp16, gamma = add_63_gamma_0_to_fp16, mean = add_63_mean_0_to_fp16, variance = add_63_variance_0_to_fp16, x = reshape_125_cast);
            tensor<fp16, [2, 2560, 8, 8]> input_271_cast = silu(x = add_63_cast);
            tensor<int32, [2]> var_5522 = const()[name = tensor<string, []>("op_5522"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5524 = const()[name = tensor<string, []>("op_5524"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_153_pad_type_0 = const()[name = tensor<string, []>("hidden_states_153_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_153_pad_0 = const()[name = tensor<string, []>("hidden_states_153_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 2560, 3, 3]> up_blocks_0_resnets_2_conv1_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_2_conv1_weight_to_fp16"), val = tensor<fp16, [1280, 2560, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(900155840)))];
            tensor<fp16, [1280]> up_blocks_0_resnets_2_conv1_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_2_conv1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(959138304)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_153_cast = conv(bias = up_blocks_0_resnets_2_conv1_bias_to_fp16, dilations = var_5524, groups = var_5405, pad = hidden_states_153_pad_0, pad_type = hidden_states_153_pad_type_0, strides = var_5522, weight = up_blocks_0_resnets_2_conv1_weight_to_fp16, x = input_271_cast);
            tensor<int32, [2]> var_5530 = const()[name = tensor<string, []>("op_5530"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5532 = const()[name = tensor<string, []>("op_5532"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_25_pad_type_0 = const()[name = tensor<string, []>("temb_25_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_25_pad_0 = const()[name = tensor<string, []>("temb_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_0_resnets_2_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_2_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(959140928)))];
            tensor<fp16, [1280]> up_blocks_0_resnets_2_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_2_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(962417792)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_25_cast = conv(bias = up_blocks_0_resnets_2_time_emb_proj_bias_to_fp16, dilations = var_5532, groups = var_5405, pad = temb_25_pad_0, pad_type = temb_25_pad_type_0, strides = var_5530, weight = up_blocks_0_resnets_2_time_emb_proj_weight_to_fp16, x = input_15_cast);
            tensor<fp16, [2, 1280, 8, 8]> input_275_cast = add(x = hidden_states_153_cast, y = temb_25_cast);
            tensor<int32, [5]> reshape_128_shape_0 = const()[name = tensor<string, []>("reshape_128_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 8, 8])];
            tensor<fp16, [2, 32, 40, 8, 8]> reshape_128_cast = reshape(shape = reshape_128_shape_0, x = input_275_cast);
            tensor<int32, [3]> reduce_mean_96_axes_0 = const()[name = tensor<string, []>("reduce_mean_96_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_96_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_96_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_96_cast = reduce_mean(axes = reduce_mean_96_axes_0, keep_dims = reduce_mean_96_keep_dims_0, x = reshape_128_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> sub_64_cast = sub(x = reshape_128_cast, y = reduce_mean_96_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> square_32_cast = square(x = sub_64_cast);
            tensor<int32, [3]> reduce_mean_98_axes_0 = const()[name = tensor<string, []>("reduce_mean_98_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_98_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_98_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_98_cast = reduce_mean(axes = reduce_mean_98_axes_0, keep_dims = reduce_mean_98_keep_dims_0, x = square_32_cast);
            tensor<fp16, []> add_64_y_0_to_fp16 = const()[name = tensor<string, []>("add_64_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_64_cast = add(x = reduce_mean_98_cast, y = add_64_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_32_cast = sqrt(x = add_64_cast);
            tensor<fp16, [2, 32, 40, 8, 8]> real_div_32_cast = real_div(x = sub_64_cast, y = sqrt_32_cast);
            tensor<int32, [4]> reshape_129_shape_0 = const()[name = tensor<string, []>("reshape_129_shape_0"), val = tensor<int32, [4]>([2, 1280, 8, 8])];
            tensor<fp16, [2, 1280, 8, 8]> reshape_129_cast = reshape(shape = reshape_129_shape_0, x = real_div_32_cast);
            tensor<fp16, [1280]> add_65_mean_0_to_fp16 = const()[name = tensor<string, []>("add_65_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(962420416)))];
            tensor<fp16, [1280]> add_65_variance_0_to_fp16 = const()[name = tensor<string, []>("add_65_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(962423040)))];
            tensor<fp16, [1280]> add_65_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_65_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(962425664)))];
            tensor<fp16, [1280]> add_65_beta_0_to_fp16 = const()[name = tensor<string, []>("add_65_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(962428288)))];
            tensor<fp16, []> add_65_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_65_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 8, 8]> add_65_cast = batch_norm(beta = add_65_beta_0_to_fp16, epsilon = add_65_epsilon_0_to_fp16, gamma = add_65_gamma_0_to_fp16, mean = add_65_mean_0_to_fp16, variance = add_65_variance_0_to_fp16, x = reshape_129_cast);
            tensor<fp16, [2, 1280, 8, 8]> input_279_cast = silu(x = add_65_cast);
            tensor<int32, [2]> var_5542 = const()[name = tensor<string, []>("op_5542"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5544 = const()[name = tensor<string, []>("op_5544"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_155_pad_type_0 = const()[name = tensor<string, []>("hidden_states_155_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_155_pad_0 = const()[name = tensor<string, []>("hidden_states_155_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> up_blocks_0_resnets_2_conv2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_2_conv2_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(962430912)))];
            tensor<fp16, [1280]> up_blocks_0_resnets_2_conv2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_2_conv2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(991922176)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_155_cast = conv(bias = up_blocks_0_resnets_2_conv2_bias_to_fp16, dilations = var_5544, groups = var_5405, pad = hidden_states_155_pad_0, pad_type = hidden_states_155_pad_type_0, strides = var_5542, weight = up_blocks_0_resnets_2_conv2_weight_to_fp16, x = input_279_cast);
            tensor<int32, [2]> var_5549 = const()[name = tensor<string, []>("op_5549"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5551 = const()[name = tensor<string, []>("op_5551"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> x_9_pad_type_0 = const()[name = tensor<string, []>("x_9_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> x_9_pad_0 = const()[name = tensor<string, []>("x_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2560, 1, 1]> up_blocks_0_resnets_2_conv_shortcut_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_2_conv_shortcut_weight_to_fp16"), val = tensor<fp16, [1280, 2560, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(991924800)))];
            tensor<fp16, [1280]> up_blocks_0_resnets_2_conv_shortcut_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_resnets_2_conv_shortcut_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(998478464)))];
            tensor<fp16, [2, 1280, 8, 8]> x_9_cast = conv(bias = up_blocks_0_resnets_2_conv_shortcut_bias_to_fp16, dilations = var_5551, groups = var_5405, pad = x_9_pad_0, pad_type = x_9_pad_type_0, strides = var_5549, weight = up_blocks_0_resnets_2_conv_shortcut_weight_to_fp16, x = input_267_cast);
            tensor<fp16, [2, 1280, 8, 8]> input_281_cast = add(x = x_9_cast, y = hidden_states_155_cast);
            tensor<fp32, []> input_283_scale_factor_height_0 = const()[name = tensor<string, []>("input_283_scale_factor_height_0"), val = tensor<fp32, []>(0x1p+1)];
            tensor<fp32, []> input_283_scale_factor_width_0 = const()[name = tensor<string, []>("input_283_scale_factor_width_0"), val = tensor<fp32, []>(0x1p+1)];
            tensor<fp16, [2, 1280, 16, 16]> input_283_cast = upsample_nearest_neighbor(scale_factor_height = input_283_scale_factor_height_0, scale_factor_width = input_283_scale_factor_width_0, x = input_281_cast);
            tensor<int32, [2]> var_5560 = const()[name = tensor<string, []>("op_5560"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5562 = const()[name = tensor<string, []>("op_5562"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_157_pad_type_0 = const()[name = tensor<string, []>("hidden_states_157_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_157_pad_0 = const()[name = tensor<string, []>("hidden_states_157_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> up_blocks_0_upsamplers_0_conv_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_upsamplers_0_conv_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(998481088)))];
            tensor<fp16, [1280]> up_blocks_0_upsamplers_0_conv_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_0_upsamplers_0_conv_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1027972352)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_157_cast = conv(bias = up_blocks_0_upsamplers_0_conv_bias_to_fp16, dilations = var_5562, groups = var_5405, pad = hidden_states_157_pad_0, pad_type = hidden_states_157_pad_type_0, strides = var_5560, weight = up_blocks_0_upsamplers_0_conv_weight_to_fp16, x = input_283_cast);
            tensor<bool, []> var_5594 = const()[name = tensor<string, []>("op_5594"), val = tensor<bool, []>(true)];
            tensor<int32, []> var_5599 = const()[name = tensor<string, []>("op_5599"), val = tensor<int32, []>(1)];
            tensor<bool, []> input_285_interleave_0 = const()[name = tensor<string, []>("input_285_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 2560, 16, 16]> input_285_cast = concat(axis = var_5599, interleave = input_285_interleave_0, values = (hidden_states_157_cast, input_169_cast));
            tensor<int32, [5]> reshape_132_shape_0 = const()[name = tensor<string, []>("reshape_132_shape_0"), val = tensor<int32, [5]>([2, 32, 80, 16, 16])];
            tensor<fp16, [2, 32, 80, 16, 16]> reshape_132_cast = reshape(shape = reshape_132_shape_0, x = input_285_cast);
            tensor<int32, [3]> reduce_mean_99_axes_0 = const()[name = tensor<string, []>("reduce_mean_99_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_99_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_99_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_99_cast = reduce_mean(axes = reduce_mean_99_axes_0, keep_dims = reduce_mean_99_keep_dims_0, x = reshape_132_cast);
            tensor<fp16, [2, 32, 80, 16, 16]> sub_66_cast = sub(x = reshape_132_cast, y = reduce_mean_99_cast);
            tensor<fp16, [2, 32, 80, 16, 16]> square_33_cast = square(x = sub_66_cast);
            tensor<int32, [3]> reduce_mean_101_axes_0 = const()[name = tensor<string, []>("reduce_mean_101_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_101_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_101_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_101_cast = reduce_mean(axes = reduce_mean_101_axes_0, keep_dims = reduce_mean_101_keep_dims_0, x = square_33_cast);
            tensor<fp16, []> add_66_y_0_to_fp16 = const()[name = tensor<string, []>("add_66_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_66_cast = add(x = reduce_mean_101_cast, y = add_66_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_33_cast = sqrt(x = add_66_cast);
            tensor<fp16, [2, 32, 80, 16, 16]> real_div_33_cast = real_div(x = sub_66_cast, y = sqrt_33_cast);
            tensor<int32, [4]> reshape_133_shape_0 = const()[name = tensor<string, []>("reshape_133_shape_0"), val = tensor<int32, [4]>([2, 2560, 16, 16])];
            tensor<fp16, [2, 2560, 16, 16]> reshape_133_cast = reshape(shape = reshape_133_shape_0, x = real_div_33_cast);
            tensor<fp16, [2560]> add_67_mean_0_to_fp16 = const()[name = tensor<string, []>("add_67_mean_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1027974976)))];
            tensor<fp16, [2560]> add_67_variance_0_to_fp16 = const()[name = tensor<string, []>("add_67_variance_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1027980160)))];
            tensor<fp16, [2560]> add_67_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_67_gamma_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1027985344)))];
            tensor<fp16, [2560]> add_67_beta_0_to_fp16 = const()[name = tensor<string, []>("add_67_beta_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1027990528)))];
            tensor<fp16, []> add_67_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_67_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 2560, 16, 16]> add_67_cast = batch_norm(beta = add_67_beta_0_to_fp16, epsilon = add_67_epsilon_0_to_fp16, gamma = add_67_gamma_0_to_fp16, mean = add_67_mean_0_to_fp16, variance = add_67_variance_0_to_fp16, x = reshape_133_cast);
            tensor<fp16, [2, 2560, 16, 16]> input_289_cast = silu(x = add_67_cast);
            tensor<int32, [2]> var_5628 = const()[name = tensor<string, []>("op_5628"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5630 = const()[name = tensor<string, []>("op_5630"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_159_pad_type_0 = const()[name = tensor<string, []>("hidden_states_159_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_159_pad_0 = const()[name = tensor<string, []>("hidden_states_159_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 2560, 3, 3]> up_blocks_1_resnets_0_conv1_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_0_conv1_weight_to_fp16"), val = tensor<fp16, [1280, 2560, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1027995712)))];
            tensor<fp16, [1280]> up_blocks_1_resnets_0_conv1_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_0_conv1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1086978176)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_159_cast = conv(bias = up_blocks_1_resnets_0_conv1_bias_to_fp16, dilations = var_5630, groups = var_5599, pad = hidden_states_159_pad_0, pad_type = hidden_states_159_pad_type_0, strides = var_5628, weight = up_blocks_1_resnets_0_conv1_weight_to_fp16, x = input_289_cast);
            tensor<int32, [2]> var_5636 = const()[name = tensor<string, []>("op_5636"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5638 = const()[name = tensor<string, []>("op_5638"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_27_pad_type_0 = const()[name = tensor<string, []>("temb_27_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_27_pad_0 = const()[name = tensor<string, []>("temb_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_1_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1086980800)))];
            tensor<fp16, [1280]> up_blocks_1_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1090257664)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_27_cast = conv(bias = up_blocks_1_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_5638, groups = var_5599, pad = temb_27_pad_0, pad_type = temb_27_pad_type_0, strides = var_5636, weight = up_blocks_1_resnets_0_time_emb_proj_weight_to_fp16, x = input_15_cast);
            tensor<fp16, [2, 1280, 16, 16]> input_293_cast = add(x = hidden_states_159_cast, y = temb_27_cast);
            tensor<int32, [5]> reshape_136_shape_0 = const()[name = tensor<string, []>("reshape_136_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 16, 16])];
            tensor<fp16, [2, 32, 40, 16, 16]> reshape_136_cast = reshape(shape = reshape_136_shape_0, x = input_293_cast);
            tensor<int32, [3]> reduce_mean_102_axes_0 = const()[name = tensor<string, []>("reduce_mean_102_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_102_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_102_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_102_cast = reduce_mean(axes = reduce_mean_102_axes_0, keep_dims = reduce_mean_102_keep_dims_0, x = reshape_136_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> sub_68_cast = sub(x = reshape_136_cast, y = reduce_mean_102_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> square_34_cast = square(x = sub_68_cast);
            tensor<int32, [3]> reduce_mean_104_axes_0 = const()[name = tensor<string, []>("reduce_mean_104_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_104_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_104_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_104_cast = reduce_mean(axes = reduce_mean_104_axes_0, keep_dims = reduce_mean_104_keep_dims_0, x = square_34_cast);
            tensor<fp16, []> add_68_y_0_to_fp16 = const()[name = tensor<string, []>("add_68_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_68_cast = add(x = reduce_mean_104_cast, y = add_68_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_34_cast = sqrt(x = add_68_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> real_div_34_cast = real_div(x = sub_68_cast, y = sqrt_34_cast);
            tensor<int32, [4]> reshape_137_shape_0 = const()[name = tensor<string, []>("reshape_137_shape_0"), val = tensor<int32, [4]>([2, 1280, 16, 16])];
            tensor<fp16, [2, 1280, 16, 16]> reshape_137_cast = reshape(shape = reshape_137_shape_0, x = real_div_34_cast);
            tensor<fp16, [1280]> add_69_mean_0_to_fp16 = const()[name = tensor<string, []>("add_69_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1090260288)))];
            tensor<fp16, [1280]> add_69_variance_0_to_fp16 = const()[name = tensor<string, []>("add_69_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1090262912)))];
            tensor<fp16, [1280]> add_69_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_69_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1090265536)))];
            tensor<fp16, [1280]> add_69_beta_0_to_fp16 = const()[name = tensor<string, []>("add_69_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1090268160)))];
            tensor<fp16, []> add_69_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_69_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 16, 16]> add_69_cast = batch_norm(beta = add_69_beta_0_to_fp16, epsilon = add_69_epsilon_0_to_fp16, gamma = add_69_gamma_0_to_fp16, mean = add_69_mean_0_to_fp16, variance = add_69_variance_0_to_fp16, x = reshape_137_cast);
            tensor<fp16, [2, 1280, 16, 16]> input_297_cast = silu(x = add_69_cast);
            tensor<int32, [2]> var_5648 = const()[name = tensor<string, []>("op_5648"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5650 = const()[name = tensor<string, []>("op_5650"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_161_pad_type_0 = const()[name = tensor<string, []>("hidden_states_161_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_161_pad_0 = const()[name = tensor<string, []>("hidden_states_161_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> up_blocks_1_resnets_0_conv2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_0_conv2_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1090270784)))];
            tensor<fp16, [1280]> up_blocks_1_resnets_0_conv2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_0_conv2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1119762048)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_161_cast = conv(bias = up_blocks_1_resnets_0_conv2_bias_to_fp16, dilations = var_5650, groups = var_5599, pad = hidden_states_161_pad_0, pad_type = hidden_states_161_pad_type_0, strides = var_5648, weight = up_blocks_1_resnets_0_conv2_weight_to_fp16, x = input_297_cast);
            tensor<int32, [2]> var_5655 = const()[name = tensor<string, []>("op_5655"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5657 = const()[name = tensor<string, []>("op_5657"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> x_11_pad_type_0 = const()[name = tensor<string, []>("x_11_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> x_11_pad_0 = const()[name = tensor<string, []>("x_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2560, 1, 1]> up_blocks_1_resnets_0_conv_shortcut_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_0_conv_shortcut_weight_to_fp16"), val = tensor<fp16, [1280, 2560, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1119764672)))];
            tensor<fp16, [1280]> up_blocks_1_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1126318336)))];
            tensor<fp16, [2, 1280, 16, 16]> x_11_cast = conv(bias = up_blocks_1_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_5657, groups = var_5599, pad = x_11_pad_0, pad_type = x_11_pad_type_0, strides = var_5655, weight = up_blocks_1_resnets_0_conv_shortcut_weight_to_fp16, x = input_285_cast);
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_163_cast = add(x = x_11_cast, y = hidden_states_161_cast);
            tensor<int32, [5]> reshape_140_shape_0 = const()[name = tensor<string, []>("reshape_140_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 16, 16])];
            tensor<fp16, [2, 32, 40, 16, 16]> reshape_140_cast = reshape(shape = reshape_140_shape_0, x = hidden_states_163_cast);
            tensor<int32, [3]> reduce_mean_105_axes_0 = const()[name = tensor<string, []>("reduce_mean_105_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_105_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_105_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_105_cast = reduce_mean(axes = reduce_mean_105_axes_0, keep_dims = reduce_mean_105_keep_dims_0, x = reshape_140_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> sub_70_cast = sub(x = reshape_140_cast, y = reduce_mean_105_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> square_35_cast = square(x = sub_70_cast);
            tensor<int32, [3]> reduce_mean_107_axes_0 = const()[name = tensor<string, []>("reduce_mean_107_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_107_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_107_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_107_cast = reduce_mean(axes = reduce_mean_107_axes_0, keep_dims = reduce_mean_107_keep_dims_0, x = square_35_cast);
            tensor<fp16, []> add_70_y_0_to_fp16 = const()[name = tensor<string, []>("add_70_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_70_cast = add(x = reduce_mean_107_cast, y = add_70_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_35_cast = sqrt(x = add_70_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> real_div_35_cast = real_div(x = sub_70_cast, y = sqrt_35_cast);
            tensor<int32, [4]> reshape_141_shape_0 = const()[name = tensor<string, []>("reshape_141_shape_0"), val = tensor<int32, [4]>([2, 1280, 16, 16])];
            tensor<fp16, [2, 1280, 16, 16]> reshape_141_cast = reshape(shape = reshape_141_shape_0, x = real_div_35_cast);
            tensor<fp16, [1280]> add_71_mean_0_to_fp16 = const()[name = tensor<string, []>("add_71_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1126320960)))];
            tensor<fp16, [1280]> add_71_variance_0_to_fp16 = const()[name = tensor<string, []>("add_71_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1126323584)))];
            tensor<fp16, [1280]> add_71_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_71_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1126326208)))];
            tensor<fp16, [1280]> add_71_beta_0_to_fp16 = const()[name = tensor<string, []>("add_71_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1126328832)))];
            tensor<fp16, []> add_71_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_71_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 16, 16]> add_71_cast = batch_norm(beta = add_71_beta_0_to_fp16, epsilon = add_71_epsilon_0_to_fp16, gamma = add_71_gamma_0_to_fp16, mean = add_71_mean_0_to_fp16, variance = add_71_variance_0_to_fp16, x = reshape_141_cast);
            tensor<int32, [2]> var_5677 = const()[name = tensor<string, []>("op_5677"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5679 = const()[name = tensor<string, []>("op_5679"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_165_pad_type_0 = const()[name = tensor<string, []>("hidden_states_165_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_165_pad_0 = const()[name = tensor<string, []>("hidden_states_165_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_1_attentions_0_proj_in_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_proj_in_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1126331456)))];
            tensor<fp16, [1280]> up_blocks_1_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_proj_in_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1129608320)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_165_cast = conv(bias = up_blocks_1_attentions_0_proj_in_bias_to_fp16, dilations = var_5679, groups = var_5599, pad = hidden_states_165_pad_0, pad_type = hidden_states_165_pad_type_0, strides = var_5677, weight = up_blocks_1_attentions_0_proj_in_weight_to_fp16, x = add_71_cast);
            tensor<int32, [4]> var_5684 = const()[name = tensor<string, []>("op_5684"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<fp16, [2, 1280, 1, 256]> inputs_43_cast = reshape(shape = var_5684, x = hidden_states_165_cast);
            tensor<int32, [1]> var_5694 = const()[name = tensor<string, []>("op_5694"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_43_cast = reduce_mean(axes = var_5694, keep_dims = var_5594, x = inputs_43_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_43_cast = sub(x = inputs_43_cast, y = channels_mean_43_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_sq_43_cast = mul(x = zero_mean_43_cast, y = zero_mean_43_cast);
            tensor<int32, [1]> var_5698 = const()[name = tensor<string, []>("op_5698"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_5699_cast = reduce_mean(axes = var_5698, keep_dims = var_5594, x = zero_mean_sq_43_cast);
            tensor<fp16, []> var_5700_to_fp16 = const()[name = tensor<string, []>("op_5700_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_5701_cast = add(x = var_5699_cast, y = var_5700_to_fp16);
            tensor<fp16, []> denom_43_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_43_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 256]> denom_43_cast = rsqrt(epsilon = denom_43_epsilon_0_to_fp16, x = var_5701_cast);
            tensor<fp16, [2, 1280, 1, 256]> out_43_cast = mul(x = zero_mean_43_cast, y = denom_43_cast);
            tensor<fp16, [1, 1280, 1, 1]> var_5705_to_fp16 = const()[name = tensor<string, []>("op_5705_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1129610944)))];
            tensor<fp16, [2, 1280, 1, 256]> var_5706_cast = add(x = out_43_cast, y = var_5705_to_fp16);
            tensor<fp16, [1, 1280, 1, 1]> var_5708_to_fp16 = const()[name = tensor<string, []>("op_5708_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1129613568)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_167_cast = mul(x = var_5706_cast, y = var_5708_to_fp16);
            tensor<int32, [2]> var_5715 = const()[name = tensor<string, []>("op_5715"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5717 = const()[name = tensor<string, []>("op_5717"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_29_pad_type_0 = const()[name = tensor<string, []>("q_29_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_29_pad_0 = const()[name = tensor<string, []>("q_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1129616192)))];
            tensor<fp16, [2, 1280, 1, 256]> q_29_cast = conv(dilations = var_5717, groups = var_5599, pad = q_29_pad_0, pad_type = q_29_pad_type_0, strides = var_5715, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_167_cast);
            tensor<int32, [2]> var_5721 = const()[name = tensor<string, []>("op_5721"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5723 = const()[name = tensor<string, []>("op_5723"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_57_pad_type_0 = const()[name = tensor<string, []>("k_57_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_57_pad_0 = const()[name = tensor<string, []>("k_57_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1132893056)))];
            tensor<fp16, [2, 1280, 1, 256]> k_57_cast = conv(dilations = var_5723, groups = var_5599, pad = k_57_pad_0, pad_type = k_57_pad_type_0, strides = var_5721, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_167_cast);
            tensor<int32, [2]> var_5727 = const()[name = tensor<string, []>("op_5727"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5729 = const()[name = tensor<string, []>("op_5729"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_29_pad_type_0 = const()[name = tensor<string, []>("v_29_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_29_pad_0 = const()[name = tensor<string, []>("v_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1136169920)))];
            tensor<fp16, [2, 1280, 1, 256]> v_29_cast = conv(dilations = var_5729, groups = var_5599, pad = v_29_pad_0, pad_type = v_29_pad_type_0, strides = var_5727, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_167_cast);
            tensor<int32, [4]> var_5733_begin_0 = const()[name = tensor<string, []>("op_5733_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5733_end_0 = const()[name = tensor<string, []>("op_5733_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_5733_end_mask_0 = const()[name = tensor<string, []>("op_5733_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5733_cast = slice_by_index(begin = var_5733_begin_0, end = var_5733_end_0, end_mask = var_5733_end_mask_0, x = q_29_cast);
            tensor<int32, [4]> var_5737_begin_0 = const()[name = tensor<string, []>("op_5737_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_5737_end_0 = const()[name = tensor<string, []>("op_5737_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_5737_end_mask_0 = const()[name = tensor<string, []>("op_5737_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5737_cast = slice_by_index(begin = var_5737_begin_0, end = var_5737_end_0, end_mask = var_5737_end_mask_0, x = q_29_cast);
            tensor<int32, [4]> var_5741_begin_0 = const()[name = tensor<string, []>("op_5741_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_5741_end_0 = const()[name = tensor<string, []>("op_5741_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_5741_end_mask_0 = const()[name = tensor<string, []>("op_5741_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5741_cast = slice_by_index(begin = var_5741_begin_0, end = var_5741_end_0, end_mask = var_5741_end_mask_0, x = q_29_cast);
            tensor<int32, [4]> var_5745_begin_0 = const()[name = tensor<string, []>("op_5745_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_5745_end_0 = const()[name = tensor<string, []>("op_5745_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_5745_end_mask_0 = const()[name = tensor<string, []>("op_5745_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5745_cast = slice_by_index(begin = var_5745_begin_0, end = var_5745_end_0, end_mask = var_5745_end_mask_0, x = q_29_cast);
            tensor<int32, [4]> var_5749_begin_0 = const()[name = tensor<string, []>("op_5749_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_5749_end_0 = const()[name = tensor<string, []>("op_5749_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_5749_end_mask_0 = const()[name = tensor<string, []>("op_5749_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5749_cast = slice_by_index(begin = var_5749_begin_0, end = var_5749_end_0, end_mask = var_5749_end_mask_0, x = q_29_cast);
            tensor<int32, [4]> var_5753_begin_0 = const()[name = tensor<string, []>("op_5753_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_5753_end_0 = const()[name = tensor<string, []>("op_5753_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_5753_end_mask_0 = const()[name = tensor<string, []>("op_5753_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5753_cast = slice_by_index(begin = var_5753_begin_0, end = var_5753_end_0, end_mask = var_5753_end_mask_0, x = q_29_cast);
            tensor<int32, [4]> var_5757_begin_0 = const()[name = tensor<string, []>("op_5757_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_5757_end_0 = const()[name = tensor<string, []>("op_5757_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_5757_end_mask_0 = const()[name = tensor<string, []>("op_5757_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5757_cast = slice_by_index(begin = var_5757_begin_0, end = var_5757_end_0, end_mask = var_5757_end_mask_0, x = q_29_cast);
            tensor<int32, [4]> var_5761_begin_0 = const()[name = tensor<string, []>("op_5761_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_5761_end_0 = const()[name = tensor<string, []>("op_5761_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_5761_end_mask_0 = const()[name = tensor<string, []>("op_5761_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5761_cast = slice_by_index(begin = var_5761_begin_0, end = var_5761_end_0, end_mask = var_5761_end_mask_0, x = q_29_cast);
            tensor<int32, [4]> var_5765_begin_0 = const()[name = tensor<string, []>("op_5765_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_5765_end_0 = const()[name = tensor<string, []>("op_5765_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_5765_end_mask_0 = const()[name = tensor<string, []>("op_5765_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5765_cast = slice_by_index(begin = var_5765_begin_0, end = var_5765_end_0, end_mask = var_5765_end_mask_0, x = q_29_cast);
            tensor<int32, [4]> var_5769_begin_0 = const()[name = tensor<string, []>("op_5769_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_5769_end_0 = const()[name = tensor<string, []>("op_5769_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_5769_end_mask_0 = const()[name = tensor<string, []>("op_5769_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5769_cast = slice_by_index(begin = var_5769_begin_0, end = var_5769_end_0, end_mask = var_5769_end_mask_0, x = q_29_cast);
            tensor<int32, [4]> var_5773_begin_0 = const()[name = tensor<string, []>("op_5773_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_5773_end_0 = const()[name = tensor<string, []>("op_5773_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_5773_end_mask_0 = const()[name = tensor<string, []>("op_5773_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5773_cast = slice_by_index(begin = var_5773_begin_0, end = var_5773_end_0, end_mask = var_5773_end_mask_0, x = q_29_cast);
            tensor<int32, [4]> var_5777_begin_0 = const()[name = tensor<string, []>("op_5777_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_5777_end_0 = const()[name = tensor<string, []>("op_5777_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_5777_end_mask_0 = const()[name = tensor<string, []>("op_5777_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5777_cast = slice_by_index(begin = var_5777_begin_0, end = var_5777_end_0, end_mask = var_5777_end_mask_0, x = q_29_cast);
            tensor<int32, [4]> var_5781_begin_0 = const()[name = tensor<string, []>("op_5781_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_5781_end_0 = const()[name = tensor<string, []>("op_5781_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_5781_end_mask_0 = const()[name = tensor<string, []>("op_5781_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5781_cast = slice_by_index(begin = var_5781_begin_0, end = var_5781_end_0, end_mask = var_5781_end_mask_0, x = q_29_cast);
            tensor<int32, [4]> var_5785_begin_0 = const()[name = tensor<string, []>("op_5785_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_5785_end_0 = const()[name = tensor<string, []>("op_5785_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_5785_end_mask_0 = const()[name = tensor<string, []>("op_5785_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5785_cast = slice_by_index(begin = var_5785_begin_0, end = var_5785_end_0, end_mask = var_5785_end_mask_0, x = q_29_cast);
            tensor<int32, [4]> var_5789_begin_0 = const()[name = tensor<string, []>("op_5789_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_5789_end_0 = const()[name = tensor<string, []>("op_5789_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_5789_end_mask_0 = const()[name = tensor<string, []>("op_5789_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5789_cast = slice_by_index(begin = var_5789_begin_0, end = var_5789_end_0, end_mask = var_5789_end_mask_0, x = q_29_cast);
            tensor<int32, [4]> var_5793_begin_0 = const()[name = tensor<string, []>("op_5793_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_5793_end_0 = const()[name = tensor<string, []>("op_5793_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_5793_end_mask_0 = const()[name = tensor<string, []>("op_5793_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5793_cast = slice_by_index(begin = var_5793_begin_0, end = var_5793_end_0, end_mask = var_5793_end_mask_0, x = q_29_cast);
            tensor<int32, [4]> var_5797_begin_0 = const()[name = tensor<string, []>("op_5797_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_5797_end_0 = const()[name = tensor<string, []>("op_5797_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_5797_end_mask_0 = const()[name = tensor<string, []>("op_5797_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5797_cast = slice_by_index(begin = var_5797_begin_0, end = var_5797_end_0, end_mask = var_5797_end_mask_0, x = q_29_cast);
            tensor<int32, [4]> var_5801_begin_0 = const()[name = tensor<string, []>("op_5801_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_5801_end_0 = const()[name = tensor<string, []>("op_5801_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_5801_end_mask_0 = const()[name = tensor<string, []>("op_5801_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5801_cast = slice_by_index(begin = var_5801_begin_0, end = var_5801_end_0, end_mask = var_5801_end_mask_0, x = q_29_cast);
            tensor<int32, [4]> var_5805_begin_0 = const()[name = tensor<string, []>("op_5805_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_5805_end_0 = const()[name = tensor<string, []>("op_5805_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_5805_end_mask_0 = const()[name = tensor<string, []>("op_5805_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5805_cast = slice_by_index(begin = var_5805_begin_0, end = var_5805_end_0, end_mask = var_5805_end_mask_0, x = q_29_cast);
            tensor<int32, [4]> var_5809_begin_0 = const()[name = tensor<string, []>("op_5809_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_5809_end_0 = const()[name = tensor<string, []>("op_5809_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_5809_end_mask_0 = const()[name = tensor<string, []>("op_5809_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5809_cast = slice_by_index(begin = var_5809_begin_0, end = var_5809_end_0, end_mask = var_5809_end_mask_0, x = q_29_cast);
            tensor<int32, [4]> k_59_perm_0 = const()[name = tensor<string, []>("k_59_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_5816_begin_0 = const()[name = tensor<string, []>("op_5816_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5816_end_0 = const()[name = tensor<string, []>("op_5816_end_0"), val = tensor<int32, [4]>([2, 256, 1, 64])];
            tensor<bool, [4]> var_5816_end_mask_0 = const()[name = tensor<string, []>("op_5816_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 1280]> transpose_17 = transpose(perm = k_59_perm_0, x = k_57_cast);
            tensor<fp16, [2, 256, 1, 64]> var_5816_cast = slice_by_index(begin = var_5816_begin_0, end = var_5816_end_0, end_mask = var_5816_end_mask_0, x = transpose_17);
            tensor<int32, [4]> var_5820_begin_0 = const()[name = tensor<string, []>("op_5820_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_5820_end_0 = const()[name = tensor<string, []>("op_5820_end_0"), val = tensor<int32, [4]>([2, 256, 1, 128])];
            tensor<bool, [4]> var_5820_end_mask_0 = const()[name = tensor<string, []>("op_5820_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5820_cast = slice_by_index(begin = var_5820_begin_0, end = var_5820_end_0, end_mask = var_5820_end_mask_0, x = transpose_17);
            tensor<int32, [4]> var_5824_begin_0 = const()[name = tensor<string, []>("op_5824_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_5824_end_0 = const()[name = tensor<string, []>("op_5824_end_0"), val = tensor<int32, [4]>([2, 256, 1, 192])];
            tensor<bool, [4]> var_5824_end_mask_0 = const()[name = tensor<string, []>("op_5824_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5824_cast = slice_by_index(begin = var_5824_begin_0, end = var_5824_end_0, end_mask = var_5824_end_mask_0, x = transpose_17);
            tensor<int32, [4]> var_5828_begin_0 = const()[name = tensor<string, []>("op_5828_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_5828_end_0 = const()[name = tensor<string, []>("op_5828_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_5828_end_mask_0 = const()[name = tensor<string, []>("op_5828_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5828_cast = slice_by_index(begin = var_5828_begin_0, end = var_5828_end_0, end_mask = var_5828_end_mask_0, x = transpose_17);
            tensor<int32, [4]> var_5832_begin_0 = const()[name = tensor<string, []>("op_5832_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_5832_end_0 = const()[name = tensor<string, []>("op_5832_end_0"), val = tensor<int32, [4]>([2, 256, 1, 320])];
            tensor<bool, [4]> var_5832_end_mask_0 = const()[name = tensor<string, []>("op_5832_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5832_cast = slice_by_index(begin = var_5832_begin_0, end = var_5832_end_0, end_mask = var_5832_end_mask_0, x = transpose_17);
            tensor<int32, [4]> var_5836_begin_0 = const()[name = tensor<string, []>("op_5836_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_5836_end_0 = const()[name = tensor<string, []>("op_5836_end_0"), val = tensor<int32, [4]>([2, 256, 1, 384])];
            tensor<bool, [4]> var_5836_end_mask_0 = const()[name = tensor<string, []>("op_5836_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5836_cast = slice_by_index(begin = var_5836_begin_0, end = var_5836_end_0, end_mask = var_5836_end_mask_0, x = transpose_17);
            tensor<int32, [4]> var_5840_begin_0 = const()[name = tensor<string, []>("op_5840_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_5840_end_0 = const()[name = tensor<string, []>("op_5840_end_0"), val = tensor<int32, [4]>([2, 256, 1, 448])];
            tensor<bool, [4]> var_5840_end_mask_0 = const()[name = tensor<string, []>("op_5840_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5840_cast = slice_by_index(begin = var_5840_begin_0, end = var_5840_end_0, end_mask = var_5840_end_mask_0, x = transpose_17);
            tensor<int32, [4]> var_5844_begin_0 = const()[name = tensor<string, []>("op_5844_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_5844_end_0 = const()[name = tensor<string, []>("op_5844_end_0"), val = tensor<int32, [4]>([2, 256, 1, 512])];
            tensor<bool, [4]> var_5844_end_mask_0 = const()[name = tensor<string, []>("op_5844_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5844_cast = slice_by_index(begin = var_5844_begin_0, end = var_5844_end_0, end_mask = var_5844_end_mask_0, x = transpose_17);
            tensor<int32, [4]> var_5848_begin_0 = const()[name = tensor<string, []>("op_5848_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_5848_end_0 = const()[name = tensor<string, []>("op_5848_end_0"), val = tensor<int32, [4]>([2, 256, 1, 576])];
            tensor<bool, [4]> var_5848_end_mask_0 = const()[name = tensor<string, []>("op_5848_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5848_cast = slice_by_index(begin = var_5848_begin_0, end = var_5848_end_0, end_mask = var_5848_end_mask_0, x = transpose_17);
            tensor<int32, [4]> var_5852_begin_0 = const()[name = tensor<string, []>("op_5852_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_5852_end_0 = const()[name = tensor<string, []>("op_5852_end_0"), val = tensor<int32, [4]>([2, 256, 1, 640])];
            tensor<bool, [4]> var_5852_end_mask_0 = const()[name = tensor<string, []>("op_5852_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5852_cast = slice_by_index(begin = var_5852_begin_0, end = var_5852_end_0, end_mask = var_5852_end_mask_0, x = transpose_17);
            tensor<int32, [4]> var_5856_begin_0 = const()[name = tensor<string, []>("op_5856_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_5856_end_0 = const()[name = tensor<string, []>("op_5856_end_0"), val = tensor<int32, [4]>([2, 256, 1, 704])];
            tensor<bool, [4]> var_5856_end_mask_0 = const()[name = tensor<string, []>("op_5856_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5856_cast = slice_by_index(begin = var_5856_begin_0, end = var_5856_end_0, end_mask = var_5856_end_mask_0, x = transpose_17);
            tensor<int32, [4]> var_5860_begin_0 = const()[name = tensor<string, []>("op_5860_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_5860_end_0 = const()[name = tensor<string, []>("op_5860_end_0"), val = tensor<int32, [4]>([2, 256, 1, 768])];
            tensor<bool, [4]> var_5860_end_mask_0 = const()[name = tensor<string, []>("op_5860_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5860_cast = slice_by_index(begin = var_5860_begin_0, end = var_5860_end_0, end_mask = var_5860_end_mask_0, x = transpose_17);
            tensor<int32, [4]> var_5864_begin_0 = const()[name = tensor<string, []>("op_5864_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_5864_end_0 = const()[name = tensor<string, []>("op_5864_end_0"), val = tensor<int32, [4]>([2, 256, 1, 832])];
            tensor<bool, [4]> var_5864_end_mask_0 = const()[name = tensor<string, []>("op_5864_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5864_cast = slice_by_index(begin = var_5864_begin_0, end = var_5864_end_0, end_mask = var_5864_end_mask_0, x = transpose_17);
            tensor<int32, [4]> var_5868_begin_0 = const()[name = tensor<string, []>("op_5868_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_5868_end_0 = const()[name = tensor<string, []>("op_5868_end_0"), val = tensor<int32, [4]>([2, 256, 1, 896])];
            tensor<bool, [4]> var_5868_end_mask_0 = const()[name = tensor<string, []>("op_5868_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5868_cast = slice_by_index(begin = var_5868_begin_0, end = var_5868_end_0, end_mask = var_5868_end_mask_0, x = transpose_17);
            tensor<int32, [4]> var_5872_begin_0 = const()[name = tensor<string, []>("op_5872_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_5872_end_0 = const()[name = tensor<string, []>("op_5872_end_0"), val = tensor<int32, [4]>([2, 256, 1, 960])];
            tensor<bool, [4]> var_5872_end_mask_0 = const()[name = tensor<string, []>("op_5872_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5872_cast = slice_by_index(begin = var_5872_begin_0, end = var_5872_end_0, end_mask = var_5872_end_mask_0, x = transpose_17);
            tensor<int32, [4]> var_5876_begin_0 = const()[name = tensor<string, []>("op_5876_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_5876_end_0 = const()[name = tensor<string, []>("op_5876_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_5876_end_mask_0 = const()[name = tensor<string, []>("op_5876_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5876_cast = slice_by_index(begin = var_5876_begin_0, end = var_5876_end_0, end_mask = var_5876_end_mask_0, x = transpose_17);
            tensor<int32, [4]> var_5880_begin_0 = const()[name = tensor<string, []>("op_5880_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_5880_end_0 = const()[name = tensor<string, []>("op_5880_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1088])];
            tensor<bool, [4]> var_5880_end_mask_0 = const()[name = tensor<string, []>("op_5880_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5880_cast = slice_by_index(begin = var_5880_begin_0, end = var_5880_end_0, end_mask = var_5880_end_mask_0, x = transpose_17);
            tensor<int32, [4]> var_5884_begin_0 = const()[name = tensor<string, []>("op_5884_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_5884_end_0 = const()[name = tensor<string, []>("op_5884_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1152])];
            tensor<bool, [4]> var_5884_end_mask_0 = const()[name = tensor<string, []>("op_5884_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5884_cast = slice_by_index(begin = var_5884_begin_0, end = var_5884_end_0, end_mask = var_5884_end_mask_0, x = transpose_17);
            tensor<int32, [4]> var_5888_begin_0 = const()[name = tensor<string, []>("op_5888_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_5888_end_0 = const()[name = tensor<string, []>("op_5888_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1216])];
            tensor<bool, [4]> var_5888_end_mask_0 = const()[name = tensor<string, []>("op_5888_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5888_cast = slice_by_index(begin = var_5888_begin_0, end = var_5888_end_0, end_mask = var_5888_end_mask_0, x = transpose_17);
            tensor<int32, [4]> var_5892_begin_0 = const()[name = tensor<string, []>("op_5892_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_5892_end_0 = const()[name = tensor<string, []>("op_5892_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1280])];
            tensor<bool, [4]> var_5892_end_mask_0 = const()[name = tensor<string, []>("op_5892_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5892_cast = slice_by_index(begin = var_5892_begin_0, end = var_5892_end_0, end_mask = var_5892_end_mask_0, x = transpose_17);
            tensor<int32, [4]> var_5894_begin_0 = const()[name = tensor<string, []>("op_5894_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5894_end_0 = const()[name = tensor<string, []>("op_5894_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_5894_end_mask_0 = const()[name = tensor<string, []>("op_5894_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5894_cast = slice_by_index(begin = var_5894_begin_0, end = var_5894_end_0, end_mask = var_5894_end_mask_0, x = v_29_cast);
            tensor<int32, [4]> var_5898_begin_0 = const()[name = tensor<string, []>("op_5898_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_5898_end_0 = const()[name = tensor<string, []>("op_5898_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_5898_end_mask_0 = const()[name = tensor<string, []>("op_5898_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5898_cast = slice_by_index(begin = var_5898_begin_0, end = var_5898_end_0, end_mask = var_5898_end_mask_0, x = v_29_cast);
            tensor<int32, [4]> var_5902_begin_0 = const()[name = tensor<string, []>("op_5902_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_5902_end_0 = const()[name = tensor<string, []>("op_5902_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_5902_end_mask_0 = const()[name = tensor<string, []>("op_5902_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5902_cast = slice_by_index(begin = var_5902_begin_0, end = var_5902_end_0, end_mask = var_5902_end_mask_0, x = v_29_cast);
            tensor<int32, [4]> var_5906_begin_0 = const()[name = tensor<string, []>("op_5906_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_5906_end_0 = const()[name = tensor<string, []>("op_5906_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_5906_end_mask_0 = const()[name = tensor<string, []>("op_5906_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5906_cast = slice_by_index(begin = var_5906_begin_0, end = var_5906_end_0, end_mask = var_5906_end_mask_0, x = v_29_cast);
            tensor<int32, [4]> var_5910_begin_0 = const()[name = tensor<string, []>("op_5910_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_5910_end_0 = const()[name = tensor<string, []>("op_5910_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_5910_end_mask_0 = const()[name = tensor<string, []>("op_5910_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5910_cast = slice_by_index(begin = var_5910_begin_0, end = var_5910_end_0, end_mask = var_5910_end_mask_0, x = v_29_cast);
            tensor<int32, [4]> var_5914_begin_0 = const()[name = tensor<string, []>("op_5914_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_5914_end_0 = const()[name = tensor<string, []>("op_5914_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_5914_end_mask_0 = const()[name = tensor<string, []>("op_5914_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5914_cast = slice_by_index(begin = var_5914_begin_0, end = var_5914_end_0, end_mask = var_5914_end_mask_0, x = v_29_cast);
            tensor<int32, [4]> var_5918_begin_0 = const()[name = tensor<string, []>("op_5918_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_5918_end_0 = const()[name = tensor<string, []>("op_5918_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_5918_end_mask_0 = const()[name = tensor<string, []>("op_5918_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5918_cast = slice_by_index(begin = var_5918_begin_0, end = var_5918_end_0, end_mask = var_5918_end_mask_0, x = v_29_cast);
            tensor<int32, [4]> var_5922_begin_0 = const()[name = tensor<string, []>("op_5922_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_5922_end_0 = const()[name = tensor<string, []>("op_5922_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_5922_end_mask_0 = const()[name = tensor<string, []>("op_5922_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5922_cast = slice_by_index(begin = var_5922_begin_0, end = var_5922_end_0, end_mask = var_5922_end_mask_0, x = v_29_cast);
            tensor<int32, [4]> var_5926_begin_0 = const()[name = tensor<string, []>("op_5926_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_5926_end_0 = const()[name = tensor<string, []>("op_5926_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_5926_end_mask_0 = const()[name = tensor<string, []>("op_5926_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5926_cast = slice_by_index(begin = var_5926_begin_0, end = var_5926_end_0, end_mask = var_5926_end_mask_0, x = v_29_cast);
            tensor<int32, [4]> var_5930_begin_0 = const()[name = tensor<string, []>("op_5930_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_5930_end_0 = const()[name = tensor<string, []>("op_5930_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_5930_end_mask_0 = const()[name = tensor<string, []>("op_5930_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5930_cast = slice_by_index(begin = var_5930_begin_0, end = var_5930_end_0, end_mask = var_5930_end_mask_0, x = v_29_cast);
            tensor<int32, [4]> var_5934_begin_0 = const()[name = tensor<string, []>("op_5934_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_5934_end_0 = const()[name = tensor<string, []>("op_5934_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_5934_end_mask_0 = const()[name = tensor<string, []>("op_5934_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5934_cast = slice_by_index(begin = var_5934_begin_0, end = var_5934_end_0, end_mask = var_5934_end_mask_0, x = v_29_cast);
            tensor<int32, [4]> var_5938_begin_0 = const()[name = tensor<string, []>("op_5938_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_5938_end_0 = const()[name = tensor<string, []>("op_5938_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_5938_end_mask_0 = const()[name = tensor<string, []>("op_5938_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5938_cast = slice_by_index(begin = var_5938_begin_0, end = var_5938_end_0, end_mask = var_5938_end_mask_0, x = v_29_cast);
            tensor<int32, [4]> var_5942_begin_0 = const()[name = tensor<string, []>("op_5942_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_5942_end_0 = const()[name = tensor<string, []>("op_5942_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_5942_end_mask_0 = const()[name = tensor<string, []>("op_5942_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5942_cast = slice_by_index(begin = var_5942_begin_0, end = var_5942_end_0, end_mask = var_5942_end_mask_0, x = v_29_cast);
            tensor<int32, [4]> var_5946_begin_0 = const()[name = tensor<string, []>("op_5946_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_5946_end_0 = const()[name = tensor<string, []>("op_5946_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_5946_end_mask_0 = const()[name = tensor<string, []>("op_5946_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5946_cast = slice_by_index(begin = var_5946_begin_0, end = var_5946_end_0, end_mask = var_5946_end_mask_0, x = v_29_cast);
            tensor<int32, [4]> var_5950_begin_0 = const()[name = tensor<string, []>("op_5950_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_5950_end_0 = const()[name = tensor<string, []>("op_5950_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_5950_end_mask_0 = const()[name = tensor<string, []>("op_5950_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5950_cast = slice_by_index(begin = var_5950_begin_0, end = var_5950_end_0, end_mask = var_5950_end_mask_0, x = v_29_cast);
            tensor<int32, [4]> var_5954_begin_0 = const()[name = tensor<string, []>("op_5954_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_5954_end_0 = const()[name = tensor<string, []>("op_5954_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_5954_end_mask_0 = const()[name = tensor<string, []>("op_5954_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5954_cast = slice_by_index(begin = var_5954_begin_0, end = var_5954_end_0, end_mask = var_5954_end_mask_0, x = v_29_cast);
            tensor<int32, [4]> var_5958_begin_0 = const()[name = tensor<string, []>("op_5958_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_5958_end_0 = const()[name = tensor<string, []>("op_5958_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_5958_end_mask_0 = const()[name = tensor<string, []>("op_5958_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5958_cast = slice_by_index(begin = var_5958_begin_0, end = var_5958_end_0, end_mask = var_5958_end_mask_0, x = v_29_cast);
            tensor<int32, [4]> var_5962_begin_0 = const()[name = tensor<string, []>("op_5962_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_5962_end_0 = const()[name = tensor<string, []>("op_5962_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_5962_end_mask_0 = const()[name = tensor<string, []>("op_5962_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5962_cast = slice_by_index(begin = var_5962_begin_0, end = var_5962_end_0, end_mask = var_5962_end_mask_0, x = v_29_cast);
            tensor<int32, [4]> var_5966_begin_0 = const()[name = tensor<string, []>("op_5966_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_5966_end_0 = const()[name = tensor<string, []>("op_5966_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_5966_end_mask_0 = const()[name = tensor<string, []>("op_5966_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5966_cast = slice_by_index(begin = var_5966_begin_0, end = var_5966_end_0, end_mask = var_5966_end_mask_0, x = v_29_cast);
            tensor<int32, [4]> var_5970_begin_0 = const()[name = tensor<string, []>("op_5970_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_5970_end_0 = const()[name = tensor<string, []>("op_5970_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_5970_end_mask_0 = const()[name = tensor<string, []>("op_5970_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5970_cast = slice_by_index(begin = var_5970_begin_0, end = var_5970_end_0, end_mask = var_5970_end_mask_0, x = v_29_cast);
            tensor<string, []> var_5974_equation_0 = const()[name = tensor<string, []>("op_5974_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5974_cast = einsum(equation = var_5974_equation_0, values = (var_5816_cast, var_5733_cast));
            tensor<fp16, []> var_5975_to_fp16 = const()[name = tensor<string, []>("op_5975_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_361_cast = mul(x = var_5974_cast, y = var_5975_to_fp16);
            tensor<string, []> var_5978_equation_0 = const()[name = tensor<string, []>("op_5978_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5978_cast = einsum(equation = var_5978_equation_0, values = (var_5820_cast, var_5737_cast));
            tensor<fp16, []> var_5979_to_fp16 = const()[name = tensor<string, []>("op_5979_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_363_cast = mul(x = var_5978_cast, y = var_5979_to_fp16);
            tensor<string, []> var_5982_equation_0 = const()[name = tensor<string, []>("op_5982_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5982_cast = einsum(equation = var_5982_equation_0, values = (var_5824_cast, var_5741_cast));
            tensor<fp16, []> var_5983_to_fp16 = const()[name = tensor<string, []>("op_5983_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_365_cast = mul(x = var_5982_cast, y = var_5983_to_fp16);
            tensor<string, []> var_5986_equation_0 = const()[name = tensor<string, []>("op_5986_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5986_cast = einsum(equation = var_5986_equation_0, values = (var_5828_cast, var_5745_cast));
            tensor<fp16, []> var_5987_to_fp16 = const()[name = tensor<string, []>("op_5987_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_367_cast = mul(x = var_5986_cast, y = var_5987_to_fp16);
            tensor<string, []> var_5990_equation_0 = const()[name = tensor<string, []>("op_5990_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5990_cast = einsum(equation = var_5990_equation_0, values = (var_5832_cast, var_5749_cast));
            tensor<fp16, []> var_5991_to_fp16 = const()[name = tensor<string, []>("op_5991_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_369_cast = mul(x = var_5990_cast, y = var_5991_to_fp16);
            tensor<string, []> var_5994_equation_0 = const()[name = tensor<string, []>("op_5994_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5994_cast = einsum(equation = var_5994_equation_0, values = (var_5836_cast, var_5753_cast));
            tensor<fp16, []> var_5995_to_fp16 = const()[name = tensor<string, []>("op_5995_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_371_cast = mul(x = var_5994_cast, y = var_5995_to_fp16);
            tensor<string, []> var_5998_equation_0 = const()[name = tensor<string, []>("op_5998_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5998_cast = einsum(equation = var_5998_equation_0, values = (var_5840_cast, var_5757_cast));
            tensor<fp16, []> var_5999_to_fp16 = const()[name = tensor<string, []>("op_5999_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_373_cast = mul(x = var_5998_cast, y = var_5999_to_fp16);
            tensor<string, []> var_6002_equation_0 = const()[name = tensor<string, []>("op_6002_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6002_cast = einsum(equation = var_6002_equation_0, values = (var_5844_cast, var_5761_cast));
            tensor<fp16, []> var_6003_to_fp16 = const()[name = tensor<string, []>("op_6003_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_375_cast = mul(x = var_6002_cast, y = var_6003_to_fp16);
            tensor<string, []> var_6006_equation_0 = const()[name = tensor<string, []>("op_6006_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6006_cast = einsum(equation = var_6006_equation_0, values = (var_5848_cast, var_5765_cast));
            tensor<fp16, []> var_6007_to_fp16 = const()[name = tensor<string, []>("op_6007_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_377_cast = mul(x = var_6006_cast, y = var_6007_to_fp16);
            tensor<string, []> var_6010_equation_0 = const()[name = tensor<string, []>("op_6010_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6010_cast = einsum(equation = var_6010_equation_0, values = (var_5852_cast, var_5769_cast));
            tensor<fp16, []> var_6011_to_fp16 = const()[name = tensor<string, []>("op_6011_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_379_cast = mul(x = var_6010_cast, y = var_6011_to_fp16);
            tensor<string, []> var_6014_equation_0 = const()[name = tensor<string, []>("op_6014_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6014_cast = einsum(equation = var_6014_equation_0, values = (var_5856_cast, var_5773_cast));
            tensor<fp16, []> var_6015_to_fp16 = const()[name = tensor<string, []>("op_6015_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_381_cast = mul(x = var_6014_cast, y = var_6015_to_fp16);
            tensor<string, []> var_6018_equation_0 = const()[name = tensor<string, []>("op_6018_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6018_cast = einsum(equation = var_6018_equation_0, values = (var_5860_cast, var_5777_cast));
            tensor<fp16, []> var_6019_to_fp16 = const()[name = tensor<string, []>("op_6019_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_383_cast = mul(x = var_6018_cast, y = var_6019_to_fp16);
            tensor<string, []> var_6022_equation_0 = const()[name = tensor<string, []>("op_6022_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6022_cast = einsum(equation = var_6022_equation_0, values = (var_5864_cast, var_5781_cast));
            tensor<fp16, []> var_6023_to_fp16 = const()[name = tensor<string, []>("op_6023_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_385_cast = mul(x = var_6022_cast, y = var_6023_to_fp16);
            tensor<string, []> var_6026_equation_0 = const()[name = tensor<string, []>("op_6026_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6026_cast = einsum(equation = var_6026_equation_0, values = (var_5868_cast, var_5785_cast));
            tensor<fp16, []> var_6027_to_fp16 = const()[name = tensor<string, []>("op_6027_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_387_cast = mul(x = var_6026_cast, y = var_6027_to_fp16);
            tensor<string, []> var_6030_equation_0 = const()[name = tensor<string, []>("op_6030_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6030_cast = einsum(equation = var_6030_equation_0, values = (var_5872_cast, var_5789_cast));
            tensor<fp16, []> var_6031_to_fp16 = const()[name = tensor<string, []>("op_6031_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_389_cast = mul(x = var_6030_cast, y = var_6031_to_fp16);
            tensor<string, []> var_6034_equation_0 = const()[name = tensor<string, []>("op_6034_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6034_cast = einsum(equation = var_6034_equation_0, values = (var_5876_cast, var_5793_cast));
            tensor<fp16, []> var_6035_to_fp16 = const()[name = tensor<string, []>("op_6035_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_391_cast = mul(x = var_6034_cast, y = var_6035_to_fp16);
            tensor<string, []> var_6038_equation_0 = const()[name = tensor<string, []>("op_6038_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6038_cast = einsum(equation = var_6038_equation_0, values = (var_5880_cast, var_5797_cast));
            tensor<fp16, []> var_6039_to_fp16 = const()[name = tensor<string, []>("op_6039_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_393_cast = mul(x = var_6038_cast, y = var_6039_to_fp16);
            tensor<string, []> var_6042_equation_0 = const()[name = tensor<string, []>("op_6042_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6042_cast = einsum(equation = var_6042_equation_0, values = (var_5884_cast, var_5801_cast));
            tensor<fp16, []> var_6043_to_fp16 = const()[name = tensor<string, []>("op_6043_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_395_cast = mul(x = var_6042_cast, y = var_6043_to_fp16);
            tensor<string, []> var_6046_equation_0 = const()[name = tensor<string, []>("op_6046_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6046_cast = einsum(equation = var_6046_equation_0, values = (var_5888_cast, var_5805_cast));
            tensor<fp16, []> var_6047_to_fp16 = const()[name = tensor<string, []>("op_6047_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_397_cast = mul(x = var_6046_cast, y = var_6047_to_fp16);
            tensor<string, []> var_6050_equation_0 = const()[name = tensor<string, []>("op_6050_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6050_cast = einsum(equation = var_6050_equation_0, values = (var_5892_cast, var_5809_cast));
            tensor<fp16, []> var_6051_to_fp16 = const()[name = tensor<string, []>("op_6051_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_399_cast = mul(x = var_6050_cast, y = var_6051_to_fp16);
            tensor<fp16, [2, 256, 1, 256]> var_6053_cast = softmax(axis = var_5599, x = aw_361_cast);
            tensor<fp16, [2, 256, 1, 256]> var_6054_cast = softmax(axis = var_5599, x = aw_363_cast);
            tensor<fp16, [2, 256, 1, 256]> var_6055_cast = softmax(axis = var_5599, x = aw_365_cast);
            tensor<fp16, [2, 256, 1, 256]> var_6056_cast = softmax(axis = var_5599, x = aw_367_cast);
            tensor<fp16, [2, 256, 1, 256]> var_6057_cast = softmax(axis = var_5599, x = aw_369_cast);
            tensor<fp16, [2, 256, 1, 256]> var_6058_cast = softmax(axis = var_5599, x = aw_371_cast);
            tensor<fp16, [2, 256, 1, 256]> var_6059_cast = softmax(axis = var_5599, x = aw_373_cast);
            tensor<fp16, [2, 256, 1, 256]> var_6060_cast = softmax(axis = var_5599, x = aw_375_cast);
            tensor<fp16, [2, 256, 1, 256]> var_6061_cast = softmax(axis = var_5599, x = aw_377_cast);
            tensor<fp16, [2, 256, 1, 256]> var_6062_cast = softmax(axis = var_5599, x = aw_379_cast);
            tensor<fp16, [2, 256, 1, 256]> var_6063_cast = softmax(axis = var_5599, x = aw_381_cast);
            tensor<fp16, [2, 256, 1, 256]> var_6064_cast = softmax(axis = var_5599, x = aw_383_cast);
            tensor<fp16, [2, 256, 1, 256]> var_6065_cast = softmax(axis = var_5599, x = aw_385_cast);
            tensor<fp16, [2, 256, 1, 256]> var_6066_cast = softmax(axis = var_5599, x = aw_387_cast);
            tensor<fp16, [2, 256, 1, 256]> var_6067_cast = softmax(axis = var_5599, x = aw_389_cast);
            tensor<fp16, [2, 256, 1, 256]> var_6068_cast = softmax(axis = var_5599, x = aw_391_cast);
            tensor<fp16, [2, 256, 1, 256]> var_6069_cast = softmax(axis = var_5599, x = aw_393_cast);
            tensor<fp16, [2, 256, 1, 256]> var_6070_cast = softmax(axis = var_5599, x = aw_395_cast);
            tensor<fp16, [2, 256, 1, 256]> var_6071_cast = softmax(axis = var_5599, x = aw_397_cast);
            tensor<fp16, [2, 256, 1, 256]> var_6072_cast = softmax(axis = var_5599, x = aw_399_cast);
            tensor<string, []> var_6074_equation_0 = const()[name = tensor<string, []>("op_6074_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6074_cast = einsum(equation = var_6074_equation_0, values = (var_5894_cast, var_6053_cast));
            tensor<string, []> var_6076_equation_0 = const()[name = tensor<string, []>("op_6076_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6076_cast = einsum(equation = var_6076_equation_0, values = (var_5898_cast, var_6054_cast));
            tensor<string, []> var_6078_equation_0 = const()[name = tensor<string, []>("op_6078_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6078_cast = einsum(equation = var_6078_equation_0, values = (var_5902_cast, var_6055_cast));
            tensor<string, []> var_6080_equation_0 = const()[name = tensor<string, []>("op_6080_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6080_cast = einsum(equation = var_6080_equation_0, values = (var_5906_cast, var_6056_cast));
            tensor<string, []> var_6082_equation_0 = const()[name = tensor<string, []>("op_6082_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6082_cast = einsum(equation = var_6082_equation_0, values = (var_5910_cast, var_6057_cast));
            tensor<string, []> var_6084_equation_0 = const()[name = tensor<string, []>("op_6084_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6084_cast = einsum(equation = var_6084_equation_0, values = (var_5914_cast, var_6058_cast));
            tensor<string, []> var_6086_equation_0 = const()[name = tensor<string, []>("op_6086_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6086_cast = einsum(equation = var_6086_equation_0, values = (var_5918_cast, var_6059_cast));
            tensor<string, []> var_6088_equation_0 = const()[name = tensor<string, []>("op_6088_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6088_cast = einsum(equation = var_6088_equation_0, values = (var_5922_cast, var_6060_cast));
            tensor<string, []> var_6090_equation_0 = const()[name = tensor<string, []>("op_6090_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6090_cast = einsum(equation = var_6090_equation_0, values = (var_5926_cast, var_6061_cast));
            tensor<string, []> var_6092_equation_0 = const()[name = tensor<string, []>("op_6092_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6092_cast = einsum(equation = var_6092_equation_0, values = (var_5930_cast, var_6062_cast));
            tensor<string, []> var_6094_equation_0 = const()[name = tensor<string, []>("op_6094_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6094_cast = einsum(equation = var_6094_equation_0, values = (var_5934_cast, var_6063_cast));
            tensor<string, []> var_6096_equation_0 = const()[name = tensor<string, []>("op_6096_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6096_cast = einsum(equation = var_6096_equation_0, values = (var_5938_cast, var_6064_cast));
            tensor<string, []> var_6098_equation_0 = const()[name = tensor<string, []>("op_6098_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6098_cast = einsum(equation = var_6098_equation_0, values = (var_5942_cast, var_6065_cast));
            tensor<string, []> var_6100_equation_0 = const()[name = tensor<string, []>("op_6100_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6100_cast = einsum(equation = var_6100_equation_0, values = (var_5946_cast, var_6066_cast));
            tensor<string, []> var_6102_equation_0 = const()[name = tensor<string, []>("op_6102_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6102_cast = einsum(equation = var_6102_equation_0, values = (var_5950_cast, var_6067_cast));
            tensor<string, []> var_6104_equation_0 = const()[name = tensor<string, []>("op_6104_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6104_cast = einsum(equation = var_6104_equation_0, values = (var_5954_cast, var_6068_cast));
            tensor<string, []> var_6106_equation_0 = const()[name = tensor<string, []>("op_6106_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6106_cast = einsum(equation = var_6106_equation_0, values = (var_5958_cast, var_6069_cast));
            tensor<string, []> var_6108_equation_0 = const()[name = tensor<string, []>("op_6108_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6108_cast = einsum(equation = var_6108_equation_0, values = (var_5962_cast, var_6070_cast));
            tensor<string, []> var_6110_equation_0 = const()[name = tensor<string, []>("op_6110_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6110_cast = einsum(equation = var_6110_equation_0, values = (var_5966_cast, var_6071_cast));
            tensor<string, []> var_6112_equation_0 = const()[name = tensor<string, []>("op_6112_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6112_cast = einsum(equation = var_6112_equation_0, values = (var_5970_cast, var_6072_cast));
            tensor<bool, []> input_301_interleave_0 = const()[name = tensor<string, []>("input_301_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 256]> input_301_cast = concat(axis = var_5599, interleave = input_301_interleave_0, values = (var_6074_cast, var_6076_cast, var_6078_cast, var_6080_cast, var_6082_cast, var_6084_cast, var_6086_cast, var_6088_cast, var_6090_cast, var_6092_cast, var_6094_cast, var_6096_cast, var_6098_cast, var_6100_cast, var_6102_cast, var_6104_cast, var_6106_cast, var_6108_cast, var_6110_cast, var_6112_cast));
            tensor<int32, [2]> var_6118 = const()[name = tensor<string, []>("op_6118"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6120 = const()[name = tensor<string, []>("op_6120"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_6122_pad_type_0 = const()[name = tensor<string, []>("op_6122_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_6122_pad_0 = const()[name = tensor<string, []>("op_6122_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1139446784)))];
            tensor<fp16, [1280]> up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1142723648)))];
            tensor<fp16, [2, 1280, 1, 256]> var_6122_cast = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_6120, groups = var_5599, pad = var_6122_pad_0, pad_type = var_6122_pad_type_0, strides = var_6118, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_301_cast);
            tensor<fp16, [2, 1280, 1, 256]> inputs_45_cast = add(x = var_6122_cast, y = inputs_43_cast);
            tensor<int32, [1]> var_6126 = const()[name = tensor<string, []>("op_6126"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_45_cast = reduce_mean(axes = var_6126, keep_dims = var_5594, x = inputs_45_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_45_cast = sub(x = inputs_45_cast, y = channels_mean_45_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_sq_45_cast = mul(x = zero_mean_45_cast, y = zero_mean_45_cast);
            tensor<int32, [1]> var_6130 = const()[name = tensor<string, []>("op_6130"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_6131_cast = reduce_mean(axes = var_6130, keep_dims = var_5594, x = zero_mean_sq_45_cast);
            tensor<fp16, []> var_6132_to_fp16 = const()[name = tensor<string, []>("op_6132_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_6133_cast = add(x = var_6131_cast, y = var_6132_to_fp16);
            tensor<fp16, []> denom_45_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_45_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 256]> denom_45_cast = rsqrt(epsilon = denom_45_epsilon_0_to_fp16, x = var_6133_cast);
            tensor<fp16, [2, 1280, 1, 256]> out_45_cast = mul(x = zero_mean_45_cast, y = denom_45_cast);
            tensor<fp16, [1, 1280, 1, 1]> var_6137_to_fp16 = const()[name = tensor<string, []>("op_6137_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1142726272)))];
            tensor<fp16, [2, 1280, 1, 256]> var_6138_cast = add(x = out_45_cast, y = var_6137_to_fp16);
            tensor<fp16, [1, 1280, 1, 1]> var_6140_to_fp16 = const()[name = tensor<string, []>("op_6140_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1142728896)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_169_cast = mul(x = var_6138_cast, y = var_6140_to_fp16);
            tensor<int32, [2]> var_6147 = const()[name = tensor<string, []>("op_6147"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6149 = const()[name = tensor<string, []>("op_6149"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_31_pad_type_0 = const()[name = tensor<string, []>("q_31_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_31_pad_0 = const()[name = tensor<string, []>("q_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1142731520)))];
            tensor<fp16, [2, 1280, 1, 256]> q_31_cast = conv(dilations = var_6149, groups = var_5599, pad = q_31_pad_0, pad_type = q_31_pad_type_0, strides = var_6147, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_169_cast);
            tensor<int32, [2]> var_6153 = const()[name = tensor<string, []>("op_6153"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6155 = const()[name = tensor<string, []>("op_6155"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_61_pad_type_0 = const()[name = tensor<string, []>("k_61_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_61_pad_0 = const()[name = tensor<string, []>("k_61_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1024, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1146008384)))];
            tensor<fp16, [2, 1280, 1, 77]> k_61_cast = conv(dilations = var_6155, groups = var_5599, pad = k_61_pad_0, pad_type = k_61_pad_type_0, strides = var_6153, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [2]> var_6159 = const()[name = tensor<string, []>("op_6159"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6161 = const()[name = tensor<string, []>("op_6161"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_31_pad_type_0 = const()[name = tensor<string, []>("v_31_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_31_pad_0 = const()[name = tensor<string, []>("v_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1024, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1148629888)))];
            tensor<fp16, [2, 1280, 1, 77]> v_31_cast = conv(dilations = var_6161, groups = var_5599, pad = v_31_pad_0, pad_type = v_31_pad_type_0, strides = var_6159, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [4]> var_6165_begin_0 = const()[name = tensor<string, []>("op_6165_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_6165_end_0 = const()[name = tensor<string, []>("op_6165_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_6165_end_mask_0 = const()[name = tensor<string, []>("op_6165_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6165_cast = slice_by_index(begin = var_6165_begin_0, end = var_6165_end_0, end_mask = var_6165_end_mask_0, x = q_31_cast);
            tensor<int32, [4]> var_6169_begin_0 = const()[name = tensor<string, []>("op_6169_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_6169_end_0 = const()[name = tensor<string, []>("op_6169_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_6169_end_mask_0 = const()[name = tensor<string, []>("op_6169_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6169_cast = slice_by_index(begin = var_6169_begin_0, end = var_6169_end_0, end_mask = var_6169_end_mask_0, x = q_31_cast);
            tensor<int32, [4]> var_6173_begin_0 = const()[name = tensor<string, []>("op_6173_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_6173_end_0 = const()[name = tensor<string, []>("op_6173_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_6173_end_mask_0 = const()[name = tensor<string, []>("op_6173_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6173_cast = slice_by_index(begin = var_6173_begin_0, end = var_6173_end_0, end_mask = var_6173_end_mask_0, x = q_31_cast);
            tensor<int32, [4]> var_6177_begin_0 = const()[name = tensor<string, []>("op_6177_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_6177_end_0 = const()[name = tensor<string, []>("op_6177_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_6177_end_mask_0 = const()[name = tensor<string, []>("op_6177_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6177_cast = slice_by_index(begin = var_6177_begin_0, end = var_6177_end_0, end_mask = var_6177_end_mask_0, x = q_31_cast);
            tensor<int32, [4]> var_6181_begin_0 = const()[name = tensor<string, []>("op_6181_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_6181_end_0 = const()[name = tensor<string, []>("op_6181_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_6181_end_mask_0 = const()[name = tensor<string, []>("op_6181_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6181_cast = slice_by_index(begin = var_6181_begin_0, end = var_6181_end_0, end_mask = var_6181_end_mask_0, x = q_31_cast);
            tensor<int32, [4]> var_6185_begin_0 = const()[name = tensor<string, []>("op_6185_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_6185_end_0 = const()[name = tensor<string, []>("op_6185_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_6185_end_mask_0 = const()[name = tensor<string, []>("op_6185_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6185_cast = slice_by_index(begin = var_6185_begin_0, end = var_6185_end_0, end_mask = var_6185_end_mask_0, x = q_31_cast);
            tensor<int32, [4]> var_6189_begin_0 = const()[name = tensor<string, []>("op_6189_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_6189_end_0 = const()[name = tensor<string, []>("op_6189_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_6189_end_mask_0 = const()[name = tensor<string, []>("op_6189_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6189_cast = slice_by_index(begin = var_6189_begin_0, end = var_6189_end_0, end_mask = var_6189_end_mask_0, x = q_31_cast);
            tensor<int32, [4]> var_6193_begin_0 = const()[name = tensor<string, []>("op_6193_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_6193_end_0 = const()[name = tensor<string, []>("op_6193_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_6193_end_mask_0 = const()[name = tensor<string, []>("op_6193_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6193_cast = slice_by_index(begin = var_6193_begin_0, end = var_6193_end_0, end_mask = var_6193_end_mask_0, x = q_31_cast);
            tensor<int32, [4]> var_6197_begin_0 = const()[name = tensor<string, []>("op_6197_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_6197_end_0 = const()[name = tensor<string, []>("op_6197_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_6197_end_mask_0 = const()[name = tensor<string, []>("op_6197_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6197_cast = slice_by_index(begin = var_6197_begin_0, end = var_6197_end_0, end_mask = var_6197_end_mask_0, x = q_31_cast);
            tensor<int32, [4]> var_6201_begin_0 = const()[name = tensor<string, []>("op_6201_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_6201_end_0 = const()[name = tensor<string, []>("op_6201_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_6201_end_mask_0 = const()[name = tensor<string, []>("op_6201_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6201_cast = slice_by_index(begin = var_6201_begin_0, end = var_6201_end_0, end_mask = var_6201_end_mask_0, x = q_31_cast);
            tensor<int32, [4]> var_6205_begin_0 = const()[name = tensor<string, []>("op_6205_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_6205_end_0 = const()[name = tensor<string, []>("op_6205_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_6205_end_mask_0 = const()[name = tensor<string, []>("op_6205_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6205_cast = slice_by_index(begin = var_6205_begin_0, end = var_6205_end_0, end_mask = var_6205_end_mask_0, x = q_31_cast);
            tensor<int32, [4]> var_6209_begin_0 = const()[name = tensor<string, []>("op_6209_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_6209_end_0 = const()[name = tensor<string, []>("op_6209_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_6209_end_mask_0 = const()[name = tensor<string, []>("op_6209_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6209_cast = slice_by_index(begin = var_6209_begin_0, end = var_6209_end_0, end_mask = var_6209_end_mask_0, x = q_31_cast);
            tensor<int32, [4]> var_6213_begin_0 = const()[name = tensor<string, []>("op_6213_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_6213_end_0 = const()[name = tensor<string, []>("op_6213_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_6213_end_mask_0 = const()[name = tensor<string, []>("op_6213_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6213_cast = slice_by_index(begin = var_6213_begin_0, end = var_6213_end_0, end_mask = var_6213_end_mask_0, x = q_31_cast);
            tensor<int32, [4]> var_6217_begin_0 = const()[name = tensor<string, []>("op_6217_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_6217_end_0 = const()[name = tensor<string, []>("op_6217_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_6217_end_mask_0 = const()[name = tensor<string, []>("op_6217_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6217_cast = slice_by_index(begin = var_6217_begin_0, end = var_6217_end_0, end_mask = var_6217_end_mask_0, x = q_31_cast);
            tensor<int32, [4]> var_6221_begin_0 = const()[name = tensor<string, []>("op_6221_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_6221_end_0 = const()[name = tensor<string, []>("op_6221_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_6221_end_mask_0 = const()[name = tensor<string, []>("op_6221_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6221_cast = slice_by_index(begin = var_6221_begin_0, end = var_6221_end_0, end_mask = var_6221_end_mask_0, x = q_31_cast);
            tensor<int32, [4]> var_6225_begin_0 = const()[name = tensor<string, []>("op_6225_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_6225_end_0 = const()[name = tensor<string, []>("op_6225_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_6225_end_mask_0 = const()[name = tensor<string, []>("op_6225_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6225_cast = slice_by_index(begin = var_6225_begin_0, end = var_6225_end_0, end_mask = var_6225_end_mask_0, x = q_31_cast);
            tensor<int32, [4]> var_6229_begin_0 = const()[name = tensor<string, []>("op_6229_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_6229_end_0 = const()[name = tensor<string, []>("op_6229_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_6229_end_mask_0 = const()[name = tensor<string, []>("op_6229_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6229_cast = slice_by_index(begin = var_6229_begin_0, end = var_6229_end_0, end_mask = var_6229_end_mask_0, x = q_31_cast);
            tensor<int32, [4]> var_6233_begin_0 = const()[name = tensor<string, []>("op_6233_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_6233_end_0 = const()[name = tensor<string, []>("op_6233_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_6233_end_mask_0 = const()[name = tensor<string, []>("op_6233_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6233_cast = slice_by_index(begin = var_6233_begin_0, end = var_6233_end_0, end_mask = var_6233_end_mask_0, x = q_31_cast);
            tensor<int32, [4]> var_6237_begin_0 = const()[name = tensor<string, []>("op_6237_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_6237_end_0 = const()[name = tensor<string, []>("op_6237_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_6237_end_mask_0 = const()[name = tensor<string, []>("op_6237_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6237_cast = slice_by_index(begin = var_6237_begin_0, end = var_6237_end_0, end_mask = var_6237_end_mask_0, x = q_31_cast);
            tensor<int32, [4]> var_6241_begin_0 = const()[name = tensor<string, []>("op_6241_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_6241_end_0 = const()[name = tensor<string, []>("op_6241_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_6241_end_mask_0 = const()[name = tensor<string, []>("op_6241_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6241_cast = slice_by_index(begin = var_6241_begin_0, end = var_6241_end_0, end_mask = var_6241_end_mask_0, x = q_31_cast);
            tensor<int32, [4]> k_63_perm_0 = const()[name = tensor<string, []>("k_63_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_6248_begin_0 = const()[name = tensor<string, []>("op_6248_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_6248_end_0 = const()[name = tensor<string, []>("op_6248_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_6248_end_mask_0 = const()[name = tensor<string, []>("op_6248_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_16 = transpose(perm = k_63_perm_0, x = k_61_cast);
            tensor<fp16, [2, 77, 1, 64]> var_6248_cast = slice_by_index(begin = var_6248_begin_0, end = var_6248_end_0, end_mask = var_6248_end_mask_0, x = transpose_16);
            tensor<int32, [4]> var_6252_begin_0 = const()[name = tensor<string, []>("op_6252_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_6252_end_0 = const()[name = tensor<string, []>("op_6252_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_6252_end_mask_0 = const()[name = tensor<string, []>("op_6252_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6252_cast = slice_by_index(begin = var_6252_begin_0, end = var_6252_end_0, end_mask = var_6252_end_mask_0, x = transpose_16);
            tensor<int32, [4]> var_6256_begin_0 = const()[name = tensor<string, []>("op_6256_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_6256_end_0 = const()[name = tensor<string, []>("op_6256_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_6256_end_mask_0 = const()[name = tensor<string, []>("op_6256_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6256_cast = slice_by_index(begin = var_6256_begin_0, end = var_6256_end_0, end_mask = var_6256_end_mask_0, x = transpose_16);
            tensor<int32, [4]> var_6260_begin_0 = const()[name = tensor<string, []>("op_6260_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_6260_end_0 = const()[name = tensor<string, []>("op_6260_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_6260_end_mask_0 = const()[name = tensor<string, []>("op_6260_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6260_cast = slice_by_index(begin = var_6260_begin_0, end = var_6260_end_0, end_mask = var_6260_end_mask_0, x = transpose_16);
            tensor<int32, [4]> var_6264_begin_0 = const()[name = tensor<string, []>("op_6264_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_6264_end_0 = const()[name = tensor<string, []>("op_6264_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_6264_end_mask_0 = const()[name = tensor<string, []>("op_6264_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6264_cast = slice_by_index(begin = var_6264_begin_0, end = var_6264_end_0, end_mask = var_6264_end_mask_0, x = transpose_16);
            tensor<int32, [4]> var_6268_begin_0 = const()[name = tensor<string, []>("op_6268_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_6268_end_0 = const()[name = tensor<string, []>("op_6268_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_6268_end_mask_0 = const()[name = tensor<string, []>("op_6268_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6268_cast = slice_by_index(begin = var_6268_begin_0, end = var_6268_end_0, end_mask = var_6268_end_mask_0, x = transpose_16);
            tensor<int32, [4]> var_6272_begin_0 = const()[name = tensor<string, []>("op_6272_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_6272_end_0 = const()[name = tensor<string, []>("op_6272_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_6272_end_mask_0 = const()[name = tensor<string, []>("op_6272_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6272_cast = slice_by_index(begin = var_6272_begin_0, end = var_6272_end_0, end_mask = var_6272_end_mask_0, x = transpose_16);
            tensor<int32, [4]> var_6276_begin_0 = const()[name = tensor<string, []>("op_6276_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_6276_end_0 = const()[name = tensor<string, []>("op_6276_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_6276_end_mask_0 = const()[name = tensor<string, []>("op_6276_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6276_cast = slice_by_index(begin = var_6276_begin_0, end = var_6276_end_0, end_mask = var_6276_end_mask_0, x = transpose_16);
            tensor<int32, [4]> var_6280_begin_0 = const()[name = tensor<string, []>("op_6280_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_6280_end_0 = const()[name = tensor<string, []>("op_6280_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_6280_end_mask_0 = const()[name = tensor<string, []>("op_6280_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6280_cast = slice_by_index(begin = var_6280_begin_0, end = var_6280_end_0, end_mask = var_6280_end_mask_0, x = transpose_16);
            tensor<int32, [4]> var_6284_begin_0 = const()[name = tensor<string, []>("op_6284_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_6284_end_0 = const()[name = tensor<string, []>("op_6284_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_6284_end_mask_0 = const()[name = tensor<string, []>("op_6284_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6284_cast = slice_by_index(begin = var_6284_begin_0, end = var_6284_end_0, end_mask = var_6284_end_mask_0, x = transpose_16);
            tensor<int32, [4]> var_6288_begin_0 = const()[name = tensor<string, []>("op_6288_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_6288_end_0 = const()[name = tensor<string, []>("op_6288_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_6288_end_mask_0 = const()[name = tensor<string, []>("op_6288_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6288_cast = slice_by_index(begin = var_6288_begin_0, end = var_6288_end_0, end_mask = var_6288_end_mask_0, x = transpose_16);
            tensor<int32, [4]> var_6292_begin_0 = const()[name = tensor<string, []>("op_6292_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_6292_end_0 = const()[name = tensor<string, []>("op_6292_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_6292_end_mask_0 = const()[name = tensor<string, []>("op_6292_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6292_cast = slice_by_index(begin = var_6292_begin_0, end = var_6292_end_0, end_mask = var_6292_end_mask_0, x = transpose_16);
            tensor<int32, [4]> var_6296_begin_0 = const()[name = tensor<string, []>("op_6296_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_6296_end_0 = const()[name = tensor<string, []>("op_6296_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_6296_end_mask_0 = const()[name = tensor<string, []>("op_6296_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6296_cast = slice_by_index(begin = var_6296_begin_0, end = var_6296_end_0, end_mask = var_6296_end_mask_0, x = transpose_16);
            tensor<int32, [4]> var_6300_begin_0 = const()[name = tensor<string, []>("op_6300_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_6300_end_0 = const()[name = tensor<string, []>("op_6300_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_6300_end_mask_0 = const()[name = tensor<string, []>("op_6300_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6300_cast = slice_by_index(begin = var_6300_begin_0, end = var_6300_end_0, end_mask = var_6300_end_mask_0, x = transpose_16);
            tensor<int32, [4]> var_6304_begin_0 = const()[name = tensor<string, []>("op_6304_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_6304_end_0 = const()[name = tensor<string, []>("op_6304_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_6304_end_mask_0 = const()[name = tensor<string, []>("op_6304_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6304_cast = slice_by_index(begin = var_6304_begin_0, end = var_6304_end_0, end_mask = var_6304_end_mask_0, x = transpose_16);
            tensor<int32, [4]> var_6308_begin_0 = const()[name = tensor<string, []>("op_6308_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_6308_end_0 = const()[name = tensor<string, []>("op_6308_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_6308_end_mask_0 = const()[name = tensor<string, []>("op_6308_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6308_cast = slice_by_index(begin = var_6308_begin_0, end = var_6308_end_0, end_mask = var_6308_end_mask_0, x = transpose_16);
            tensor<int32, [4]> var_6312_begin_0 = const()[name = tensor<string, []>("op_6312_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_6312_end_0 = const()[name = tensor<string, []>("op_6312_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_6312_end_mask_0 = const()[name = tensor<string, []>("op_6312_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6312_cast = slice_by_index(begin = var_6312_begin_0, end = var_6312_end_0, end_mask = var_6312_end_mask_0, x = transpose_16);
            tensor<int32, [4]> var_6316_begin_0 = const()[name = tensor<string, []>("op_6316_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_6316_end_0 = const()[name = tensor<string, []>("op_6316_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_6316_end_mask_0 = const()[name = tensor<string, []>("op_6316_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6316_cast = slice_by_index(begin = var_6316_begin_0, end = var_6316_end_0, end_mask = var_6316_end_mask_0, x = transpose_16);
            tensor<int32, [4]> var_6320_begin_0 = const()[name = tensor<string, []>("op_6320_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_6320_end_0 = const()[name = tensor<string, []>("op_6320_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_6320_end_mask_0 = const()[name = tensor<string, []>("op_6320_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6320_cast = slice_by_index(begin = var_6320_begin_0, end = var_6320_end_0, end_mask = var_6320_end_mask_0, x = transpose_16);
            tensor<int32, [4]> var_6324_begin_0 = const()[name = tensor<string, []>("op_6324_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_6324_end_0 = const()[name = tensor<string, []>("op_6324_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_6324_end_mask_0 = const()[name = tensor<string, []>("op_6324_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6324_cast = slice_by_index(begin = var_6324_begin_0, end = var_6324_end_0, end_mask = var_6324_end_mask_0, x = transpose_16);
            tensor<int32, [4]> var_6326_begin_0 = const()[name = tensor<string, []>("op_6326_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_6326_end_0 = const()[name = tensor<string, []>("op_6326_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_6326_end_mask_0 = const()[name = tensor<string, []>("op_6326_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6326_cast = slice_by_index(begin = var_6326_begin_0, end = var_6326_end_0, end_mask = var_6326_end_mask_0, x = v_31_cast);
            tensor<int32, [4]> var_6330_begin_0 = const()[name = tensor<string, []>("op_6330_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_6330_end_0 = const()[name = tensor<string, []>("op_6330_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_6330_end_mask_0 = const()[name = tensor<string, []>("op_6330_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6330_cast = slice_by_index(begin = var_6330_begin_0, end = var_6330_end_0, end_mask = var_6330_end_mask_0, x = v_31_cast);
            tensor<int32, [4]> var_6334_begin_0 = const()[name = tensor<string, []>("op_6334_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_6334_end_0 = const()[name = tensor<string, []>("op_6334_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_6334_end_mask_0 = const()[name = tensor<string, []>("op_6334_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6334_cast = slice_by_index(begin = var_6334_begin_0, end = var_6334_end_0, end_mask = var_6334_end_mask_0, x = v_31_cast);
            tensor<int32, [4]> var_6338_begin_0 = const()[name = tensor<string, []>("op_6338_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_6338_end_0 = const()[name = tensor<string, []>("op_6338_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_6338_end_mask_0 = const()[name = tensor<string, []>("op_6338_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6338_cast = slice_by_index(begin = var_6338_begin_0, end = var_6338_end_0, end_mask = var_6338_end_mask_0, x = v_31_cast);
            tensor<int32, [4]> var_6342_begin_0 = const()[name = tensor<string, []>("op_6342_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_6342_end_0 = const()[name = tensor<string, []>("op_6342_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_6342_end_mask_0 = const()[name = tensor<string, []>("op_6342_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6342_cast = slice_by_index(begin = var_6342_begin_0, end = var_6342_end_0, end_mask = var_6342_end_mask_0, x = v_31_cast);
            tensor<int32, [4]> var_6346_begin_0 = const()[name = tensor<string, []>("op_6346_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_6346_end_0 = const()[name = tensor<string, []>("op_6346_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_6346_end_mask_0 = const()[name = tensor<string, []>("op_6346_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6346_cast = slice_by_index(begin = var_6346_begin_0, end = var_6346_end_0, end_mask = var_6346_end_mask_0, x = v_31_cast);
            tensor<int32, [4]> var_6350_begin_0 = const()[name = tensor<string, []>("op_6350_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_6350_end_0 = const()[name = tensor<string, []>("op_6350_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_6350_end_mask_0 = const()[name = tensor<string, []>("op_6350_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6350_cast = slice_by_index(begin = var_6350_begin_0, end = var_6350_end_0, end_mask = var_6350_end_mask_0, x = v_31_cast);
            tensor<int32, [4]> var_6354_begin_0 = const()[name = tensor<string, []>("op_6354_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_6354_end_0 = const()[name = tensor<string, []>("op_6354_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_6354_end_mask_0 = const()[name = tensor<string, []>("op_6354_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6354_cast = slice_by_index(begin = var_6354_begin_0, end = var_6354_end_0, end_mask = var_6354_end_mask_0, x = v_31_cast);
            tensor<int32, [4]> var_6358_begin_0 = const()[name = tensor<string, []>("op_6358_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_6358_end_0 = const()[name = tensor<string, []>("op_6358_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_6358_end_mask_0 = const()[name = tensor<string, []>("op_6358_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6358_cast = slice_by_index(begin = var_6358_begin_0, end = var_6358_end_0, end_mask = var_6358_end_mask_0, x = v_31_cast);
            tensor<int32, [4]> var_6362_begin_0 = const()[name = tensor<string, []>("op_6362_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_6362_end_0 = const()[name = tensor<string, []>("op_6362_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_6362_end_mask_0 = const()[name = tensor<string, []>("op_6362_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6362_cast = slice_by_index(begin = var_6362_begin_0, end = var_6362_end_0, end_mask = var_6362_end_mask_0, x = v_31_cast);
            tensor<int32, [4]> var_6366_begin_0 = const()[name = tensor<string, []>("op_6366_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_6366_end_0 = const()[name = tensor<string, []>("op_6366_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_6366_end_mask_0 = const()[name = tensor<string, []>("op_6366_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6366_cast = slice_by_index(begin = var_6366_begin_0, end = var_6366_end_0, end_mask = var_6366_end_mask_0, x = v_31_cast);
            tensor<int32, [4]> var_6370_begin_0 = const()[name = tensor<string, []>("op_6370_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_6370_end_0 = const()[name = tensor<string, []>("op_6370_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_6370_end_mask_0 = const()[name = tensor<string, []>("op_6370_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6370_cast = slice_by_index(begin = var_6370_begin_0, end = var_6370_end_0, end_mask = var_6370_end_mask_0, x = v_31_cast);
            tensor<int32, [4]> var_6374_begin_0 = const()[name = tensor<string, []>("op_6374_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_6374_end_0 = const()[name = tensor<string, []>("op_6374_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_6374_end_mask_0 = const()[name = tensor<string, []>("op_6374_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6374_cast = slice_by_index(begin = var_6374_begin_0, end = var_6374_end_0, end_mask = var_6374_end_mask_0, x = v_31_cast);
            tensor<int32, [4]> var_6378_begin_0 = const()[name = tensor<string, []>("op_6378_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_6378_end_0 = const()[name = tensor<string, []>("op_6378_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_6378_end_mask_0 = const()[name = tensor<string, []>("op_6378_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6378_cast = slice_by_index(begin = var_6378_begin_0, end = var_6378_end_0, end_mask = var_6378_end_mask_0, x = v_31_cast);
            tensor<int32, [4]> var_6382_begin_0 = const()[name = tensor<string, []>("op_6382_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_6382_end_0 = const()[name = tensor<string, []>("op_6382_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_6382_end_mask_0 = const()[name = tensor<string, []>("op_6382_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6382_cast = slice_by_index(begin = var_6382_begin_0, end = var_6382_end_0, end_mask = var_6382_end_mask_0, x = v_31_cast);
            tensor<int32, [4]> var_6386_begin_0 = const()[name = tensor<string, []>("op_6386_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_6386_end_0 = const()[name = tensor<string, []>("op_6386_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_6386_end_mask_0 = const()[name = tensor<string, []>("op_6386_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6386_cast = slice_by_index(begin = var_6386_begin_0, end = var_6386_end_0, end_mask = var_6386_end_mask_0, x = v_31_cast);
            tensor<int32, [4]> var_6390_begin_0 = const()[name = tensor<string, []>("op_6390_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_6390_end_0 = const()[name = tensor<string, []>("op_6390_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_6390_end_mask_0 = const()[name = tensor<string, []>("op_6390_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6390_cast = slice_by_index(begin = var_6390_begin_0, end = var_6390_end_0, end_mask = var_6390_end_mask_0, x = v_31_cast);
            tensor<int32, [4]> var_6394_begin_0 = const()[name = tensor<string, []>("op_6394_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_6394_end_0 = const()[name = tensor<string, []>("op_6394_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_6394_end_mask_0 = const()[name = tensor<string, []>("op_6394_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6394_cast = slice_by_index(begin = var_6394_begin_0, end = var_6394_end_0, end_mask = var_6394_end_mask_0, x = v_31_cast);
            tensor<int32, [4]> var_6398_begin_0 = const()[name = tensor<string, []>("op_6398_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_6398_end_0 = const()[name = tensor<string, []>("op_6398_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_6398_end_mask_0 = const()[name = tensor<string, []>("op_6398_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6398_cast = slice_by_index(begin = var_6398_begin_0, end = var_6398_end_0, end_mask = var_6398_end_mask_0, x = v_31_cast);
            tensor<int32, [4]> var_6402_begin_0 = const()[name = tensor<string, []>("op_6402_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_6402_end_0 = const()[name = tensor<string, []>("op_6402_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_6402_end_mask_0 = const()[name = tensor<string, []>("op_6402_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6402_cast = slice_by_index(begin = var_6402_begin_0, end = var_6402_end_0, end_mask = var_6402_end_mask_0, x = v_31_cast);
            tensor<string, []> var_6406_equation_0 = const()[name = tensor<string, []>("op_6406_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_6406_cast = einsum(equation = var_6406_equation_0, values = (var_6248_cast, var_6165_cast));
            tensor<fp16, []> var_6407_to_fp16 = const()[name = tensor<string, []>("op_6407_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_401_cast = mul(x = var_6406_cast, y = var_6407_to_fp16);
            tensor<string, []> var_6410_equation_0 = const()[name = tensor<string, []>("op_6410_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_6410_cast = einsum(equation = var_6410_equation_0, values = (var_6252_cast, var_6169_cast));
            tensor<fp16, []> var_6411_to_fp16 = const()[name = tensor<string, []>("op_6411_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_403_cast = mul(x = var_6410_cast, y = var_6411_to_fp16);
            tensor<string, []> var_6414_equation_0 = const()[name = tensor<string, []>("op_6414_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_6414_cast = einsum(equation = var_6414_equation_0, values = (var_6256_cast, var_6173_cast));
            tensor<fp16, []> var_6415_to_fp16 = const()[name = tensor<string, []>("op_6415_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_405_cast = mul(x = var_6414_cast, y = var_6415_to_fp16);
            tensor<string, []> var_6418_equation_0 = const()[name = tensor<string, []>("op_6418_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_6418_cast = einsum(equation = var_6418_equation_0, values = (var_6260_cast, var_6177_cast));
            tensor<fp16, []> var_6419_to_fp16 = const()[name = tensor<string, []>("op_6419_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_407_cast = mul(x = var_6418_cast, y = var_6419_to_fp16);
            tensor<string, []> var_6422_equation_0 = const()[name = tensor<string, []>("op_6422_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_6422_cast = einsum(equation = var_6422_equation_0, values = (var_6264_cast, var_6181_cast));
            tensor<fp16, []> var_6423_to_fp16 = const()[name = tensor<string, []>("op_6423_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_409_cast = mul(x = var_6422_cast, y = var_6423_to_fp16);
            tensor<string, []> var_6426_equation_0 = const()[name = tensor<string, []>("op_6426_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_6426_cast = einsum(equation = var_6426_equation_0, values = (var_6268_cast, var_6185_cast));
            tensor<fp16, []> var_6427_to_fp16 = const()[name = tensor<string, []>("op_6427_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_411_cast = mul(x = var_6426_cast, y = var_6427_to_fp16);
            tensor<string, []> var_6430_equation_0 = const()[name = tensor<string, []>("op_6430_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_6430_cast = einsum(equation = var_6430_equation_0, values = (var_6272_cast, var_6189_cast));
            tensor<fp16, []> var_6431_to_fp16 = const()[name = tensor<string, []>("op_6431_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_413_cast = mul(x = var_6430_cast, y = var_6431_to_fp16);
            tensor<string, []> var_6434_equation_0 = const()[name = tensor<string, []>("op_6434_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_6434_cast = einsum(equation = var_6434_equation_0, values = (var_6276_cast, var_6193_cast));
            tensor<fp16, []> var_6435_to_fp16 = const()[name = tensor<string, []>("op_6435_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_415_cast = mul(x = var_6434_cast, y = var_6435_to_fp16);
            tensor<string, []> var_6438_equation_0 = const()[name = tensor<string, []>("op_6438_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_6438_cast = einsum(equation = var_6438_equation_0, values = (var_6280_cast, var_6197_cast));
            tensor<fp16, []> var_6439_to_fp16 = const()[name = tensor<string, []>("op_6439_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_417_cast = mul(x = var_6438_cast, y = var_6439_to_fp16);
            tensor<string, []> var_6442_equation_0 = const()[name = tensor<string, []>("op_6442_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_6442_cast = einsum(equation = var_6442_equation_0, values = (var_6284_cast, var_6201_cast));
            tensor<fp16, []> var_6443_to_fp16 = const()[name = tensor<string, []>("op_6443_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_419_cast = mul(x = var_6442_cast, y = var_6443_to_fp16);
            tensor<string, []> var_6446_equation_0 = const()[name = tensor<string, []>("op_6446_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_6446_cast = einsum(equation = var_6446_equation_0, values = (var_6288_cast, var_6205_cast));
            tensor<fp16, []> var_6447_to_fp16 = const()[name = tensor<string, []>("op_6447_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_421_cast = mul(x = var_6446_cast, y = var_6447_to_fp16);
            tensor<string, []> var_6450_equation_0 = const()[name = tensor<string, []>("op_6450_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_6450_cast = einsum(equation = var_6450_equation_0, values = (var_6292_cast, var_6209_cast));
            tensor<fp16, []> var_6451_to_fp16 = const()[name = tensor<string, []>("op_6451_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_423_cast = mul(x = var_6450_cast, y = var_6451_to_fp16);
            tensor<string, []> var_6454_equation_0 = const()[name = tensor<string, []>("op_6454_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_6454_cast = einsum(equation = var_6454_equation_0, values = (var_6296_cast, var_6213_cast));
            tensor<fp16, []> var_6455_to_fp16 = const()[name = tensor<string, []>("op_6455_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_425_cast = mul(x = var_6454_cast, y = var_6455_to_fp16);
            tensor<string, []> var_6458_equation_0 = const()[name = tensor<string, []>("op_6458_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_6458_cast = einsum(equation = var_6458_equation_0, values = (var_6300_cast, var_6217_cast));
            tensor<fp16, []> var_6459_to_fp16 = const()[name = tensor<string, []>("op_6459_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_427_cast = mul(x = var_6458_cast, y = var_6459_to_fp16);
            tensor<string, []> var_6462_equation_0 = const()[name = tensor<string, []>("op_6462_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_6462_cast = einsum(equation = var_6462_equation_0, values = (var_6304_cast, var_6221_cast));
            tensor<fp16, []> var_6463_to_fp16 = const()[name = tensor<string, []>("op_6463_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_429_cast = mul(x = var_6462_cast, y = var_6463_to_fp16);
            tensor<string, []> var_6466_equation_0 = const()[name = tensor<string, []>("op_6466_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_6466_cast = einsum(equation = var_6466_equation_0, values = (var_6308_cast, var_6225_cast));
            tensor<fp16, []> var_6467_to_fp16 = const()[name = tensor<string, []>("op_6467_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_431_cast = mul(x = var_6466_cast, y = var_6467_to_fp16);
            tensor<string, []> var_6470_equation_0 = const()[name = tensor<string, []>("op_6470_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_6470_cast = einsum(equation = var_6470_equation_0, values = (var_6312_cast, var_6229_cast));
            tensor<fp16, []> var_6471_to_fp16 = const()[name = tensor<string, []>("op_6471_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_433_cast = mul(x = var_6470_cast, y = var_6471_to_fp16);
            tensor<string, []> var_6474_equation_0 = const()[name = tensor<string, []>("op_6474_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_6474_cast = einsum(equation = var_6474_equation_0, values = (var_6316_cast, var_6233_cast));
            tensor<fp16, []> var_6475_to_fp16 = const()[name = tensor<string, []>("op_6475_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_435_cast = mul(x = var_6474_cast, y = var_6475_to_fp16);
            tensor<string, []> var_6478_equation_0 = const()[name = tensor<string, []>("op_6478_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_6478_cast = einsum(equation = var_6478_equation_0, values = (var_6320_cast, var_6237_cast));
            tensor<fp16, []> var_6479_to_fp16 = const()[name = tensor<string, []>("op_6479_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_437_cast = mul(x = var_6478_cast, y = var_6479_to_fp16);
            tensor<string, []> var_6482_equation_0 = const()[name = tensor<string, []>("op_6482_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_6482_cast = einsum(equation = var_6482_equation_0, values = (var_6324_cast, var_6241_cast));
            tensor<fp16, []> var_6483_to_fp16 = const()[name = tensor<string, []>("op_6483_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_439_cast = mul(x = var_6482_cast, y = var_6483_to_fp16);
            tensor<fp16, [2, 77, 1, 256]> var_6485_cast = softmax(axis = var_5599, x = aw_401_cast);
            tensor<fp16, [2, 77, 1, 256]> var_6486_cast = softmax(axis = var_5599, x = aw_403_cast);
            tensor<fp16, [2, 77, 1, 256]> var_6487_cast = softmax(axis = var_5599, x = aw_405_cast);
            tensor<fp16, [2, 77, 1, 256]> var_6488_cast = softmax(axis = var_5599, x = aw_407_cast);
            tensor<fp16, [2, 77, 1, 256]> var_6489_cast = softmax(axis = var_5599, x = aw_409_cast);
            tensor<fp16, [2, 77, 1, 256]> var_6490_cast = softmax(axis = var_5599, x = aw_411_cast);
            tensor<fp16, [2, 77, 1, 256]> var_6491_cast = softmax(axis = var_5599, x = aw_413_cast);
            tensor<fp16, [2, 77, 1, 256]> var_6492_cast = softmax(axis = var_5599, x = aw_415_cast);
            tensor<fp16, [2, 77, 1, 256]> var_6493_cast = softmax(axis = var_5599, x = aw_417_cast);
            tensor<fp16, [2, 77, 1, 256]> var_6494_cast = softmax(axis = var_5599, x = aw_419_cast);
            tensor<fp16, [2, 77, 1, 256]> var_6495_cast = softmax(axis = var_5599, x = aw_421_cast);
            tensor<fp16, [2, 77, 1, 256]> var_6496_cast = softmax(axis = var_5599, x = aw_423_cast);
            tensor<fp16, [2, 77, 1, 256]> var_6497_cast = softmax(axis = var_5599, x = aw_425_cast);
            tensor<fp16, [2, 77, 1, 256]> var_6498_cast = softmax(axis = var_5599, x = aw_427_cast);
            tensor<fp16, [2, 77, 1, 256]> var_6499_cast = softmax(axis = var_5599, x = aw_429_cast);
            tensor<fp16, [2, 77, 1, 256]> var_6500_cast = softmax(axis = var_5599, x = aw_431_cast);
            tensor<fp16, [2, 77, 1, 256]> var_6501_cast = softmax(axis = var_5599, x = aw_433_cast);
            tensor<fp16, [2, 77, 1, 256]> var_6502_cast = softmax(axis = var_5599, x = aw_435_cast);
            tensor<fp16, [2, 77, 1, 256]> var_6503_cast = softmax(axis = var_5599, x = aw_437_cast);
            tensor<fp16, [2, 77, 1, 256]> var_6504_cast = softmax(axis = var_5599, x = aw_439_cast);
            tensor<string, []> var_6506_equation_0 = const()[name = tensor<string, []>("op_6506_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6506_cast = einsum(equation = var_6506_equation_0, values = (var_6326_cast, var_6485_cast));
            tensor<string, []> var_6508_equation_0 = const()[name = tensor<string, []>("op_6508_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6508_cast = einsum(equation = var_6508_equation_0, values = (var_6330_cast, var_6486_cast));
            tensor<string, []> var_6510_equation_0 = const()[name = tensor<string, []>("op_6510_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6510_cast = einsum(equation = var_6510_equation_0, values = (var_6334_cast, var_6487_cast));
            tensor<string, []> var_6512_equation_0 = const()[name = tensor<string, []>("op_6512_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6512_cast = einsum(equation = var_6512_equation_0, values = (var_6338_cast, var_6488_cast));
            tensor<string, []> var_6514_equation_0 = const()[name = tensor<string, []>("op_6514_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6514_cast = einsum(equation = var_6514_equation_0, values = (var_6342_cast, var_6489_cast));
            tensor<string, []> var_6516_equation_0 = const()[name = tensor<string, []>("op_6516_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6516_cast = einsum(equation = var_6516_equation_0, values = (var_6346_cast, var_6490_cast));
            tensor<string, []> var_6518_equation_0 = const()[name = tensor<string, []>("op_6518_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6518_cast = einsum(equation = var_6518_equation_0, values = (var_6350_cast, var_6491_cast));
            tensor<string, []> var_6520_equation_0 = const()[name = tensor<string, []>("op_6520_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6520_cast = einsum(equation = var_6520_equation_0, values = (var_6354_cast, var_6492_cast));
            tensor<string, []> var_6522_equation_0 = const()[name = tensor<string, []>("op_6522_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6522_cast = einsum(equation = var_6522_equation_0, values = (var_6358_cast, var_6493_cast));
            tensor<string, []> var_6524_equation_0 = const()[name = tensor<string, []>("op_6524_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6524_cast = einsum(equation = var_6524_equation_0, values = (var_6362_cast, var_6494_cast));
            tensor<string, []> var_6526_equation_0 = const()[name = tensor<string, []>("op_6526_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6526_cast = einsum(equation = var_6526_equation_0, values = (var_6366_cast, var_6495_cast));
            tensor<string, []> var_6528_equation_0 = const()[name = tensor<string, []>("op_6528_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6528_cast = einsum(equation = var_6528_equation_0, values = (var_6370_cast, var_6496_cast));
            tensor<string, []> var_6530_equation_0 = const()[name = tensor<string, []>("op_6530_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6530_cast = einsum(equation = var_6530_equation_0, values = (var_6374_cast, var_6497_cast));
            tensor<string, []> var_6532_equation_0 = const()[name = tensor<string, []>("op_6532_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6532_cast = einsum(equation = var_6532_equation_0, values = (var_6378_cast, var_6498_cast));
            tensor<string, []> var_6534_equation_0 = const()[name = tensor<string, []>("op_6534_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6534_cast = einsum(equation = var_6534_equation_0, values = (var_6382_cast, var_6499_cast));
            tensor<string, []> var_6536_equation_0 = const()[name = tensor<string, []>("op_6536_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6536_cast = einsum(equation = var_6536_equation_0, values = (var_6386_cast, var_6500_cast));
            tensor<string, []> var_6538_equation_0 = const()[name = tensor<string, []>("op_6538_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6538_cast = einsum(equation = var_6538_equation_0, values = (var_6390_cast, var_6501_cast));
            tensor<string, []> var_6540_equation_0 = const()[name = tensor<string, []>("op_6540_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6540_cast = einsum(equation = var_6540_equation_0, values = (var_6394_cast, var_6502_cast));
            tensor<string, []> var_6542_equation_0 = const()[name = tensor<string, []>("op_6542_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6542_cast = einsum(equation = var_6542_equation_0, values = (var_6398_cast, var_6503_cast));
            tensor<string, []> var_6544_equation_0 = const()[name = tensor<string, []>("op_6544_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_6544_cast = einsum(equation = var_6544_equation_0, values = (var_6402_cast, var_6504_cast));
            tensor<bool, []> input_303_interleave_0 = const()[name = tensor<string, []>("input_303_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 256]> input_303_cast = concat(axis = var_5599, interleave = input_303_interleave_0, values = (var_6506_cast, var_6508_cast, var_6510_cast, var_6512_cast, var_6514_cast, var_6516_cast, var_6518_cast, var_6520_cast, var_6522_cast, var_6524_cast, var_6526_cast, var_6528_cast, var_6530_cast, var_6532_cast, var_6534_cast, var_6536_cast, var_6538_cast, var_6540_cast, var_6542_cast, var_6544_cast));
            tensor<int32, [2]> var_6550 = const()[name = tensor<string, []>("op_6550"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6552 = const()[name = tensor<string, []>("op_6552"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_6554_pad_type_0 = const()[name = tensor<string, []>("op_6554_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_6554_pad_0 = const()[name = tensor<string, []>("op_6554_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1151251392)))];
            tensor<fp16, [1280]> up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1154528256)))];
            tensor<fp16, [2, 1280, 1, 256]> var_6554_cast = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_6552, groups = var_5599, pad = var_6554_pad_0, pad_type = var_6554_pad_type_0, strides = var_6550, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_303_cast);
            tensor<fp16, [2, 1280, 1, 256]> inputs_47_cast = add(x = var_6554_cast, y = inputs_45_cast);
            tensor<int32, [1]> var_6558 = const()[name = tensor<string, []>("op_6558"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_47_cast = reduce_mean(axes = var_6558, keep_dims = var_5594, x = inputs_47_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_47_cast = sub(x = inputs_47_cast, y = channels_mean_47_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_sq_47_cast = mul(x = zero_mean_47_cast, y = zero_mean_47_cast);
            tensor<int32, [1]> var_6562 = const()[name = tensor<string, []>("op_6562"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_6563_cast = reduce_mean(axes = var_6562, keep_dims = var_5594, x = zero_mean_sq_47_cast);
            tensor<fp16, []> var_6564_to_fp16 = const()[name = tensor<string, []>("op_6564_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_6565_cast = add(x = var_6563_cast, y = var_6564_to_fp16);
            tensor<fp16, []> denom_47_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_47_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 256]> denom_47_cast = rsqrt(epsilon = denom_47_epsilon_0_to_fp16, x = var_6565_cast);
            tensor<fp16, [2, 1280, 1, 256]> out_47_cast = mul(x = zero_mean_47_cast, y = denom_47_cast);
            tensor<fp16, [1, 1280, 1, 1]> var_6569_to_fp16 = const()[name = tensor<string, []>("op_6569_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1154530880)))];
            tensor<fp16, [2, 1280, 1, 256]> var_6570_cast = add(x = out_47_cast, y = var_6569_to_fp16);
            tensor<fp16, [1, 1280, 1, 1]> var_6572_to_fp16 = const()[name = tensor<string, []>("op_6572_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1154533504)))];
            tensor<fp16, [2, 1280, 1, 256]> input_305_cast = mul(x = var_6570_cast, y = var_6572_to_fp16);
            tensor<int32, [2]> var_6580 = const()[name = tensor<string, []>("op_6580"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6582 = const()[name = tensor<string, []>("op_6582"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_6584_pad_type_0 = const()[name = tensor<string, []>("op_6584_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_6584_pad_0 = const()[name = tensor<string, []>("op_6584_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1154536128)))];
            tensor<fp16, [10240]> up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1180750592)))];
            tensor<fp16, [2, 10240, 1, 256]> var_6584_cast = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_6582, groups = var_5599, pad = var_6584_pad_0, pad_type = var_6584_pad_type_0, strides = var_6580, weight = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_305_cast);
            tensor<int32, [2]> var_6585_split_sizes_0 = const()[name = tensor<string, []>("op_6585_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_6585_axis_0 = const()[name = tensor<string, []>("op_6585_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 256]> var_6585_cast_0, tensor<fp16, [2, 5120, 1, 256]> var_6585_cast_1 = split(axis = var_6585_axis_0, split_sizes = var_6585_split_sizes_0, x = var_6584_cast);
            tensor<string, []> var_6587_mode_0 = const()[name = tensor<string, []>("op_6587_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 256]> var_6587_cast = gelu(mode = var_6587_mode_0, x = var_6585_cast_1);
            tensor<fp16, [2, 5120, 1, 256]> input_307_cast = mul(x = var_6585_cast_0, y = var_6587_cast);
            tensor<int32, [2]> var_6591 = const()[name = tensor<string, []>("op_6591"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6593 = const()[name = tensor<string, []>("op_6593"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_6595_pad_type_0 = const()[name = tensor<string, []>("op_6595_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_6595_pad_0 = const()[name = tensor<string, []>("op_6595_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1180771136)))];
            tensor<fp16, [1280]> up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1193878400)))];
            tensor<fp16, [2, 1280, 1, 256]> var_6595_cast = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_6593, groups = var_5599, pad = var_6595_pad_0, pad_type = var_6595_pad_type_0, strides = var_6591, weight = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_307_cast);
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_173_cast = add(x = var_6595_cast, y = inputs_47_cast);
            tensor<int32, [4]> var_6597 = const()[name = tensor<string, []>("op_6597"), val = tensor<int32, [4]>([2, 1280, 16, 16])];
            tensor<fp16, [2, 1280, 16, 16]> input_309_cast = reshape(shape = var_6597, x = hidden_states_173_cast);
            tensor<int32, [2]> var_6601 = const()[name = tensor<string, []>("op_6601"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6603 = const()[name = tensor<string, []>("op_6603"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_175_pad_type_0 = const()[name = tensor<string, []>("hidden_states_175_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_175_pad_0 = const()[name = tensor<string, []>("hidden_states_175_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_1_attentions_0_proj_out_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_proj_out_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1193881024)))];
            tensor<fp16, [1280]> up_blocks_1_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_0_proj_out_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1197157888)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_175_cast = conv(bias = up_blocks_1_attentions_0_proj_out_bias_to_fp16, dilations = var_6603, groups = var_5599, pad = hidden_states_175_pad_0, pad_type = hidden_states_175_pad_type_0, strides = var_6601, weight = up_blocks_1_attentions_0_proj_out_weight_to_fp16, x = input_309_cast);
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_177_cast = add(x = hidden_states_175_cast, y = hidden_states_163_cast);
            tensor<bool, []> input_311_interleave_0 = const()[name = tensor<string, []>("input_311_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 2560, 16, 16]> input_311_cast = concat(axis = var_5599, interleave = input_311_interleave_0, values = (hidden_states_177_cast, input_143_cast));
            tensor<int32, [5]> reshape_144_shape_0 = const()[name = tensor<string, []>("reshape_144_shape_0"), val = tensor<int32, [5]>([2, 32, 80, 16, 16])];
            tensor<fp16, [2, 32, 80, 16, 16]> reshape_144_cast = reshape(shape = reshape_144_shape_0, x = input_311_cast);
            tensor<int32, [3]> reduce_mean_108_axes_0 = const()[name = tensor<string, []>("reduce_mean_108_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_108_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_108_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_108_cast = reduce_mean(axes = reduce_mean_108_axes_0, keep_dims = reduce_mean_108_keep_dims_0, x = reshape_144_cast);
            tensor<fp16, [2, 32, 80, 16, 16]> sub_72_cast = sub(x = reshape_144_cast, y = reduce_mean_108_cast);
            tensor<fp16, [2, 32, 80, 16, 16]> square_36_cast = square(x = sub_72_cast);
            tensor<int32, [3]> reduce_mean_110_axes_0 = const()[name = tensor<string, []>("reduce_mean_110_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_110_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_110_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_110_cast = reduce_mean(axes = reduce_mean_110_axes_0, keep_dims = reduce_mean_110_keep_dims_0, x = square_36_cast);
            tensor<fp16, []> add_72_y_0_to_fp16 = const()[name = tensor<string, []>("add_72_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_72_cast = add(x = reduce_mean_110_cast, y = add_72_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_36_cast = sqrt(x = add_72_cast);
            tensor<fp16, [2, 32, 80, 16, 16]> real_div_36_cast = real_div(x = sub_72_cast, y = sqrt_36_cast);
            tensor<int32, [4]> reshape_145_shape_0 = const()[name = tensor<string, []>("reshape_145_shape_0"), val = tensor<int32, [4]>([2, 2560, 16, 16])];
            tensor<fp16, [2, 2560, 16, 16]> reshape_145_cast = reshape(shape = reshape_145_shape_0, x = real_div_36_cast);
            tensor<fp16, [2560]> add_73_mean_0_to_fp16 = const()[name = tensor<string, []>("add_73_mean_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1197160512)))];
            tensor<fp16, [2560]> add_73_variance_0_to_fp16 = const()[name = tensor<string, []>("add_73_variance_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1197165696)))];
            tensor<fp16, [2560]> add_73_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_73_gamma_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1197170880)))];
            tensor<fp16, [2560]> add_73_beta_0_to_fp16 = const()[name = tensor<string, []>("add_73_beta_0_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1197176064)))];
            tensor<fp16, []> add_73_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_73_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 2560, 16, 16]> add_73_cast = batch_norm(beta = add_73_beta_0_to_fp16, epsilon = add_73_epsilon_0_to_fp16, gamma = add_73_gamma_0_to_fp16, mean = add_73_mean_0_to_fp16, variance = add_73_variance_0_to_fp16, x = reshape_145_cast);
            tensor<fp16, [2, 2560, 16, 16]> input_315_cast = silu(x = add_73_cast);
            tensor<int32, [2]> var_6621 = const()[name = tensor<string, []>("op_6621"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6623 = const()[name = tensor<string, []>("op_6623"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_179_pad_type_0 = const()[name = tensor<string, []>("hidden_states_179_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_179_pad_0 = const()[name = tensor<string, []>("hidden_states_179_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 2560, 3, 3]> up_blocks_1_resnets_1_conv1_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_1_conv1_weight_to_fp16"), val = tensor<fp16, [1280, 2560, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1197181248)))];
            tensor<fp16, [1280]> up_blocks_1_resnets_1_conv1_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_1_conv1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1256163712)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_179_cast = conv(bias = up_blocks_1_resnets_1_conv1_bias_to_fp16, dilations = var_6623, groups = var_5599, pad = hidden_states_179_pad_0, pad_type = hidden_states_179_pad_type_0, strides = var_6621, weight = up_blocks_1_resnets_1_conv1_weight_to_fp16, x = input_315_cast);
            tensor<int32, [2]> var_6629 = const()[name = tensor<string, []>("op_6629"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6631 = const()[name = tensor<string, []>("op_6631"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_29_pad_type_0 = const()[name = tensor<string, []>("temb_29_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_29_pad_0 = const()[name = tensor<string, []>("temb_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_1_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1256166336)))];
            tensor<fp16, [1280]> up_blocks_1_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1259443200)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_29_cast = conv(bias = up_blocks_1_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_6631, groups = var_5599, pad = temb_29_pad_0, pad_type = temb_29_pad_type_0, strides = var_6629, weight = up_blocks_1_resnets_1_time_emb_proj_weight_to_fp16, x = input_15_cast);
            tensor<fp16, [2, 1280, 16, 16]> input_319_cast = add(x = hidden_states_179_cast, y = temb_29_cast);
            tensor<int32, [5]> reshape_148_shape_0 = const()[name = tensor<string, []>("reshape_148_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 16, 16])];
            tensor<fp16, [2, 32, 40, 16, 16]> reshape_148_cast = reshape(shape = reshape_148_shape_0, x = input_319_cast);
            tensor<int32, [3]> reduce_mean_111_axes_0 = const()[name = tensor<string, []>("reduce_mean_111_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_111_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_111_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_111_cast = reduce_mean(axes = reduce_mean_111_axes_0, keep_dims = reduce_mean_111_keep_dims_0, x = reshape_148_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> sub_74_cast = sub(x = reshape_148_cast, y = reduce_mean_111_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> square_37_cast = square(x = sub_74_cast);
            tensor<int32, [3]> reduce_mean_113_axes_0 = const()[name = tensor<string, []>("reduce_mean_113_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_113_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_113_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_113_cast = reduce_mean(axes = reduce_mean_113_axes_0, keep_dims = reduce_mean_113_keep_dims_0, x = square_37_cast);
            tensor<fp16, []> add_74_y_0_to_fp16 = const()[name = tensor<string, []>("add_74_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_74_cast = add(x = reduce_mean_113_cast, y = add_74_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_37_cast = sqrt(x = add_74_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> real_div_37_cast = real_div(x = sub_74_cast, y = sqrt_37_cast);
            tensor<int32, [4]> reshape_149_shape_0 = const()[name = tensor<string, []>("reshape_149_shape_0"), val = tensor<int32, [4]>([2, 1280, 16, 16])];
            tensor<fp16, [2, 1280, 16, 16]> reshape_149_cast = reshape(shape = reshape_149_shape_0, x = real_div_37_cast);
            tensor<fp16, [1280]> add_75_mean_0_to_fp16 = const()[name = tensor<string, []>("add_75_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1259445824)))];
            tensor<fp16, [1280]> add_75_variance_0_to_fp16 = const()[name = tensor<string, []>("add_75_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1259448448)))];
            tensor<fp16, [1280]> add_75_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_75_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1259451072)))];
            tensor<fp16, [1280]> add_75_beta_0_to_fp16 = const()[name = tensor<string, []>("add_75_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1259453696)))];
            tensor<fp16, []> add_75_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_75_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 16, 16]> add_75_cast = batch_norm(beta = add_75_beta_0_to_fp16, epsilon = add_75_epsilon_0_to_fp16, gamma = add_75_gamma_0_to_fp16, mean = add_75_mean_0_to_fp16, variance = add_75_variance_0_to_fp16, x = reshape_149_cast);
            tensor<fp16, [2, 1280, 16, 16]> input_323_cast = silu(x = add_75_cast);
            tensor<int32, [2]> var_6641 = const()[name = tensor<string, []>("op_6641"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6643 = const()[name = tensor<string, []>("op_6643"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_181_pad_type_0 = const()[name = tensor<string, []>("hidden_states_181_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_181_pad_0 = const()[name = tensor<string, []>("hidden_states_181_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> up_blocks_1_resnets_1_conv2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_1_conv2_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1259456320)))];
            tensor<fp16, [1280]> up_blocks_1_resnets_1_conv2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_1_conv2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1288947584)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_181_cast = conv(bias = up_blocks_1_resnets_1_conv2_bias_to_fp16, dilations = var_6643, groups = var_5599, pad = hidden_states_181_pad_0, pad_type = hidden_states_181_pad_type_0, strides = var_6641, weight = up_blocks_1_resnets_1_conv2_weight_to_fp16, x = input_323_cast);
            tensor<int32, [2]> var_6648 = const()[name = tensor<string, []>("op_6648"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6650 = const()[name = tensor<string, []>("op_6650"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> x_13_pad_type_0 = const()[name = tensor<string, []>("x_13_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> x_13_pad_0 = const()[name = tensor<string, []>("x_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 2560, 1, 1]> up_blocks_1_resnets_1_conv_shortcut_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_1_conv_shortcut_weight_to_fp16"), val = tensor<fp16, [1280, 2560, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1288950208)))];
            tensor<fp16, [1280]> up_blocks_1_resnets_1_conv_shortcut_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_1_conv_shortcut_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1295503872)))];
            tensor<fp16, [2, 1280, 16, 16]> x_13_cast = conv(bias = up_blocks_1_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_6650, groups = var_5599, pad = x_13_pad_0, pad_type = x_13_pad_type_0, strides = var_6648, weight = up_blocks_1_resnets_1_conv_shortcut_weight_to_fp16, x = input_311_cast);
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_183_cast = add(x = x_13_cast, y = hidden_states_181_cast);
            tensor<int32, [5]> reshape_152_shape_0 = const()[name = tensor<string, []>("reshape_152_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 16, 16])];
            tensor<fp16, [2, 32, 40, 16, 16]> reshape_152_cast = reshape(shape = reshape_152_shape_0, x = hidden_states_183_cast);
            tensor<int32, [3]> reduce_mean_114_axes_0 = const()[name = tensor<string, []>("reduce_mean_114_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_114_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_114_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_114_cast = reduce_mean(axes = reduce_mean_114_axes_0, keep_dims = reduce_mean_114_keep_dims_0, x = reshape_152_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> sub_76_cast = sub(x = reshape_152_cast, y = reduce_mean_114_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> square_38_cast = square(x = sub_76_cast);
            tensor<int32, [3]> reduce_mean_116_axes_0 = const()[name = tensor<string, []>("reduce_mean_116_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_116_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_116_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_116_cast = reduce_mean(axes = reduce_mean_116_axes_0, keep_dims = reduce_mean_116_keep_dims_0, x = square_38_cast);
            tensor<fp16, []> add_76_y_0_to_fp16 = const()[name = tensor<string, []>("add_76_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_76_cast = add(x = reduce_mean_116_cast, y = add_76_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_38_cast = sqrt(x = add_76_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> real_div_38_cast = real_div(x = sub_76_cast, y = sqrt_38_cast);
            tensor<int32, [4]> reshape_153_shape_0 = const()[name = tensor<string, []>("reshape_153_shape_0"), val = tensor<int32, [4]>([2, 1280, 16, 16])];
            tensor<fp16, [2, 1280, 16, 16]> reshape_153_cast = reshape(shape = reshape_153_shape_0, x = real_div_38_cast);
            tensor<fp16, [1280]> add_77_mean_0_to_fp16 = const()[name = tensor<string, []>("add_77_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1295506496)))];
            tensor<fp16, [1280]> add_77_variance_0_to_fp16 = const()[name = tensor<string, []>("add_77_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1295509120)))];
            tensor<fp16, [1280]> add_77_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_77_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1295511744)))];
            tensor<fp16, [1280]> add_77_beta_0_to_fp16 = const()[name = tensor<string, []>("add_77_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1295514368)))];
            tensor<fp16, []> add_77_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_77_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 16, 16]> add_77_cast = batch_norm(beta = add_77_beta_0_to_fp16, epsilon = add_77_epsilon_0_to_fp16, gamma = add_77_gamma_0_to_fp16, mean = add_77_mean_0_to_fp16, variance = add_77_variance_0_to_fp16, x = reshape_153_cast);
            tensor<int32, [2]> var_6670 = const()[name = tensor<string, []>("op_6670"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6672 = const()[name = tensor<string, []>("op_6672"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_185_pad_type_0 = const()[name = tensor<string, []>("hidden_states_185_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_185_pad_0 = const()[name = tensor<string, []>("hidden_states_185_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_1_attentions_1_proj_in_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_proj_in_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1295516992)))];
            tensor<fp16, [1280]> up_blocks_1_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_proj_in_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1298793856)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_185_cast = conv(bias = up_blocks_1_attentions_1_proj_in_bias_to_fp16, dilations = var_6672, groups = var_5599, pad = hidden_states_185_pad_0, pad_type = hidden_states_185_pad_type_0, strides = var_6670, weight = up_blocks_1_attentions_1_proj_in_weight_to_fp16, x = add_77_cast);
            tensor<int32, [4]> var_6677 = const()[name = tensor<string, []>("op_6677"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<fp16, [2, 1280, 1, 256]> inputs_49_cast = reshape(shape = var_6677, x = hidden_states_185_cast);
            tensor<int32, [1]> var_6687 = const()[name = tensor<string, []>("op_6687"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_49_cast = reduce_mean(axes = var_6687, keep_dims = var_5594, x = inputs_49_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_49_cast = sub(x = inputs_49_cast, y = channels_mean_49_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_sq_49_cast = mul(x = zero_mean_49_cast, y = zero_mean_49_cast);
            tensor<int32, [1]> var_6691 = const()[name = tensor<string, []>("op_6691"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_6692_cast = reduce_mean(axes = var_6691, keep_dims = var_5594, x = zero_mean_sq_49_cast);
            tensor<fp16, []> var_6693_to_fp16 = const()[name = tensor<string, []>("op_6693_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_6694_cast = add(x = var_6692_cast, y = var_6693_to_fp16);
            tensor<fp16, []> denom_49_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_49_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 256]> denom_49_cast = rsqrt(epsilon = denom_49_epsilon_0_to_fp16, x = var_6694_cast);
            tensor<fp16, [2, 1280, 1, 256]> out_49_cast = mul(x = zero_mean_49_cast, y = denom_49_cast);
            tensor<fp16, [1, 1280, 1, 1]> var_6698_to_fp16 = const()[name = tensor<string, []>("op_6698_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1298796480)))];
            tensor<fp16, [2, 1280, 1, 256]> var_6699_cast = add(x = out_49_cast, y = var_6698_to_fp16);
            tensor<fp16, [1, 1280, 1, 1]> var_6701_to_fp16 = const()[name = tensor<string, []>("op_6701_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1298799104)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_187_cast = mul(x = var_6699_cast, y = var_6701_to_fp16);
            tensor<int32, [2]> var_6708 = const()[name = tensor<string, []>("op_6708"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6710 = const()[name = tensor<string, []>("op_6710"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_33_pad_type_0 = const()[name = tensor<string, []>("q_33_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_33_pad_0 = const()[name = tensor<string, []>("q_33_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1298801728)))];
            tensor<fp16, [2, 1280, 1, 256]> q_33_cast = conv(dilations = var_6710, groups = var_5599, pad = q_33_pad_0, pad_type = q_33_pad_type_0, strides = var_6708, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_187_cast);
            tensor<int32, [2]> var_6714 = const()[name = tensor<string, []>("op_6714"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6716 = const()[name = tensor<string, []>("op_6716"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_65_pad_type_0 = const()[name = tensor<string, []>("k_65_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_65_pad_0 = const()[name = tensor<string, []>("k_65_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1302078592)))];
            tensor<fp16, [2, 1280, 1, 256]> k_65_cast = conv(dilations = var_6716, groups = var_5599, pad = k_65_pad_0, pad_type = k_65_pad_type_0, strides = var_6714, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_187_cast);
            tensor<int32, [2]> var_6720 = const()[name = tensor<string, []>("op_6720"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6722 = const()[name = tensor<string, []>("op_6722"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_33_pad_type_0 = const()[name = tensor<string, []>("v_33_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_33_pad_0 = const()[name = tensor<string, []>("v_33_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1305355456)))];
            tensor<fp16, [2, 1280, 1, 256]> v_33_cast = conv(dilations = var_6722, groups = var_5599, pad = v_33_pad_0, pad_type = v_33_pad_type_0, strides = var_6720, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_187_cast);
            tensor<int32, [4]> var_6726_begin_0 = const()[name = tensor<string, []>("op_6726_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_6726_end_0 = const()[name = tensor<string, []>("op_6726_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_6726_end_mask_0 = const()[name = tensor<string, []>("op_6726_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6726_cast = slice_by_index(begin = var_6726_begin_0, end = var_6726_end_0, end_mask = var_6726_end_mask_0, x = q_33_cast);
            tensor<int32, [4]> var_6730_begin_0 = const()[name = tensor<string, []>("op_6730_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_6730_end_0 = const()[name = tensor<string, []>("op_6730_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_6730_end_mask_0 = const()[name = tensor<string, []>("op_6730_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6730_cast = slice_by_index(begin = var_6730_begin_0, end = var_6730_end_0, end_mask = var_6730_end_mask_0, x = q_33_cast);
            tensor<int32, [4]> var_6734_begin_0 = const()[name = tensor<string, []>("op_6734_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_6734_end_0 = const()[name = tensor<string, []>("op_6734_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_6734_end_mask_0 = const()[name = tensor<string, []>("op_6734_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6734_cast = slice_by_index(begin = var_6734_begin_0, end = var_6734_end_0, end_mask = var_6734_end_mask_0, x = q_33_cast);
            tensor<int32, [4]> var_6738_begin_0 = const()[name = tensor<string, []>("op_6738_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_6738_end_0 = const()[name = tensor<string, []>("op_6738_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_6738_end_mask_0 = const()[name = tensor<string, []>("op_6738_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6738_cast = slice_by_index(begin = var_6738_begin_0, end = var_6738_end_0, end_mask = var_6738_end_mask_0, x = q_33_cast);
            tensor<int32, [4]> var_6742_begin_0 = const()[name = tensor<string, []>("op_6742_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_6742_end_0 = const()[name = tensor<string, []>("op_6742_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_6742_end_mask_0 = const()[name = tensor<string, []>("op_6742_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6742_cast = slice_by_index(begin = var_6742_begin_0, end = var_6742_end_0, end_mask = var_6742_end_mask_0, x = q_33_cast);
            tensor<int32, [4]> var_6746_begin_0 = const()[name = tensor<string, []>("op_6746_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_6746_end_0 = const()[name = tensor<string, []>("op_6746_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_6746_end_mask_0 = const()[name = tensor<string, []>("op_6746_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6746_cast = slice_by_index(begin = var_6746_begin_0, end = var_6746_end_0, end_mask = var_6746_end_mask_0, x = q_33_cast);
            tensor<int32, [4]> var_6750_begin_0 = const()[name = tensor<string, []>("op_6750_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_6750_end_0 = const()[name = tensor<string, []>("op_6750_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_6750_end_mask_0 = const()[name = tensor<string, []>("op_6750_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6750_cast = slice_by_index(begin = var_6750_begin_0, end = var_6750_end_0, end_mask = var_6750_end_mask_0, x = q_33_cast);
            tensor<int32, [4]> var_6754_begin_0 = const()[name = tensor<string, []>("op_6754_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_6754_end_0 = const()[name = tensor<string, []>("op_6754_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_6754_end_mask_0 = const()[name = tensor<string, []>("op_6754_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6754_cast = slice_by_index(begin = var_6754_begin_0, end = var_6754_end_0, end_mask = var_6754_end_mask_0, x = q_33_cast);
            tensor<int32, [4]> var_6758_begin_0 = const()[name = tensor<string, []>("op_6758_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_6758_end_0 = const()[name = tensor<string, []>("op_6758_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_6758_end_mask_0 = const()[name = tensor<string, []>("op_6758_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6758_cast = slice_by_index(begin = var_6758_begin_0, end = var_6758_end_0, end_mask = var_6758_end_mask_0, x = q_33_cast);
            tensor<int32, [4]> var_6762_begin_0 = const()[name = tensor<string, []>("op_6762_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_6762_end_0 = const()[name = tensor<string, []>("op_6762_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_6762_end_mask_0 = const()[name = tensor<string, []>("op_6762_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6762_cast = slice_by_index(begin = var_6762_begin_0, end = var_6762_end_0, end_mask = var_6762_end_mask_0, x = q_33_cast);
            tensor<int32, [4]> var_6766_begin_0 = const()[name = tensor<string, []>("op_6766_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_6766_end_0 = const()[name = tensor<string, []>("op_6766_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_6766_end_mask_0 = const()[name = tensor<string, []>("op_6766_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6766_cast = slice_by_index(begin = var_6766_begin_0, end = var_6766_end_0, end_mask = var_6766_end_mask_0, x = q_33_cast);
            tensor<int32, [4]> var_6770_begin_0 = const()[name = tensor<string, []>("op_6770_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_6770_end_0 = const()[name = tensor<string, []>("op_6770_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_6770_end_mask_0 = const()[name = tensor<string, []>("op_6770_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6770_cast = slice_by_index(begin = var_6770_begin_0, end = var_6770_end_0, end_mask = var_6770_end_mask_0, x = q_33_cast);
            tensor<int32, [4]> var_6774_begin_0 = const()[name = tensor<string, []>("op_6774_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_6774_end_0 = const()[name = tensor<string, []>("op_6774_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_6774_end_mask_0 = const()[name = tensor<string, []>("op_6774_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6774_cast = slice_by_index(begin = var_6774_begin_0, end = var_6774_end_0, end_mask = var_6774_end_mask_0, x = q_33_cast);
            tensor<int32, [4]> var_6778_begin_0 = const()[name = tensor<string, []>("op_6778_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_6778_end_0 = const()[name = tensor<string, []>("op_6778_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_6778_end_mask_0 = const()[name = tensor<string, []>("op_6778_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6778_cast = slice_by_index(begin = var_6778_begin_0, end = var_6778_end_0, end_mask = var_6778_end_mask_0, x = q_33_cast);
            tensor<int32, [4]> var_6782_begin_0 = const()[name = tensor<string, []>("op_6782_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_6782_end_0 = const()[name = tensor<string, []>("op_6782_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_6782_end_mask_0 = const()[name = tensor<string, []>("op_6782_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6782_cast = slice_by_index(begin = var_6782_begin_0, end = var_6782_end_0, end_mask = var_6782_end_mask_0, x = q_33_cast);
            tensor<int32, [4]> var_6786_begin_0 = const()[name = tensor<string, []>("op_6786_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_6786_end_0 = const()[name = tensor<string, []>("op_6786_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_6786_end_mask_0 = const()[name = tensor<string, []>("op_6786_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6786_cast = slice_by_index(begin = var_6786_begin_0, end = var_6786_end_0, end_mask = var_6786_end_mask_0, x = q_33_cast);
            tensor<int32, [4]> var_6790_begin_0 = const()[name = tensor<string, []>("op_6790_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_6790_end_0 = const()[name = tensor<string, []>("op_6790_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_6790_end_mask_0 = const()[name = tensor<string, []>("op_6790_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6790_cast = slice_by_index(begin = var_6790_begin_0, end = var_6790_end_0, end_mask = var_6790_end_mask_0, x = q_33_cast);
            tensor<int32, [4]> var_6794_begin_0 = const()[name = tensor<string, []>("op_6794_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_6794_end_0 = const()[name = tensor<string, []>("op_6794_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_6794_end_mask_0 = const()[name = tensor<string, []>("op_6794_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6794_cast = slice_by_index(begin = var_6794_begin_0, end = var_6794_end_0, end_mask = var_6794_end_mask_0, x = q_33_cast);
            tensor<int32, [4]> var_6798_begin_0 = const()[name = tensor<string, []>("op_6798_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_6798_end_0 = const()[name = tensor<string, []>("op_6798_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_6798_end_mask_0 = const()[name = tensor<string, []>("op_6798_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6798_cast = slice_by_index(begin = var_6798_begin_0, end = var_6798_end_0, end_mask = var_6798_end_mask_0, x = q_33_cast);
            tensor<int32, [4]> var_6802_begin_0 = const()[name = tensor<string, []>("op_6802_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_6802_end_0 = const()[name = tensor<string, []>("op_6802_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_6802_end_mask_0 = const()[name = tensor<string, []>("op_6802_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6802_cast = slice_by_index(begin = var_6802_begin_0, end = var_6802_end_0, end_mask = var_6802_end_mask_0, x = q_33_cast);
            tensor<int32, [4]> k_67_perm_0 = const()[name = tensor<string, []>("k_67_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_6809_begin_0 = const()[name = tensor<string, []>("op_6809_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_6809_end_0 = const()[name = tensor<string, []>("op_6809_end_0"), val = tensor<int32, [4]>([2, 256, 1, 64])];
            tensor<bool, [4]> var_6809_end_mask_0 = const()[name = tensor<string, []>("op_6809_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 1280]> transpose_15 = transpose(perm = k_67_perm_0, x = k_65_cast);
            tensor<fp16, [2, 256, 1, 64]> var_6809_cast = slice_by_index(begin = var_6809_begin_0, end = var_6809_end_0, end_mask = var_6809_end_mask_0, x = transpose_15);
            tensor<int32, [4]> var_6813_begin_0 = const()[name = tensor<string, []>("op_6813_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_6813_end_0 = const()[name = tensor<string, []>("op_6813_end_0"), val = tensor<int32, [4]>([2, 256, 1, 128])];
            tensor<bool, [4]> var_6813_end_mask_0 = const()[name = tensor<string, []>("op_6813_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_6813_cast = slice_by_index(begin = var_6813_begin_0, end = var_6813_end_0, end_mask = var_6813_end_mask_0, x = transpose_15);
            tensor<int32, [4]> var_6817_begin_0 = const()[name = tensor<string, []>("op_6817_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_6817_end_0 = const()[name = tensor<string, []>("op_6817_end_0"), val = tensor<int32, [4]>([2, 256, 1, 192])];
            tensor<bool, [4]> var_6817_end_mask_0 = const()[name = tensor<string, []>("op_6817_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_6817_cast = slice_by_index(begin = var_6817_begin_0, end = var_6817_end_0, end_mask = var_6817_end_mask_0, x = transpose_15);
            tensor<int32, [4]> var_6821_begin_0 = const()[name = tensor<string, []>("op_6821_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_6821_end_0 = const()[name = tensor<string, []>("op_6821_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_6821_end_mask_0 = const()[name = tensor<string, []>("op_6821_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_6821_cast = slice_by_index(begin = var_6821_begin_0, end = var_6821_end_0, end_mask = var_6821_end_mask_0, x = transpose_15);
            tensor<int32, [4]> var_6825_begin_0 = const()[name = tensor<string, []>("op_6825_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_6825_end_0 = const()[name = tensor<string, []>("op_6825_end_0"), val = tensor<int32, [4]>([2, 256, 1, 320])];
            tensor<bool, [4]> var_6825_end_mask_0 = const()[name = tensor<string, []>("op_6825_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_6825_cast = slice_by_index(begin = var_6825_begin_0, end = var_6825_end_0, end_mask = var_6825_end_mask_0, x = transpose_15);
            tensor<int32, [4]> var_6829_begin_0 = const()[name = tensor<string, []>("op_6829_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_6829_end_0 = const()[name = tensor<string, []>("op_6829_end_0"), val = tensor<int32, [4]>([2, 256, 1, 384])];
            tensor<bool, [4]> var_6829_end_mask_0 = const()[name = tensor<string, []>("op_6829_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_6829_cast = slice_by_index(begin = var_6829_begin_0, end = var_6829_end_0, end_mask = var_6829_end_mask_0, x = transpose_15);
            tensor<int32, [4]> var_6833_begin_0 = const()[name = tensor<string, []>("op_6833_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_6833_end_0 = const()[name = tensor<string, []>("op_6833_end_0"), val = tensor<int32, [4]>([2, 256, 1, 448])];
            tensor<bool, [4]> var_6833_end_mask_0 = const()[name = tensor<string, []>("op_6833_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_6833_cast = slice_by_index(begin = var_6833_begin_0, end = var_6833_end_0, end_mask = var_6833_end_mask_0, x = transpose_15);
            tensor<int32, [4]> var_6837_begin_0 = const()[name = tensor<string, []>("op_6837_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_6837_end_0 = const()[name = tensor<string, []>("op_6837_end_0"), val = tensor<int32, [4]>([2, 256, 1, 512])];
            tensor<bool, [4]> var_6837_end_mask_0 = const()[name = tensor<string, []>("op_6837_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_6837_cast = slice_by_index(begin = var_6837_begin_0, end = var_6837_end_0, end_mask = var_6837_end_mask_0, x = transpose_15);
            tensor<int32, [4]> var_6841_begin_0 = const()[name = tensor<string, []>("op_6841_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_6841_end_0 = const()[name = tensor<string, []>("op_6841_end_0"), val = tensor<int32, [4]>([2, 256, 1, 576])];
            tensor<bool, [4]> var_6841_end_mask_0 = const()[name = tensor<string, []>("op_6841_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_6841_cast = slice_by_index(begin = var_6841_begin_0, end = var_6841_end_0, end_mask = var_6841_end_mask_0, x = transpose_15);
            tensor<int32, [4]> var_6845_begin_0 = const()[name = tensor<string, []>("op_6845_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_6845_end_0 = const()[name = tensor<string, []>("op_6845_end_0"), val = tensor<int32, [4]>([2, 256, 1, 640])];
            tensor<bool, [4]> var_6845_end_mask_0 = const()[name = tensor<string, []>("op_6845_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_6845_cast = slice_by_index(begin = var_6845_begin_0, end = var_6845_end_0, end_mask = var_6845_end_mask_0, x = transpose_15);
            tensor<int32, [4]> var_6849_begin_0 = const()[name = tensor<string, []>("op_6849_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_6849_end_0 = const()[name = tensor<string, []>("op_6849_end_0"), val = tensor<int32, [4]>([2, 256, 1, 704])];
            tensor<bool, [4]> var_6849_end_mask_0 = const()[name = tensor<string, []>("op_6849_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_6849_cast = slice_by_index(begin = var_6849_begin_0, end = var_6849_end_0, end_mask = var_6849_end_mask_0, x = transpose_15);
            tensor<int32, [4]> var_6853_begin_0 = const()[name = tensor<string, []>("op_6853_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_6853_end_0 = const()[name = tensor<string, []>("op_6853_end_0"), val = tensor<int32, [4]>([2, 256, 1, 768])];
            tensor<bool, [4]> var_6853_end_mask_0 = const()[name = tensor<string, []>("op_6853_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_6853_cast = slice_by_index(begin = var_6853_begin_0, end = var_6853_end_0, end_mask = var_6853_end_mask_0, x = transpose_15);
            tensor<int32, [4]> var_6857_begin_0 = const()[name = tensor<string, []>("op_6857_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_6857_end_0 = const()[name = tensor<string, []>("op_6857_end_0"), val = tensor<int32, [4]>([2, 256, 1, 832])];
            tensor<bool, [4]> var_6857_end_mask_0 = const()[name = tensor<string, []>("op_6857_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_6857_cast = slice_by_index(begin = var_6857_begin_0, end = var_6857_end_0, end_mask = var_6857_end_mask_0, x = transpose_15);
            tensor<int32, [4]> var_6861_begin_0 = const()[name = tensor<string, []>("op_6861_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_6861_end_0 = const()[name = tensor<string, []>("op_6861_end_0"), val = tensor<int32, [4]>([2, 256, 1, 896])];
            tensor<bool, [4]> var_6861_end_mask_0 = const()[name = tensor<string, []>("op_6861_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_6861_cast = slice_by_index(begin = var_6861_begin_0, end = var_6861_end_0, end_mask = var_6861_end_mask_0, x = transpose_15);
            tensor<int32, [4]> var_6865_begin_0 = const()[name = tensor<string, []>("op_6865_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_6865_end_0 = const()[name = tensor<string, []>("op_6865_end_0"), val = tensor<int32, [4]>([2, 256, 1, 960])];
            tensor<bool, [4]> var_6865_end_mask_0 = const()[name = tensor<string, []>("op_6865_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_6865_cast = slice_by_index(begin = var_6865_begin_0, end = var_6865_end_0, end_mask = var_6865_end_mask_0, x = transpose_15);
            tensor<int32, [4]> var_6869_begin_0 = const()[name = tensor<string, []>("op_6869_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_6869_end_0 = const()[name = tensor<string, []>("op_6869_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_6869_end_mask_0 = const()[name = tensor<string, []>("op_6869_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_6869_cast = slice_by_index(begin = var_6869_begin_0, end = var_6869_end_0, end_mask = var_6869_end_mask_0, x = transpose_15);
            tensor<int32, [4]> var_6873_begin_0 = const()[name = tensor<string, []>("op_6873_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_6873_end_0 = const()[name = tensor<string, []>("op_6873_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1088])];
            tensor<bool, [4]> var_6873_end_mask_0 = const()[name = tensor<string, []>("op_6873_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_6873_cast = slice_by_index(begin = var_6873_begin_0, end = var_6873_end_0, end_mask = var_6873_end_mask_0, x = transpose_15);
            tensor<int32, [4]> var_6877_begin_0 = const()[name = tensor<string, []>("op_6877_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_6877_end_0 = const()[name = tensor<string, []>("op_6877_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1152])];
            tensor<bool, [4]> var_6877_end_mask_0 = const()[name = tensor<string, []>("op_6877_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_6877_cast = slice_by_index(begin = var_6877_begin_0, end = var_6877_end_0, end_mask = var_6877_end_mask_0, x = transpose_15);
            tensor<int32, [4]> var_6881_begin_0 = const()[name = tensor<string, []>("op_6881_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_6881_end_0 = const()[name = tensor<string, []>("op_6881_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1216])];
            tensor<bool, [4]> var_6881_end_mask_0 = const()[name = tensor<string, []>("op_6881_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_6881_cast = slice_by_index(begin = var_6881_begin_0, end = var_6881_end_0, end_mask = var_6881_end_mask_0, x = transpose_15);
            tensor<int32, [4]> var_6885_begin_0 = const()[name = tensor<string, []>("op_6885_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_6885_end_0 = const()[name = tensor<string, []>("op_6885_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1280])];
            tensor<bool, [4]> var_6885_end_mask_0 = const()[name = tensor<string, []>("op_6885_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_6885_cast = slice_by_index(begin = var_6885_begin_0, end = var_6885_end_0, end_mask = var_6885_end_mask_0, x = transpose_15);
            tensor<int32, [4]> var_6887_begin_0 = const()[name = tensor<string, []>("op_6887_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_6887_end_0 = const()[name = tensor<string, []>("op_6887_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_6887_end_mask_0 = const()[name = tensor<string, []>("op_6887_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6887_cast = slice_by_index(begin = var_6887_begin_0, end = var_6887_end_0, end_mask = var_6887_end_mask_0, x = v_33_cast);
            tensor<int32, [4]> var_6891_begin_0 = const()[name = tensor<string, []>("op_6891_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_6891_end_0 = const()[name = tensor<string, []>("op_6891_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_6891_end_mask_0 = const()[name = tensor<string, []>("op_6891_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6891_cast = slice_by_index(begin = var_6891_begin_0, end = var_6891_end_0, end_mask = var_6891_end_mask_0, x = v_33_cast);
            tensor<int32, [4]> var_6895_begin_0 = const()[name = tensor<string, []>("op_6895_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_6895_end_0 = const()[name = tensor<string, []>("op_6895_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_6895_end_mask_0 = const()[name = tensor<string, []>("op_6895_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6895_cast = slice_by_index(begin = var_6895_begin_0, end = var_6895_end_0, end_mask = var_6895_end_mask_0, x = v_33_cast);
            tensor<int32, [4]> var_6899_begin_0 = const()[name = tensor<string, []>("op_6899_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_6899_end_0 = const()[name = tensor<string, []>("op_6899_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_6899_end_mask_0 = const()[name = tensor<string, []>("op_6899_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6899_cast = slice_by_index(begin = var_6899_begin_0, end = var_6899_end_0, end_mask = var_6899_end_mask_0, x = v_33_cast);
            tensor<int32, [4]> var_6903_begin_0 = const()[name = tensor<string, []>("op_6903_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_6903_end_0 = const()[name = tensor<string, []>("op_6903_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_6903_end_mask_0 = const()[name = tensor<string, []>("op_6903_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6903_cast = slice_by_index(begin = var_6903_begin_0, end = var_6903_end_0, end_mask = var_6903_end_mask_0, x = v_33_cast);
            tensor<int32, [4]> var_6907_begin_0 = const()[name = tensor<string, []>("op_6907_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_6907_end_0 = const()[name = tensor<string, []>("op_6907_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_6907_end_mask_0 = const()[name = tensor<string, []>("op_6907_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6907_cast = slice_by_index(begin = var_6907_begin_0, end = var_6907_end_0, end_mask = var_6907_end_mask_0, x = v_33_cast);
            tensor<int32, [4]> var_6911_begin_0 = const()[name = tensor<string, []>("op_6911_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_6911_end_0 = const()[name = tensor<string, []>("op_6911_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_6911_end_mask_0 = const()[name = tensor<string, []>("op_6911_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6911_cast = slice_by_index(begin = var_6911_begin_0, end = var_6911_end_0, end_mask = var_6911_end_mask_0, x = v_33_cast);
            tensor<int32, [4]> var_6915_begin_0 = const()[name = tensor<string, []>("op_6915_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_6915_end_0 = const()[name = tensor<string, []>("op_6915_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_6915_end_mask_0 = const()[name = tensor<string, []>("op_6915_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6915_cast = slice_by_index(begin = var_6915_begin_0, end = var_6915_end_0, end_mask = var_6915_end_mask_0, x = v_33_cast);
            tensor<int32, [4]> var_6919_begin_0 = const()[name = tensor<string, []>("op_6919_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_6919_end_0 = const()[name = tensor<string, []>("op_6919_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_6919_end_mask_0 = const()[name = tensor<string, []>("op_6919_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6919_cast = slice_by_index(begin = var_6919_begin_0, end = var_6919_end_0, end_mask = var_6919_end_mask_0, x = v_33_cast);
            tensor<int32, [4]> var_6923_begin_0 = const()[name = tensor<string, []>("op_6923_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_6923_end_0 = const()[name = tensor<string, []>("op_6923_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_6923_end_mask_0 = const()[name = tensor<string, []>("op_6923_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6923_cast = slice_by_index(begin = var_6923_begin_0, end = var_6923_end_0, end_mask = var_6923_end_mask_0, x = v_33_cast);
            tensor<int32, [4]> var_6927_begin_0 = const()[name = tensor<string, []>("op_6927_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_6927_end_0 = const()[name = tensor<string, []>("op_6927_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_6927_end_mask_0 = const()[name = tensor<string, []>("op_6927_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6927_cast = slice_by_index(begin = var_6927_begin_0, end = var_6927_end_0, end_mask = var_6927_end_mask_0, x = v_33_cast);
            tensor<int32, [4]> var_6931_begin_0 = const()[name = tensor<string, []>("op_6931_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_6931_end_0 = const()[name = tensor<string, []>("op_6931_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_6931_end_mask_0 = const()[name = tensor<string, []>("op_6931_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6931_cast = slice_by_index(begin = var_6931_begin_0, end = var_6931_end_0, end_mask = var_6931_end_mask_0, x = v_33_cast);
            tensor<int32, [4]> var_6935_begin_0 = const()[name = tensor<string, []>("op_6935_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_6935_end_0 = const()[name = tensor<string, []>("op_6935_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_6935_end_mask_0 = const()[name = tensor<string, []>("op_6935_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6935_cast = slice_by_index(begin = var_6935_begin_0, end = var_6935_end_0, end_mask = var_6935_end_mask_0, x = v_33_cast);
            tensor<int32, [4]> var_6939_begin_0 = const()[name = tensor<string, []>("op_6939_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_6939_end_0 = const()[name = tensor<string, []>("op_6939_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_6939_end_mask_0 = const()[name = tensor<string, []>("op_6939_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6939_cast = slice_by_index(begin = var_6939_begin_0, end = var_6939_end_0, end_mask = var_6939_end_mask_0, x = v_33_cast);
            tensor<int32, [4]> var_6943_begin_0 = const()[name = tensor<string, []>("op_6943_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_6943_end_0 = const()[name = tensor<string, []>("op_6943_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_6943_end_mask_0 = const()[name = tensor<string, []>("op_6943_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6943_cast = slice_by_index(begin = var_6943_begin_0, end = var_6943_end_0, end_mask = var_6943_end_mask_0, x = v_33_cast);
            tensor<int32, [4]> var_6947_begin_0 = const()[name = tensor<string, []>("op_6947_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_6947_end_0 = const()[name = tensor<string, []>("op_6947_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_6947_end_mask_0 = const()[name = tensor<string, []>("op_6947_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6947_cast = slice_by_index(begin = var_6947_begin_0, end = var_6947_end_0, end_mask = var_6947_end_mask_0, x = v_33_cast);
            tensor<int32, [4]> var_6951_begin_0 = const()[name = tensor<string, []>("op_6951_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_6951_end_0 = const()[name = tensor<string, []>("op_6951_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_6951_end_mask_0 = const()[name = tensor<string, []>("op_6951_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6951_cast = slice_by_index(begin = var_6951_begin_0, end = var_6951_end_0, end_mask = var_6951_end_mask_0, x = v_33_cast);
            tensor<int32, [4]> var_6955_begin_0 = const()[name = tensor<string, []>("op_6955_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_6955_end_0 = const()[name = tensor<string, []>("op_6955_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_6955_end_mask_0 = const()[name = tensor<string, []>("op_6955_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6955_cast = slice_by_index(begin = var_6955_begin_0, end = var_6955_end_0, end_mask = var_6955_end_mask_0, x = v_33_cast);
            tensor<int32, [4]> var_6959_begin_0 = const()[name = tensor<string, []>("op_6959_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_6959_end_0 = const()[name = tensor<string, []>("op_6959_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_6959_end_mask_0 = const()[name = tensor<string, []>("op_6959_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6959_cast = slice_by_index(begin = var_6959_begin_0, end = var_6959_end_0, end_mask = var_6959_end_mask_0, x = v_33_cast);
            tensor<int32, [4]> var_6963_begin_0 = const()[name = tensor<string, []>("op_6963_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_6963_end_0 = const()[name = tensor<string, []>("op_6963_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_6963_end_mask_0 = const()[name = tensor<string, []>("op_6963_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_6963_cast = slice_by_index(begin = var_6963_begin_0, end = var_6963_end_0, end_mask = var_6963_end_mask_0, x = v_33_cast);
            tensor<string, []> var_6967_equation_0 = const()[name = tensor<string, []>("op_6967_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6967_cast = einsum(equation = var_6967_equation_0, values = (var_6809_cast, var_6726_cast));
            tensor<fp16, []> var_6968_to_fp16 = const()[name = tensor<string, []>("op_6968_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_441_cast = mul(x = var_6967_cast, y = var_6968_to_fp16);
            tensor<string, []> var_6971_equation_0 = const()[name = tensor<string, []>("op_6971_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6971_cast = einsum(equation = var_6971_equation_0, values = (var_6813_cast, var_6730_cast));
            tensor<fp16, []> var_6972_to_fp16 = const()[name = tensor<string, []>("op_6972_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_443_cast = mul(x = var_6971_cast, y = var_6972_to_fp16);
            tensor<string, []> var_6975_equation_0 = const()[name = tensor<string, []>("op_6975_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6975_cast = einsum(equation = var_6975_equation_0, values = (var_6817_cast, var_6734_cast));
            tensor<fp16, []> var_6976_to_fp16 = const()[name = tensor<string, []>("op_6976_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_445_cast = mul(x = var_6975_cast, y = var_6976_to_fp16);
            tensor<string, []> var_6979_equation_0 = const()[name = tensor<string, []>("op_6979_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6979_cast = einsum(equation = var_6979_equation_0, values = (var_6821_cast, var_6738_cast));
            tensor<fp16, []> var_6980_to_fp16 = const()[name = tensor<string, []>("op_6980_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_447_cast = mul(x = var_6979_cast, y = var_6980_to_fp16);
            tensor<string, []> var_6983_equation_0 = const()[name = tensor<string, []>("op_6983_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6983_cast = einsum(equation = var_6983_equation_0, values = (var_6825_cast, var_6742_cast));
            tensor<fp16, []> var_6984_to_fp16 = const()[name = tensor<string, []>("op_6984_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_449_cast = mul(x = var_6983_cast, y = var_6984_to_fp16);
            tensor<string, []> var_6987_equation_0 = const()[name = tensor<string, []>("op_6987_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6987_cast = einsum(equation = var_6987_equation_0, values = (var_6829_cast, var_6746_cast));
            tensor<fp16, []> var_6988_to_fp16 = const()[name = tensor<string, []>("op_6988_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_451_cast = mul(x = var_6987_cast, y = var_6988_to_fp16);
            tensor<string, []> var_6991_equation_0 = const()[name = tensor<string, []>("op_6991_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6991_cast = einsum(equation = var_6991_equation_0, values = (var_6833_cast, var_6750_cast));
            tensor<fp16, []> var_6992_to_fp16 = const()[name = tensor<string, []>("op_6992_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_453_cast = mul(x = var_6991_cast, y = var_6992_to_fp16);
            tensor<string, []> var_6995_equation_0 = const()[name = tensor<string, []>("op_6995_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6995_cast = einsum(equation = var_6995_equation_0, values = (var_6837_cast, var_6754_cast));
            tensor<fp16, []> var_6996_to_fp16 = const()[name = tensor<string, []>("op_6996_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_455_cast = mul(x = var_6995_cast, y = var_6996_to_fp16);
            tensor<string, []> var_6999_equation_0 = const()[name = tensor<string, []>("op_6999_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6999_cast = einsum(equation = var_6999_equation_0, values = (var_6841_cast, var_6758_cast));
            tensor<fp16, []> var_7000_to_fp16 = const()[name = tensor<string, []>("op_7000_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_457_cast = mul(x = var_6999_cast, y = var_7000_to_fp16);
            tensor<string, []> var_7003_equation_0 = const()[name = tensor<string, []>("op_7003_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7003_cast = einsum(equation = var_7003_equation_0, values = (var_6845_cast, var_6762_cast));
            tensor<fp16, []> var_7004_to_fp16 = const()[name = tensor<string, []>("op_7004_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_459_cast = mul(x = var_7003_cast, y = var_7004_to_fp16);
            tensor<string, []> var_7007_equation_0 = const()[name = tensor<string, []>("op_7007_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7007_cast = einsum(equation = var_7007_equation_0, values = (var_6849_cast, var_6766_cast));
            tensor<fp16, []> var_7008_to_fp16 = const()[name = tensor<string, []>("op_7008_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_461_cast = mul(x = var_7007_cast, y = var_7008_to_fp16);
            tensor<string, []> var_7011_equation_0 = const()[name = tensor<string, []>("op_7011_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7011_cast = einsum(equation = var_7011_equation_0, values = (var_6853_cast, var_6770_cast));
            tensor<fp16, []> var_7012_to_fp16 = const()[name = tensor<string, []>("op_7012_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_463_cast = mul(x = var_7011_cast, y = var_7012_to_fp16);
            tensor<string, []> var_7015_equation_0 = const()[name = tensor<string, []>("op_7015_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7015_cast = einsum(equation = var_7015_equation_0, values = (var_6857_cast, var_6774_cast));
            tensor<fp16, []> var_7016_to_fp16 = const()[name = tensor<string, []>("op_7016_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_465_cast = mul(x = var_7015_cast, y = var_7016_to_fp16);
            tensor<string, []> var_7019_equation_0 = const()[name = tensor<string, []>("op_7019_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7019_cast = einsum(equation = var_7019_equation_0, values = (var_6861_cast, var_6778_cast));
            tensor<fp16, []> var_7020_to_fp16 = const()[name = tensor<string, []>("op_7020_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_467_cast = mul(x = var_7019_cast, y = var_7020_to_fp16);
            tensor<string, []> var_7023_equation_0 = const()[name = tensor<string, []>("op_7023_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7023_cast = einsum(equation = var_7023_equation_0, values = (var_6865_cast, var_6782_cast));
            tensor<fp16, []> var_7024_to_fp16 = const()[name = tensor<string, []>("op_7024_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_469_cast = mul(x = var_7023_cast, y = var_7024_to_fp16);
            tensor<string, []> var_7027_equation_0 = const()[name = tensor<string, []>("op_7027_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7027_cast = einsum(equation = var_7027_equation_0, values = (var_6869_cast, var_6786_cast));
            tensor<fp16, []> var_7028_to_fp16 = const()[name = tensor<string, []>("op_7028_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_471_cast = mul(x = var_7027_cast, y = var_7028_to_fp16);
            tensor<string, []> var_7031_equation_0 = const()[name = tensor<string, []>("op_7031_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7031_cast = einsum(equation = var_7031_equation_0, values = (var_6873_cast, var_6790_cast));
            tensor<fp16, []> var_7032_to_fp16 = const()[name = tensor<string, []>("op_7032_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_473_cast = mul(x = var_7031_cast, y = var_7032_to_fp16);
            tensor<string, []> var_7035_equation_0 = const()[name = tensor<string, []>("op_7035_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7035_cast = einsum(equation = var_7035_equation_0, values = (var_6877_cast, var_6794_cast));
            tensor<fp16, []> var_7036_to_fp16 = const()[name = tensor<string, []>("op_7036_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_475_cast = mul(x = var_7035_cast, y = var_7036_to_fp16);
            tensor<string, []> var_7039_equation_0 = const()[name = tensor<string, []>("op_7039_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7039_cast = einsum(equation = var_7039_equation_0, values = (var_6881_cast, var_6798_cast));
            tensor<fp16, []> var_7040_to_fp16 = const()[name = tensor<string, []>("op_7040_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_477_cast = mul(x = var_7039_cast, y = var_7040_to_fp16);
            tensor<string, []> var_7043_equation_0 = const()[name = tensor<string, []>("op_7043_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7043_cast = einsum(equation = var_7043_equation_0, values = (var_6885_cast, var_6802_cast));
            tensor<fp16, []> var_7044_to_fp16 = const()[name = tensor<string, []>("op_7044_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_479_cast = mul(x = var_7043_cast, y = var_7044_to_fp16);
            tensor<fp16, [2, 256, 1, 256]> var_7046_cast = softmax(axis = var_5599, x = aw_441_cast);
            tensor<fp16, [2, 256, 1, 256]> var_7047_cast = softmax(axis = var_5599, x = aw_443_cast);
            tensor<fp16, [2, 256, 1, 256]> var_7048_cast = softmax(axis = var_5599, x = aw_445_cast);
            tensor<fp16, [2, 256, 1, 256]> var_7049_cast = softmax(axis = var_5599, x = aw_447_cast);
            tensor<fp16, [2, 256, 1, 256]> var_7050_cast = softmax(axis = var_5599, x = aw_449_cast);
            tensor<fp16, [2, 256, 1, 256]> var_7051_cast = softmax(axis = var_5599, x = aw_451_cast);
            tensor<fp16, [2, 256, 1, 256]> var_7052_cast = softmax(axis = var_5599, x = aw_453_cast);
            tensor<fp16, [2, 256, 1, 256]> var_7053_cast = softmax(axis = var_5599, x = aw_455_cast);
            tensor<fp16, [2, 256, 1, 256]> var_7054_cast = softmax(axis = var_5599, x = aw_457_cast);
            tensor<fp16, [2, 256, 1, 256]> var_7055_cast = softmax(axis = var_5599, x = aw_459_cast);
            tensor<fp16, [2, 256, 1, 256]> var_7056_cast = softmax(axis = var_5599, x = aw_461_cast);
            tensor<fp16, [2, 256, 1, 256]> var_7057_cast = softmax(axis = var_5599, x = aw_463_cast);
            tensor<fp16, [2, 256, 1, 256]> var_7058_cast = softmax(axis = var_5599, x = aw_465_cast);
            tensor<fp16, [2, 256, 1, 256]> var_7059_cast = softmax(axis = var_5599, x = aw_467_cast);
            tensor<fp16, [2, 256, 1, 256]> var_7060_cast = softmax(axis = var_5599, x = aw_469_cast);
            tensor<fp16, [2, 256, 1, 256]> var_7061_cast = softmax(axis = var_5599, x = aw_471_cast);
            tensor<fp16, [2, 256, 1, 256]> var_7062_cast = softmax(axis = var_5599, x = aw_473_cast);
            tensor<fp16, [2, 256, 1, 256]> var_7063_cast = softmax(axis = var_5599, x = aw_475_cast);
            tensor<fp16, [2, 256, 1, 256]> var_7064_cast = softmax(axis = var_5599, x = aw_477_cast);
            tensor<fp16, [2, 256, 1, 256]> var_7065_cast = softmax(axis = var_5599, x = aw_479_cast);
            tensor<string, []> var_7067_equation_0 = const()[name = tensor<string, []>("op_7067_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7067_cast = einsum(equation = var_7067_equation_0, values = (var_6887_cast, var_7046_cast));
            tensor<string, []> var_7069_equation_0 = const()[name = tensor<string, []>("op_7069_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7069_cast = einsum(equation = var_7069_equation_0, values = (var_6891_cast, var_7047_cast));
            tensor<string, []> var_7071_equation_0 = const()[name = tensor<string, []>("op_7071_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7071_cast = einsum(equation = var_7071_equation_0, values = (var_6895_cast, var_7048_cast));
            tensor<string, []> var_7073_equation_0 = const()[name = tensor<string, []>("op_7073_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7073_cast = einsum(equation = var_7073_equation_0, values = (var_6899_cast, var_7049_cast));
            tensor<string, []> var_7075_equation_0 = const()[name = tensor<string, []>("op_7075_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7075_cast = einsum(equation = var_7075_equation_0, values = (var_6903_cast, var_7050_cast));
            tensor<string, []> var_7077_equation_0 = const()[name = tensor<string, []>("op_7077_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7077_cast = einsum(equation = var_7077_equation_0, values = (var_6907_cast, var_7051_cast));
            tensor<string, []> var_7079_equation_0 = const()[name = tensor<string, []>("op_7079_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7079_cast = einsum(equation = var_7079_equation_0, values = (var_6911_cast, var_7052_cast));
            tensor<string, []> var_7081_equation_0 = const()[name = tensor<string, []>("op_7081_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7081_cast = einsum(equation = var_7081_equation_0, values = (var_6915_cast, var_7053_cast));
            tensor<string, []> var_7083_equation_0 = const()[name = tensor<string, []>("op_7083_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7083_cast = einsum(equation = var_7083_equation_0, values = (var_6919_cast, var_7054_cast));
            tensor<string, []> var_7085_equation_0 = const()[name = tensor<string, []>("op_7085_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7085_cast = einsum(equation = var_7085_equation_0, values = (var_6923_cast, var_7055_cast));
            tensor<string, []> var_7087_equation_0 = const()[name = tensor<string, []>("op_7087_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7087_cast = einsum(equation = var_7087_equation_0, values = (var_6927_cast, var_7056_cast));
            tensor<string, []> var_7089_equation_0 = const()[name = tensor<string, []>("op_7089_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7089_cast = einsum(equation = var_7089_equation_0, values = (var_6931_cast, var_7057_cast));
            tensor<string, []> var_7091_equation_0 = const()[name = tensor<string, []>("op_7091_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7091_cast = einsum(equation = var_7091_equation_0, values = (var_6935_cast, var_7058_cast));
            tensor<string, []> var_7093_equation_0 = const()[name = tensor<string, []>("op_7093_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7093_cast = einsum(equation = var_7093_equation_0, values = (var_6939_cast, var_7059_cast));
            tensor<string, []> var_7095_equation_0 = const()[name = tensor<string, []>("op_7095_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7095_cast = einsum(equation = var_7095_equation_0, values = (var_6943_cast, var_7060_cast));
            tensor<string, []> var_7097_equation_0 = const()[name = tensor<string, []>("op_7097_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7097_cast = einsum(equation = var_7097_equation_0, values = (var_6947_cast, var_7061_cast));
            tensor<string, []> var_7099_equation_0 = const()[name = tensor<string, []>("op_7099_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7099_cast = einsum(equation = var_7099_equation_0, values = (var_6951_cast, var_7062_cast));
            tensor<string, []> var_7101_equation_0 = const()[name = tensor<string, []>("op_7101_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7101_cast = einsum(equation = var_7101_equation_0, values = (var_6955_cast, var_7063_cast));
            tensor<string, []> var_7103_equation_0 = const()[name = tensor<string, []>("op_7103_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7103_cast = einsum(equation = var_7103_equation_0, values = (var_6959_cast, var_7064_cast));
            tensor<string, []> var_7105_equation_0 = const()[name = tensor<string, []>("op_7105_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7105_cast = einsum(equation = var_7105_equation_0, values = (var_6963_cast, var_7065_cast));
            tensor<bool, []> input_327_interleave_0 = const()[name = tensor<string, []>("input_327_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 256]> input_327_cast = concat(axis = var_5599, interleave = input_327_interleave_0, values = (var_7067_cast, var_7069_cast, var_7071_cast, var_7073_cast, var_7075_cast, var_7077_cast, var_7079_cast, var_7081_cast, var_7083_cast, var_7085_cast, var_7087_cast, var_7089_cast, var_7091_cast, var_7093_cast, var_7095_cast, var_7097_cast, var_7099_cast, var_7101_cast, var_7103_cast, var_7105_cast));
            tensor<int32, [2]> var_7111 = const()[name = tensor<string, []>("op_7111"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7113 = const()[name = tensor<string, []>("op_7113"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_7115_pad_type_0 = const()[name = tensor<string, []>("op_7115_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_7115_pad_0 = const()[name = tensor<string, []>("op_7115_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1308632320)))];
            tensor<fp16, [1280]> up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1311909184)))];
            tensor<fp16, [2, 1280, 1, 256]> var_7115_cast = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_7113, groups = var_5599, pad = var_7115_pad_0, pad_type = var_7115_pad_type_0, strides = var_7111, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_327_cast);
            tensor<fp16, [2, 1280, 1, 256]> inputs_51_cast = add(x = var_7115_cast, y = inputs_49_cast);
            tensor<int32, [1]> var_7119 = const()[name = tensor<string, []>("op_7119"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_51_cast = reduce_mean(axes = var_7119, keep_dims = var_5594, x = inputs_51_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_51_cast = sub(x = inputs_51_cast, y = channels_mean_51_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_sq_51_cast = mul(x = zero_mean_51_cast, y = zero_mean_51_cast);
            tensor<int32, [1]> var_7123 = const()[name = tensor<string, []>("op_7123"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_7124_cast = reduce_mean(axes = var_7123, keep_dims = var_5594, x = zero_mean_sq_51_cast);
            tensor<fp16, []> var_7125_to_fp16 = const()[name = tensor<string, []>("op_7125_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_7126_cast = add(x = var_7124_cast, y = var_7125_to_fp16);
            tensor<fp16, []> denom_51_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_51_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 256]> denom_51_cast = rsqrt(epsilon = denom_51_epsilon_0_to_fp16, x = var_7126_cast);
            tensor<fp16, [2, 1280, 1, 256]> out_51_cast = mul(x = zero_mean_51_cast, y = denom_51_cast);
            tensor<fp16, [1, 1280, 1, 1]> var_7130_to_fp16 = const()[name = tensor<string, []>("op_7130_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1311911808)))];
            tensor<fp16, [2, 1280, 1, 256]> var_7131_cast = add(x = out_51_cast, y = var_7130_to_fp16);
            tensor<fp16, [1, 1280, 1, 1]> var_7133_to_fp16 = const()[name = tensor<string, []>("op_7133_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1311914432)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_189_cast = mul(x = var_7131_cast, y = var_7133_to_fp16);
            tensor<int32, [2]> var_7140 = const()[name = tensor<string, []>("op_7140"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7142 = const()[name = tensor<string, []>("op_7142"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_35_pad_type_0 = const()[name = tensor<string, []>("q_35_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_35_pad_0 = const()[name = tensor<string, []>("q_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1311917056)))];
            tensor<fp16, [2, 1280, 1, 256]> q_35_cast = conv(dilations = var_7142, groups = var_5599, pad = q_35_pad_0, pad_type = q_35_pad_type_0, strides = var_7140, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_189_cast);
            tensor<int32, [2]> var_7146 = const()[name = tensor<string, []>("op_7146"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7148 = const()[name = tensor<string, []>("op_7148"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_69_pad_type_0 = const()[name = tensor<string, []>("k_69_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_69_pad_0 = const()[name = tensor<string, []>("k_69_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1024, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1315193920)))];
            tensor<fp16, [2, 1280, 1, 77]> k_69_cast = conv(dilations = var_7148, groups = var_5599, pad = k_69_pad_0, pad_type = k_69_pad_type_0, strides = var_7146, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [2]> var_7152 = const()[name = tensor<string, []>("op_7152"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7154 = const()[name = tensor<string, []>("op_7154"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_35_pad_type_0 = const()[name = tensor<string, []>("v_35_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_35_pad_0 = const()[name = tensor<string, []>("v_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1024, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1317815424)))];
            tensor<fp16, [2, 1280, 1, 77]> v_35_cast = conv(dilations = var_7154, groups = var_5599, pad = v_35_pad_0, pad_type = v_35_pad_type_0, strides = var_7152, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [4]> var_7158_begin_0 = const()[name = tensor<string, []>("op_7158_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7158_end_0 = const()[name = tensor<string, []>("op_7158_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_7158_end_mask_0 = const()[name = tensor<string, []>("op_7158_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7158_cast = slice_by_index(begin = var_7158_begin_0, end = var_7158_end_0, end_mask = var_7158_end_mask_0, x = q_35_cast);
            tensor<int32, [4]> var_7162_begin_0 = const()[name = tensor<string, []>("op_7162_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_7162_end_0 = const()[name = tensor<string, []>("op_7162_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_7162_end_mask_0 = const()[name = tensor<string, []>("op_7162_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7162_cast = slice_by_index(begin = var_7162_begin_0, end = var_7162_end_0, end_mask = var_7162_end_mask_0, x = q_35_cast);
            tensor<int32, [4]> var_7166_begin_0 = const()[name = tensor<string, []>("op_7166_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_7166_end_0 = const()[name = tensor<string, []>("op_7166_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_7166_end_mask_0 = const()[name = tensor<string, []>("op_7166_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7166_cast = slice_by_index(begin = var_7166_begin_0, end = var_7166_end_0, end_mask = var_7166_end_mask_0, x = q_35_cast);
            tensor<int32, [4]> var_7170_begin_0 = const()[name = tensor<string, []>("op_7170_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_7170_end_0 = const()[name = tensor<string, []>("op_7170_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_7170_end_mask_0 = const()[name = tensor<string, []>("op_7170_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7170_cast = slice_by_index(begin = var_7170_begin_0, end = var_7170_end_0, end_mask = var_7170_end_mask_0, x = q_35_cast);
            tensor<int32, [4]> var_7174_begin_0 = const()[name = tensor<string, []>("op_7174_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_7174_end_0 = const()[name = tensor<string, []>("op_7174_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_7174_end_mask_0 = const()[name = tensor<string, []>("op_7174_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7174_cast = slice_by_index(begin = var_7174_begin_0, end = var_7174_end_0, end_mask = var_7174_end_mask_0, x = q_35_cast);
            tensor<int32, [4]> var_7178_begin_0 = const()[name = tensor<string, []>("op_7178_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_7178_end_0 = const()[name = tensor<string, []>("op_7178_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_7178_end_mask_0 = const()[name = tensor<string, []>("op_7178_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7178_cast = slice_by_index(begin = var_7178_begin_0, end = var_7178_end_0, end_mask = var_7178_end_mask_0, x = q_35_cast);
            tensor<int32, [4]> var_7182_begin_0 = const()[name = tensor<string, []>("op_7182_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_7182_end_0 = const()[name = tensor<string, []>("op_7182_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_7182_end_mask_0 = const()[name = tensor<string, []>("op_7182_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7182_cast = slice_by_index(begin = var_7182_begin_0, end = var_7182_end_0, end_mask = var_7182_end_mask_0, x = q_35_cast);
            tensor<int32, [4]> var_7186_begin_0 = const()[name = tensor<string, []>("op_7186_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_7186_end_0 = const()[name = tensor<string, []>("op_7186_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_7186_end_mask_0 = const()[name = tensor<string, []>("op_7186_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7186_cast = slice_by_index(begin = var_7186_begin_0, end = var_7186_end_0, end_mask = var_7186_end_mask_0, x = q_35_cast);
            tensor<int32, [4]> var_7190_begin_0 = const()[name = tensor<string, []>("op_7190_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_7190_end_0 = const()[name = tensor<string, []>("op_7190_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_7190_end_mask_0 = const()[name = tensor<string, []>("op_7190_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7190_cast = slice_by_index(begin = var_7190_begin_0, end = var_7190_end_0, end_mask = var_7190_end_mask_0, x = q_35_cast);
            tensor<int32, [4]> var_7194_begin_0 = const()[name = tensor<string, []>("op_7194_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_7194_end_0 = const()[name = tensor<string, []>("op_7194_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_7194_end_mask_0 = const()[name = tensor<string, []>("op_7194_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7194_cast = slice_by_index(begin = var_7194_begin_0, end = var_7194_end_0, end_mask = var_7194_end_mask_0, x = q_35_cast);
            tensor<int32, [4]> var_7198_begin_0 = const()[name = tensor<string, []>("op_7198_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_7198_end_0 = const()[name = tensor<string, []>("op_7198_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_7198_end_mask_0 = const()[name = tensor<string, []>("op_7198_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7198_cast = slice_by_index(begin = var_7198_begin_0, end = var_7198_end_0, end_mask = var_7198_end_mask_0, x = q_35_cast);
            tensor<int32, [4]> var_7202_begin_0 = const()[name = tensor<string, []>("op_7202_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_7202_end_0 = const()[name = tensor<string, []>("op_7202_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_7202_end_mask_0 = const()[name = tensor<string, []>("op_7202_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7202_cast = slice_by_index(begin = var_7202_begin_0, end = var_7202_end_0, end_mask = var_7202_end_mask_0, x = q_35_cast);
            tensor<int32, [4]> var_7206_begin_0 = const()[name = tensor<string, []>("op_7206_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_7206_end_0 = const()[name = tensor<string, []>("op_7206_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_7206_end_mask_0 = const()[name = tensor<string, []>("op_7206_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7206_cast = slice_by_index(begin = var_7206_begin_0, end = var_7206_end_0, end_mask = var_7206_end_mask_0, x = q_35_cast);
            tensor<int32, [4]> var_7210_begin_0 = const()[name = tensor<string, []>("op_7210_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_7210_end_0 = const()[name = tensor<string, []>("op_7210_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_7210_end_mask_0 = const()[name = tensor<string, []>("op_7210_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7210_cast = slice_by_index(begin = var_7210_begin_0, end = var_7210_end_0, end_mask = var_7210_end_mask_0, x = q_35_cast);
            tensor<int32, [4]> var_7214_begin_0 = const()[name = tensor<string, []>("op_7214_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_7214_end_0 = const()[name = tensor<string, []>("op_7214_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_7214_end_mask_0 = const()[name = tensor<string, []>("op_7214_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7214_cast = slice_by_index(begin = var_7214_begin_0, end = var_7214_end_0, end_mask = var_7214_end_mask_0, x = q_35_cast);
            tensor<int32, [4]> var_7218_begin_0 = const()[name = tensor<string, []>("op_7218_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_7218_end_0 = const()[name = tensor<string, []>("op_7218_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_7218_end_mask_0 = const()[name = tensor<string, []>("op_7218_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7218_cast = slice_by_index(begin = var_7218_begin_0, end = var_7218_end_0, end_mask = var_7218_end_mask_0, x = q_35_cast);
            tensor<int32, [4]> var_7222_begin_0 = const()[name = tensor<string, []>("op_7222_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_7222_end_0 = const()[name = tensor<string, []>("op_7222_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_7222_end_mask_0 = const()[name = tensor<string, []>("op_7222_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7222_cast = slice_by_index(begin = var_7222_begin_0, end = var_7222_end_0, end_mask = var_7222_end_mask_0, x = q_35_cast);
            tensor<int32, [4]> var_7226_begin_0 = const()[name = tensor<string, []>("op_7226_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_7226_end_0 = const()[name = tensor<string, []>("op_7226_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_7226_end_mask_0 = const()[name = tensor<string, []>("op_7226_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7226_cast = slice_by_index(begin = var_7226_begin_0, end = var_7226_end_0, end_mask = var_7226_end_mask_0, x = q_35_cast);
            tensor<int32, [4]> var_7230_begin_0 = const()[name = tensor<string, []>("op_7230_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_7230_end_0 = const()[name = tensor<string, []>("op_7230_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_7230_end_mask_0 = const()[name = tensor<string, []>("op_7230_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7230_cast = slice_by_index(begin = var_7230_begin_0, end = var_7230_end_0, end_mask = var_7230_end_mask_0, x = q_35_cast);
            tensor<int32, [4]> var_7234_begin_0 = const()[name = tensor<string, []>("op_7234_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_7234_end_0 = const()[name = tensor<string, []>("op_7234_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_7234_end_mask_0 = const()[name = tensor<string, []>("op_7234_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7234_cast = slice_by_index(begin = var_7234_begin_0, end = var_7234_end_0, end_mask = var_7234_end_mask_0, x = q_35_cast);
            tensor<int32, [4]> k_71_perm_0 = const()[name = tensor<string, []>("k_71_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_7241_begin_0 = const()[name = tensor<string, []>("op_7241_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7241_end_0 = const()[name = tensor<string, []>("op_7241_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_7241_end_mask_0 = const()[name = tensor<string, []>("op_7241_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_14 = transpose(perm = k_71_perm_0, x = k_69_cast);
            tensor<fp16, [2, 77, 1, 64]> var_7241_cast = slice_by_index(begin = var_7241_begin_0, end = var_7241_end_0, end_mask = var_7241_end_mask_0, x = transpose_14);
            tensor<int32, [4]> var_7245_begin_0 = const()[name = tensor<string, []>("op_7245_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_7245_end_0 = const()[name = tensor<string, []>("op_7245_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_7245_end_mask_0 = const()[name = tensor<string, []>("op_7245_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7245_cast = slice_by_index(begin = var_7245_begin_0, end = var_7245_end_0, end_mask = var_7245_end_mask_0, x = transpose_14);
            tensor<int32, [4]> var_7249_begin_0 = const()[name = tensor<string, []>("op_7249_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_7249_end_0 = const()[name = tensor<string, []>("op_7249_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_7249_end_mask_0 = const()[name = tensor<string, []>("op_7249_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7249_cast = slice_by_index(begin = var_7249_begin_0, end = var_7249_end_0, end_mask = var_7249_end_mask_0, x = transpose_14);
            tensor<int32, [4]> var_7253_begin_0 = const()[name = tensor<string, []>("op_7253_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_7253_end_0 = const()[name = tensor<string, []>("op_7253_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_7253_end_mask_0 = const()[name = tensor<string, []>("op_7253_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7253_cast = slice_by_index(begin = var_7253_begin_0, end = var_7253_end_0, end_mask = var_7253_end_mask_0, x = transpose_14);
            tensor<int32, [4]> var_7257_begin_0 = const()[name = tensor<string, []>("op_7257_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_7257_end_0 = const()[name = tensor<string, []>("op_7257_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_7257_end_mask_0 = const()[name = tensor<string, []>("op_7257_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7257_cast = slice_by_index(begin = var_7257_begin_0, end = var_7257_end_0, end_mask = var_7257_end_mask_0, x = transpose_14);
            tensor<int32, [4]> var_7261_begin_0 = const()[name = tensor<string, []>("op_7261_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_7261_end_0 = const()[name = tensor<string, []>("op_7261_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_7261_end_mask_0 = const()[name = tensor<string, []>("op_7261_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7261_cast = slice_by_index(begin = var_7261_begin_0, end = var_7261_end_0, end_mask = var_7261_end_mask_0, x = transpose_14);
            tensor<int32, [4]> var_7265_begin_0 = const()[name = tensor<string, []>("op_7265_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_7265_end_0 = const()[name = tensor<string, []>("op_7265_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_7265_end_mask_0 = const()[name = tensor<string, []>("op_7265_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7265_cast = slice_by_index(begin = var_7265_begin_0, end = var_7265_end_0, end_mask = var_7265_end_mask_0, x = transpose_14);
            tensor<int32, [4]> var_7269_begin_0 = const()[name = tensor<string, []>("op_7269_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_7269_end_0 = const()[name = tensor<string, []>("op_7269_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_7269_end_mask_0 = const()[name = tensor<string, []>("op_7269_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7269_cast = slice_by_index(begin = var_7269_begin_0, end = var_7269_end_0, end_mask = var_7269_end_mask_0, x = transpose_14);
            tensor<int32, [4]> var_7273_begin_0 = const()[name = tensor<string, []>("op_7273_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_7273_end_0 = const()[name = tensor<string, []>("op_7273_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_7273_end_mask_0 = const()[name = tensor<string, []>("op_7273_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7273_cast = slice_by_index(begin = var_7273_begin_0, end = var_7273_end_0, end_mask = var_7273_end_mask_0, x = transpose_14);
            tensor<int32, [4]> var_7277_begin_0 = const()[name = tensor<string, []>("op_7277_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_7277_end_0 = const()[name = tensor<string, []>("op_7277_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_7277_end_mask_0 = const()[name = tensor<string, []>("op_7277_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7277_cast = slice_by_index(begin = var_7277_begin_0, end = var_7277_end_0, end_mask = var_7277_end_mask_0, x = transpose_14);
            tensor<int32, [4]> var_7281_begin_0 = const()[name = tensor<string, []>("op_7281_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_7281_end_0 = const()[name = tensor<string, []>("op_7281_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_7281_end_mask_0 = const()[name = tensor<string, []>("op_7281_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7281_cast = slice_by_index(begin = var_7281_begin_0, end = var_7281_end_0, end_mask = var_7281_end_mask_0, x = transpose_14);
            tensor<int32, [4]> var_7285_begin_0 = const()[name = tensor<string, []>("op_7285_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_7285_end_0 = const()[name = tensor<string, []>("op_7285_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_7285_end_mask_0 = const()[name = tensor<string, []>("op_7285_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7285_cast = slice_by_index(begin = var_7285_begin_0, end = var_7285_end_0, end_mask = var_7285_end_mask_0, x = transpose_14);
            tensor<int32, [4]> var_7289_begin_0 = const()[name = tensor<string, []>("op_7289_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_7289_end_0 = const()[name = tensor<string, []>("op_7289_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_7289_end_mask_0 = const()[name = tensor<string, []>("op_7289_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7289_cast = slice_by_index(begin = var_7289_begin_0, end = var_7289_end_0, end_mask = var_7289_end_mask_0, x = transpose_14);
            tensor<int32, [4]> var_7293_begin_0 = const()[name = tensor<string, []>("op_7293_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_7293_end_0 = const()[name = tensor<string, []>("op_7293_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_7293_end_mask_0 = const()[name = tensor<string, []>("op_7293_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7293_cast = slice_by_index(begin = var_7293_begin_0, end = var_7293_end_0, end_mask = var_7293_end_mask_0, x = transpose_14);
            tensor<int32, [4]> var_7297_begin_0 = const()[name = tensor<string, []>("op_7297_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_7297_end_0 = const()[name = tensor<string, []>("op_7297_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_7297_end_mask_0 = const()[name = tensor<string, []>("op_7297_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7297_cast = slice_by_index(begin = var_7297_begin_0, end = var_7297_end_0, end_mask = var_7297_end_mask_0, x = transpose_14);
            tensor<int32, [4]> var_7301_begin_0 = const()[name = tensor<string, []>("op_7301_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_7301_end_0 = const()[name = tensor<string, []>("op_7301_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_7301_end_mask_0 = const()[name = tensor<string, []>("op_7301_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7301_cast = slice_by_index(begin = var_7301_begin_0, end = var_7301_end_0, end_mask = var_7301_end_mask_0, x = transpose_14);
            tensor<int32, [4]> var_7305_begin_0 = const()[name = tensor<string, []>("op_7305_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_7305_end_0 = const()[name = tensor<string, []>("op_7305_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_7305_end_mask_0 = const()[name = tensor<string, []>("op_7305_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7305_cast = slice_by_index(begin = var_7305_begin_0, end = var_7305_end_0, end_mask = var_7305_end_mask_0, x = transpose_14);
            tensor<int32, [4]> var_7309_begin_0 = const()[name = tensor<string, []>("op_7309_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_7309_end_0 = const()[name = tensor<string, []>("op_7309_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_7309_end_mask_0 = const()[name = tensor<string, []>("op_7309_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7309_cast = slice_by_index(begin = var_7309_begin_0, end = var_7309_end_0, end_mask = var_7309_end_mask_0, x = transpose_14);
            tensor<int32, [4]> var_7313_begin_0 = const()[name = tensor<string, []>("op_7313_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_7313_end_0 = const()[name = tensor<string, []>("op_7313_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_7313_end_mask_0 = const()[name = tensor<string, []>("op_7313_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7313_cast = slice_by_index(begin = var_7313_begin_0, end = var_7313_end_0, end_mask = var_7313_end_mask_0, x = transpose_14);
            tensor<int32, [4]> var_7317_begin_0 = const()[name = tensor<string, []>("op_7317_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_7317_end_0 = const()[name = tensor<string, []>("op_7317_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_7317_end_mask_0 = const()[name = tensor<string, []>("op_7317_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7317_cast = slice_by_index(begin = var_7317_begin_0, end = var_7317_end_0, end_mask = var_7317_end_mask_0, x = transpose_14);
            tensor<int32, [4]> var_7319_begin_0 = const()[name = tensor<string, []>("op_7319_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7319_end_0 = const()[name = tensor<string, []>("op_7319_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_7319_end_mask_0 = const()[name = tensor<string, []>("op_7319_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7319_cast = slice_by_index(begin = var_7319_begin_0, end = var_7319_end_0, end_mask = var_7319_end_mask_0, x = v_35_cast);
            tensor<int32, [4]> var_7323_begin_0 = const()[name = tensor<string, []>("op_7323_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_7323_end_0 = const()[name = tensor<string, []>("op_7323_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_7323_end_mask_0 = const()[name = tensor<string, []>("op_7323_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7323_cast = slice_by_index(begin = var_7323_begin_0, end = var_7323_end_0, end_mask = var_7323_end_mask_0, x = v_35_cast);
            tensor<int32, [4]> var_7327_begin_0 = const()[name = tensor<string, []>("op_7327_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_7327_end_0 = const()[name = tensor<string, []>("op_7327_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_7327_end_mask_0 = const()[name = tensor<string, []>("op_7327_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7327_cast = slice_by_index(begin = var_7327_begin_0, end = var_7327_end_0, end_mask = var_7327_end_mask_0, x = v_35_cast);
            tensor<int32, [4]> var_7331_begin_0 = const()[name = tensor<string, []>("op_7331_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_7331_end_0 = const()[name = tensor<string, []>("op_7331_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_7331_end_mask_0 = const()[name = tensor<string, []>("op_7331_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7331_cast = slice_by_index(begin = var_7331_begin_0, end = var_7331_end_0, end_mask = var_7331_end_mask_0, x = v_35_cast);
            tensor<int32, [4]> var_7335_begin_0 = const()[name = tensor<string, []>("op_7335_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_7335_end_0 = const()[name = tensor<string, []>("op_7335_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_7335_end_mask_0 = const()[name = tensor<string, []>("op_7335_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7335_cast = slice_by_index(begin = var_7335_begin_0, end = var_7335_end_0, end_mask = var_7335_end_mask_0, x = v_35_cast);
            tensor<int32, [4]> var_7339_begin_0 = const()[name = tensor<string, []>("op_7339_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_7339_end_0 = const()[name = tensor<string, []>("op_7339_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_7339_end_mask_0 = const()[name = tensor<string, []>("op_7339_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7339_cast = slice_by_index(begin = var_7339_begin_0, end = var_7339_end_0, end_mask = var_7339_end_mask_0, x = v_35_cast);
            tensor<int32, [4]> var_7343_begin_0 = const()[name = tensor<string, []>("op_7343_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_7343_end_0 = const()[name = tensor<string, []>("op_7343_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_7343_end_mask_0 = const()[name = tensor<string, []>("op_7343_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7343_cast = slice_by_index(begin = var_7343_begin_0, end = var_7343_end_0, end_mask = var_7343_end_mask_0, x = v_35_cast);
            tensor<int32, [4]> var_7347_begin_0 = const()[name = tensor<string, []>("op_7347_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_7347_end_0 = const()[name = tensor<string, []>("op_7347_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_7347_end_mask_0 = const()[name = tensor<string, []>("op_7347_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7347_cast = slice_by_index(begin = var_7347_begin_0, end = var_7347_end_0, end_mask = var_7347_end_mask_0, x = v_35_cast);
            tensor<int32, [4]> var_7351_begin_0 = const()[name = tensor<string, []>("op_7351_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_7351_end_0 = const()[name = tensor<string, []>("op_7351_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_7351_end_mask_0 = const()[name = tensor<string, []>("op_7351_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7351_cast = slice_by_index(begin = var_7351_begin_0, end = var_7351_end_0, end_mask = var_7351_end_mask_0, x = v_35_cast);
            tensor<int32, [4]> var_7355_begin_0 = const()[name = tensor<string, []>("op_7355_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_7355_end_0 = const()[name = tensor<string, []>("op_7355_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_7355_end_mask_0 = const()[name = tensor<string, []>("op_7355_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7355_cast = slice_by_index(begin = var_7355_begin_0, end = var_7355_end_0, end_mask = var_7355_end_mask_0, x = v_35_cast);
            tensor<int32, [4]> var_7359_begin_0 = const()[name = tensor<string, []>("op_7359_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_7359_end_0 = const()[name = tensor<string, []>("op_7359_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_7359_end_mask_0 = const()[name = tensor<string, []>("op_7359_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7359_cast = slice_by_index(begin = var_7359_begin_0, end = var_7359_end_0, end_mask = var_7359_end_mask_0, x = v_35_cast);
            tensor<int32, [4]> var_7363_begin_0 = const()[name = tensor<string, []>("op_7363_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_7363_end_0 = const()[name = tensor<string, []>("op_7363_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_7363_end_mask_0 = const()[name = tensor<string, []>("op_7363_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7363_cast = slice_by_index(begin = var_7363_begin_0, end = var_7363_end_0, end_mask = var_7363_end_mask_0, x = v_35_cast);
            tensor<int32, [4]> var_7367_begin_0 = const()[name = tensor<string, []>("op_7367_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_7367_end_0 = const()[name = tensor<string, []>("op_7367_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_7367_end_mask_0 = const()[name = tensor<string, []>("op_7367_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7367_cast = slice_by_index(begin = var_7367_begin_0, end = var_7367_end_0, end_mask = var_7367_end_mask_0, x = v_35_cast);
            tensor<int32, [4]> var_7371_begin_0 = const()[name = tensor<string, []>("op_7371_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_7371_end_0 = const()[name = tensor<string, []>("op_7371_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_7371_end_mask_0 = const()[name = tensor<string, []>("op_7371_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7371_cast = slice_by_index(begin = var_7371_begin_0, end = var_7371_end_0, end_mask = var_7371_end_mask_0, x = v_35_cast);
            tensor<int32, [4]> var_7375_begin_0 = const()[name = tensor<string, []>("op_7375_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_7375_end_0 = const()[name = tensor<string, []>("op_7375_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_7375_end_mask_0 = const()[name = tensor<string, []>("op_7375_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7375_cast = slice_by_index(begin = var_7375_begin_0, end = var_7375_end_0, end_mask = var_7375_end_mask_0, x = v_35_cast);
            tensor<int32, [4]> var_7379_begin_0 = const()[name = tensor<string, []>("op_7379_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_7379_end_0 = const()[name = tensor<string, []>("op_7379_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_7379_end_mask_0 = const()[name = tensor<string, []>("op_7379_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7379_cast = slice_by_index(begin = var_7379_begin_0, end = var_7379_end_0, end_mask = var_7379_end_mask_0, x = v_35_cast);
            tensor<int32, [4]> var_7383_begin_0 = const()[name = tensor<string, []>("op_7383_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_7383_end_0 = const()[name = tensor<string, []>("op_7383_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_7383_end_mask_0 = const()[name = tensor<string, []>("op_7383_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7383_cast = slice_by_index(begin = var_7383_begin_0, end = var_7383_end_0, end_mask = var_7383_end_mask_0, x = v_35_cast);
            tensor<int32, [4]> var_7387_begin_0 = const()[name = tensor<string, []>("op_7387_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_7387_end_0 = const()[name = tensor<string, []>("op_7387_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_7387_end_mask_0 = const()[name = tensor<string, []>("op_7387_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7387_cast = slice_by_index(begin = var_7387_begin_0, end = var_7387_end_0, end_mask = var_7387_end_mask_0, x = v_35_cast);
            tensor<int32, [4]> var_7391_begin_0 = const()[name = tensor<string, []>("op_7391_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_7391_end_0 = const()[name = tensor<string, []>("op_7391_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_7391_end_mask_0 = const()[name = tensor<string, []>("op_7391_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7391_cast = slice_by_index(begin = var_7391_begin_0, end = var_7391_end_0, end_mask = var_7391_end_mask_0, x = v_35_cast);
            tensor<int32, [4]> var_7395_begin_0 = const()[name = tensor<string, []>("op_7395_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_7395_end_0 = const()[name = tensor<string, []>("op_7395_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_7395_end_mask_0 = const()[name = tensor<string, []>("op_7395_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7395_cast = slice_by_index(begin = var_7395_begin_0, end = var_7395_end_0, end_mask = var_7395_end_mask_0, x = v_35_cast);
            tensor<string, []> var_7399_equation_0 = const()[name = tensor<string, []>("op_7399_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_7399_cast = einsum(equation = var_7399_equation_0, values = (var_7241_cast, var_7158_cast));
            tensor<fp16, []> var_7400_to_fp16 = const()[name = tensor<string, []>("op_7400_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_481_cast = mul(x = var_7399_cast, y = var_7400_to_fp16);
            tensor<string, []> var_7403_equation_0 = const()[name = tensor<string, []>("op_7403_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_7403_cast = einsum(equation = var_7403_equation_0, values = (var_7245_cast, var_7162_cast));
            tensor<fp16, []> var_7404_to_fp16 = const()[name = tensor<string, []>("op_7404_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_483_cast = mul(x = var_7403_cast, y = var_7404_to_fp16);
            tensor<string, []> var_7407_equation_0 = const()[name = tensor<string, []>("op_7407_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_7407_cast = einsum(equation = var_7407_equation_0, values = (var_7249_cast, var_7166_cast));
            tensor<fp16, []> var_7408_to_fp16 = const()[name = tensor<string, []>("op_7408_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_485_cast = mul(x = var_7407_cast, y = var_7408_to_fp16);
            tensor<string, []> var_7411_equation_0 = const()[name = tensor<string, []>("op_7411_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_7411_cast = einsum(equation = var_7411_equation_0, values = (var_7253_cast, var_7170_cast));
            tensor<fp16, []> var_7412_to_fp16 = const()[name = tensor<string, []>("op_7412_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_487_cast = mul(x = var_7411_cast, y = var_7412_to_fp16);
            tensor<string, []> var_7415_equation_0 = const()[name = tensor<string, []>("op_7415_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_7415_cast = einsum(equation = var_7415_equation_0, values = (var_7257_cast, var_7174_cast));
            tensor<fp16, []> var_7416_to_fp16 = const()[name = tensor<string, []>("op_7416_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_489_cast = mul(x = var_7415_cast, y = var_7416_to_fp16);
            tensor<string, []> var_7419_equation_0 = const()[name = tensor<string, []>("op_7419_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_7419_cast = einsum(equation = var_7419_equation_0, values = (var_7261_cast, var_7178_cast));
            tensor<fp16, []> var_7420_to_fp16 = const()[name = tensor<string, []>("op_7420_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_491_cast = mul(x = var_7419_cast, y = var_7420_to_fp16);
            tensor<string, []> var_7423_equation_0 = const()[name = tensor<string, []>("op_7423_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_7423_cast = einsum(equation = var_7423_equation_0, values = (var_7265_cast, var_7182_cast));
            tensor<fp16, []> var_7424_to_fp16 = const()[name = tensor<string, []>("op_7424_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_493_cast = mul(x = var_7423_cast, y = var_7424_to_fp16);
            tensor<string, []> var_7427_equation_0 = const()[name = tensor<string, []>("op_7427_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_7427_cast = einsum(equation = var_7427_equation_0, values = (var_7269_cast, var_7186_cast));
            tensor<fp16, []> var_7428_to_fp16 = const()[name = tensor<string, []>("op_7428_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_495_cast = mul(x = var_7427_cast, y = var_7428_to_fp16);
            tensor<string, []> var_7431_equation_0 = const()[name = tensor<string, []>("op_7431_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_7431_cast = einsum(equation = var_7431_equation_0, values = (var_7273_cast, var_7190_cast));
            tensor<fp16, []> var_7432_to_fp16 = const()[name = tensor<string, []>("op_7432_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_497_cast = mul(x = var_7431_cast, y = var_7432_to_fp16);
            tensor<string, []> var_7435_equation_0 = const()[name = tensor<string, []>("op_7435_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_7435_cast = einsum(equation = var_7435_equation_0, values = (var_7277_cast, var_7194_cast));
            tensor<fp16, []> var_7436_to_fp16 = const()[name = tensor<string, []>("op_7436_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_499_cast = mul(x = var_7435_cast, y = var_7436_to_fp16);
            tensor<string, []> var_7439_equation_0 = const()[name = tensor<string, []>("op_7439_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_7439_cast = einsum(equation = var_7439_equation_0, values = (var_7281_cast, var_7198_cast));
            tensor<fp16, []> var_7440_to_fp16 = const()[name = tensor<string, []>("op_7440_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_501_cast = mul(x = var_7439_cast, y = var_7440_to_fp16);
            tensor<string, []> var_7443_equation_0 = const()[name = tensor<string, []>("op_7443_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_7443_cast = einsum(equation = var_7443_equation_0, values = (var_7285_cast, var_7202_cast));
            tensor<fp16, []> var_7444_to_fp16 = const()[name = tensor<string, []>("op_7444_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_503_cast = mul(x = var_7443_cast, y = var_7444_to_fp16);
            tensor<string, []> var_7447_equation_0 = const()[name = tensor<string, []>("op_7447_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_7447_cast = einsum(equation = var_7447_equation_0, values = (var_7289_cast, var_7206_cast));
            tensor<fp16, []> var_7448_to_fp16 = const()[name = tensor<string, []>("op_7448_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_505_cast = mul(x = var_7447_cast, y = var_7448_to_fp16);
            tensor<string, []> var_7451_equation_0 = const()[name = tensor<string, []>("op_7451_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_7451_cast = einsum(equation = var_7451_equation_0, values = (var_7293_cast, var_7210_cast));
            tensor<fp16, []> var_7452_to_fp16 = const()[name = tensor<string, []>("op_7452_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_507_cast = mul(x = var_7451_cast, y = var_7452_to_fp16);
            tensor<string, []> var_7455_equation_0 = const()[name = tensor<string, []>("op_7455_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_7455_cast = einsum(equation = var_7455_equation_0, values = (var_7297_cast, var_7214_cast));
            tensor<fp16, []> var_7456_to_fp16 = const()[name = tensor<string, []>("op_7456_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_509_cast = mul(x = var_7455_cast, y = var_7456_to_fp16);
            tensor<string, []> var_7459_equation_0 = const()[name = tensor<string, []>("op_7459_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_7459_cast = einsum(equation = var_7459_equation_0, values = (var_7301_cast, var_7218_cast));
            tensor<fp16, []> var_7460_to_fp16 = const()[name = tensor<string, []>("op_7460_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_511_cast = mul(x = var_7459_cast, y = var_7460_to_fp16);
            tensor<string, []> var_7463_equation_0 = const()[name = tensor<string, []>("op_7463_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_7463_cast = einsum(equation = var_7463_equation_0, values = (var_7305_cast, var_7222_cast));
            tensor<fp16, []> var_7464_to_fp16 = const()[name = tensor<string, []>("op_7464_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_513_cast = mul(x = var_7463_cast, y = var_7464_to_fp16);
            tensor<string, []> var_7467_equation_0 = const()[name = tensor<string, []>("op_7467_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_7467_cast = einsum(equation = var_7467_equation_0, values = (var_7309_cast, var_7226_cast));
            tensor<fp16, []> var_7468_to_fp16 = const()[name = tensor<string, []>("op_7468_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_515_cast = mul(x = var_7467_cast, y = var_7468_to_fp16);
            tensor<string, []> var_7471_equation_0 = const()[name = tensor<string, []>("op_7471_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_7471_cast = einsum(equation = var_7471_equation_0, values = (var_7313_cast, var_7230_cast));
            tensor<fp16, []> var_7472_to_fp16 = const()[name = tensor<string, []>("op_7472_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_517_cast = mul(x = var_7471_cast, y = var_7472_to_fp16);
            tensor<string, []> var_7475_equation_0 = const()[name = tensor<string, []>("op_7475_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_7475_cast = einsum(equation = var_7475_equation_0, values = (var_7317_cast, var_7234_cast));
            tensor<fp16, []> var_7476_to_fp16 = const()[name = tensor<string, []>("op_7476_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_519_cast = mul(x = var_7475_cast, y = var_7476_to_fp16);
            tensor<fp16, [2, 77, 1, 256]> var_7478_cast = softmax(axis = var_5599, x = aw_481_cast);
            tensor<fp16, [2, 77, 1, 256]> var_7479_cast = softmax(axis = var_5599, x = aw_483_cast);
            tensor<fp16, [2, 77, 1, 256]> var_7480_cast = softmax(axis = var_5599, x = aw_485_cast);
            tensor<fp16, [2, 77, 1, 256]> var_7481_cast = softmax(axis = var_5599, x = aw_487_cast);
            tensor<fp16, [2, 77, 1, 256]> var_7482_cast = softmax(axis = var_5599, x = aw_489_cast);
            tensor<fp16, [2, 77, 1, 256]> var_7483_cast = softmax(axis = var_5599, x = aw_491_cast);
            tensor<fp16, [2, 77, 1, 256]> var_7484_cast = softmax(axis = var_5599, x = aw_493_cast);
            tensor<fp16, [2, 77, 1, 256]> var_7485_cast = softmax(axis = var_5599, x = aw_495_cast);
            tensor<fp16, [2, 77, 1, 256]> var_7486_cast = softmax(axis = var_5599, x = aw_497_cast);
            tensor<fp16, [2, 77, 1, 256]> var_7487_cast = softmax(axis = var_5599, x = aw_499_cast);
            tensor<fp16, [2, 77, 1, 256]> var_7488_cast = softmax(axis = var_5599, x = aw_501_cast);
            tensor<fp16, [2, 77, 1, 256]> var_7489_cast = softmax(axis = var_5599, x = aw_503_cast);
            tensor<fp16, [2, 77, 1, 256]> var_7490_cast = softmax(axis = var_5599, x = aw_505_cast);
            tensor<fp16, [2, 77, 1, 256]> var_7491_cast = softmax(axis = var_5599, x = aw_507_cast);
            tensor<fp16, [2, 77, 1, 256]> var_7492_cast = softmax(axis = var_5599, x = aw_509_cast);
            tensor<fp16, [2, 77, 1, 256]> var_7493_cast = softmax(axis = var_5599, x = aw_511_cast);
            tensor<fp16, [2, 77, 1, 256]> var_7494_cast = softmax(axis = var_5599, x = aw_513_cast);
            tensor<fp16, [2, 77, 1, 256]> var_7495_cast = softmax(axis = var_5599, x = aw_515_cast);
            tensor<fp16, [2, 77, 1, 256]> var_7496_cast = softmax(axis = var_5599, x = aw_517_cast);
            tensor<fp16, [2, 77, 1, 256]> var_7497_cast = softmax(axis = var_5599, x = aw_519_cast);
            tensor<string, []> var_7499_equation_0 = const()[name = tensor<string, []>("op_7499_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7499_cast = einsum(equation = var_7499_equation_0, values = (var_7319_cast, var_7478_cast));
            tensor<string, []> var_7501_equation_0 = const()[name = tensor<string, []>("op_7501_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7501_cast = einsum(equation = var_7501_equation_0, values = (var_7323_cast, var_7479_cast));
            tensor<string, []> var_7503_equation_0 = const()[name = tensor<string, []>("op_7503_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7503_cast = einsum(equation = var_7503_equation_0, values = (var_7327_cast, var_7480_cast));
            tensor<string, []> var_7505_equation_0 = const()[name = tensor<string, []>("op_7505_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7505_cast = einsum(equation = var_7505_equation_0, values = (var_7331_cast, var_7481_cast));
            tensor<string, []> var_7507_equation_0 = const()[name = tensor<string, []>("op_7507_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7507_cast = einsum(equation = var_7507_equation_0, values = (var_7335_cast, var_7482_cast));
            tensor<string, []> var_7509_equation_0 = const()[name = tensor<string, []>("op_7509_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7509_cast = einsum(equation = var_7509_equation_0, values = (var_7339_cast, var_7483_cast));
            tensor<string, []> var_7511_equation_0 = const()[name = tensor<string, []>("op_7511_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7511_cast = einsum(equation = var_7511_equation_0, values = (var_7343_cast, var_7484_cast));
            tensor<string, []> var_7513_equation_0 = const()[name = tensor<string, []>("op_7513_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7513_cast = einsum(equation = var_7513_equation_0, values = (var_7347_cast, var_7485_cast));
            tensor<string, []> var_7515_equation_0 = const()[name = tensor<string, []>("op_7515_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7515_cast = einsum(equation = var_7515_equation_0, values = (var_7351_cast, var_7486_cast));
            tensor<string, []> var_7517_equation_0 = const()[name = tensor<string, []>("op_7517_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7517_cast = einsum(equation = var_7517_equation_0, values = (var_7355_cast, var_7487_cast));
            tensor<string, []> var_7519_equation_0 = const()[name = tensor<string, []>("op_7519_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7519_cast = einsum(equation = var_7519_equation_0, values = (var_7359_cast, var_7488_cast));
            tensor<string, []> var_7521_equation_0 = const()[name = tensor<string, []>("op_7521_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7521_cast = einsum(equation = var_7521_equation_0, values = (var_7363_cast, var_7489_cast));
            tensor<string, []> var_7523_equation_0 = const()[name = tensor<string, []>("op_7523_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7523_cast = einsum(equation = var_7523_equation_0, values = (var_7367_cast, var_7490_cast));
            tensor<string, []> var_7525_equation_0 = const()[name = tensor<string, []>("op_7525_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7525_cast = einsum(equation = var_7525_equation_0, values = (var_7371_cast, var_7491_cast));
            tensor<string, []> var_7527_equation_0 = const()[name = tensor<string, []>("op_7527_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7527_cast = einsum(equation = var_7527_equation_0, values = (var_7375_cast, var_7492_cast));
            tensor<string, []> var_7529_equation_0 = const()[name = tensor<string, []>("op_7529_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7529_cast = einsum(equation = var_7529_equation_0, values = (var_7379_cast, var_7493_cast));
            tensor<string, []> var_7531_equation_0 = const()[name = tensor<string, []>("op_7531_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7531_cast = einsum(equation = var_7531_equation_0, values = (var_7383_cast, var_7494_cast));
            tensor<string, []> var_7533_equation_0 = const()[name = tensor<string, []>("op_7533_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7533_cast = einsum(equation = var_7533_equation_0, values = (var_7387_cast, var_7495_cast));
            tensor<string, []> var_7535_equation_0 = const()[name = tensor<string, []>("op_7535_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7535_cast = einsum(equation = var_7535_equation_0, values = (var_7391_cast, var_7496_cast));
            tensor<string, []> var_7537_equation_0 = const()[name = tensor<string, []>("op_7537_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7537_cast = einsum(equation = var_7537_equation_0, values = (var_7395_cast, var_7497_cast));
            tensor<bool, []> input_329_interleave_0 = const()[name = tensor<string, []>("input_329_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 256]> input_329_cast = concat(axis = var_5599, interleave = input_329_interleave_0, values = (var_7499_cast, var_7501_cast, var_7503_cast, var_7505_cast, var_7507_cast, var_7509_cast, var_7511_cast, var_7513_cast, var_7515_cast, var_7517_cast, var_7519_cast, var_7521_cast, var_7523_cast, var_7525_cast, var_7527_cast, var_7529_cast, var_7531_cast, var_7533_cast, var_7535_cast, var_7537_cast));
            tensor<int32, [2]> var_7543 = const()[name = tensor<string, []>("op_7543"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7545 = const()[name = tensor<string, []>("op_7545"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_7547_pad_type_0 = const()[name = tensor<string, []>("op_7547_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_7547_pad_0 = const()[name = tensor<string, []>("op_7547_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1320436928)))];
            tensor<fp16, [1280]> up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1323713792)))];
            tensor<fp16, [2, 1280, 1, 256]> var_7547_cast = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_7545, groups = var_5599, pad = var_7547_pad_0, pad_type = var_7547_pad_type_0, strides = var_7543, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_329_cast);
            tensor<fp16, [2, 1280, 1, 256]> inputs_53_cast = add(x = var_7547_cast, y = inputs_51_cast);
            tensor<int32, [1]> var_7551 = const()[name = tensor<string, []>("op_7551"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_53_cast = reduce_mean(axes = var_7551, keep_dims = var_5594, x = inputs_53_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_53_cast = sub(x = inputs_53_cast, y = channels_mean_53_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_sq_53_cast = mul(x = zero_mean_53_cast, y = zero_mean_53_cast);
            tensor<int32, [1]> var_7555 = const()[name = tensor<string, []>("op_7555"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_7556_cast = reduce_mean(axes = var_7555, keep_dims = var_5594, x = zero_mean_sq_53_cast);
            tensor<fp16, []> var_7557_to_fp16 = const()[name = tensor<string, []>("op_7557_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_7558_cast = add(x = var_7556_cast, y = var_7557_to_fp16);
            tensor<fp16, []> denom_53_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_53_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 256]> denom_53_cast = rsqrt(epsilon = denom_53_epsilon_0_to_fp16, x = var_7558_cast);
            tensor<fp16, [2, 1280, 1, 256]> out_53_cast = mul(x = zero_mean_53_cast, y = denom_53_cast);
            tensor<fp16, [1, 1280, 1, 1]> var_7562_to_fp16 = const()[name = tensor<string, []>("op_7562_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1323716416)))];
            tensor<fp16, [2, 1280, 1, 256]> var_7563_cast = add(x = out_53_cast, y = var_7562_to_fp16);
            tensor<fp16, [1, 1280, 1, 1]> var_7565_to_fp16 = const()[name = tensor<string, []>("op_7565_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1323719040)))];
            tensor<fp16, [2, 1280, 1, 256]> input_331_cast = mul(x = var_7563_cast, y = var_7565_to_fp16);
            tensor<int32, [2]> var_7573 = const()[name = tensor<string, []>("op_7573"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7575 = const()[name = tensor<string, []>("op_7575"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_7577_pad_type_0 = const()[name = tensor<string, []>("op_7577_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_7577_pad_0 = const()[name = tensor<string, []>("op_7577_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1323721664)))];
            tensor<fp16, [10240]> up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1349936128)))];
            tensor<fp16, [2, 10240, 1, 256]> var_7577_cast = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_7575, groups = var_5599, pad = var_7577_pad_0, pad_type = var_7577_pad_type_0, strides = var_7573, weight = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_331_cast);
            tensor<int32, [2]> var_7578_split_sizes_0 = const()[name = tensor<string, []>("op_7578_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_7578_axis_0 = const()[name = tensor<string, []>("op_7578_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 256]> var_7578_cast_0, tensor<fp16, [2, 5120, 1, 256]> var_7578_cast_1 = split(axis = var_7578_axis_0, split_sizes = var_7578_split_sizes_0, x = var_7577_cast);
            tensor<string, []> var_7580_mode_0 = const()[name = tensor<string, []>("op_7580_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 256]> var_7580_cast = gelu(mode = var_7580_mode_0, x = var_7578_cast_1);
            tensor<fp16, [2, 5120, 1, 256]> input_333_cast = mul(x = var_7578_cast_0, y = var_7580_cast);
            tensor<int32, [2]> var_7584 = const()[name = tensor<string, []>("op_7584"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7586 = const()[name = tensor<string, []>("op_7586"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_7588_pad_type_0 = const()[name = tensor<string, []>("op_7588_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_7588_pad_0 = const()[name = tensor<string, []>("op_7588_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1349956672)))];
            tensor<fp16, [1280]> up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1363063936)))];
            tensor<fp16, [2, 1280, 1, 256]> var_7588_cast = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_7586, groups = var_5599, pad = var_7588_pad_0, pad_type = var_7588_pad_type_0, strides = var_7584, weight = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_333_cast);
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_193_cast = add(x = var_7588_cast, y = inputs_53_cast);
            tensor<int32, [4]> var_7590 = const()[name = tensor<string, []>("op_7590"), val = tensor<int32, [4]>([2, 1280, 16, 16])];
            tensor<fp16, [2, 1280, 16, 16]> input_335_cast = reshape(shape = var_7590, x = hidden_states_193_cast);
            tensor<int32, [2]> var_7594 = const()[name = tensor<string, []>("op_7594"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7596 = const()[name = tensor<string, []>("op_7596"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_195_pad_type_0 = const()[name = tensor<string, []>("hidden_states_195_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_195_pad_0 = const()[name = tensor<string, []>("hidden_states_195_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_1_attentions_1_proj_out_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_proj_out_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1363066560)))];
            tensor<fp16, [1280]> up_blocks_1_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_1_proj_out_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1366343424)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_195_cast = conv(bias = up_blocks_1_attentions_1_proj_out_bias_to_fp16, dilations = var_7596, groups = var_5599, pad = hidden_states_195_pad_0, pad_type = hidden_states_195_pad_type_0, strides = var_7594, weight = up_blocks_1_attentions_1_proj_out_weight_to_fp16, x = input_335_cast);
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_197_cast = add(x = hidden_states_195_cast, y = hidden_states_183_cast);
            tensor<bool, []> input_337_interleave_0 = const()[name = tensor<string, []>("input_337_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1920, 16, 16]> input_337_cast = concat(axis = var_5599, interleave = input_337_interleave_0, values = (hidden_states_197_cast, input_117_cast));
            tensor<int32, [5]> reshape_156_shape_0 = const()[name = tensor<string, []>("reshape_156_shape_0"), val = tensor<int32, [5]>([2, 32, 60, 16, 16])];
            tensor<fp16, [2, 32, 60, 16, 16]> reshape_156_cast = reshape(shape = reshape_156_shape_0, x = input_337_cast);
            tensor<int32, [3]> reduce_mean_117_axes_0 = const()[name = tensor<string, []>("reduce_mean_117_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_117_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_117_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_117_cast = reduce_mean(axes = reduce_mean_117_axes_0, keep_dims = reduce_mean_117_keep_dims_0, x = reshape_156_cast);
            tensor<fp16, [2, 32, 60, 16, 16]> sub_78_cast = sub(x = reshape_156_cast, y = reduce_mean_117_cast);
            tensor<fp16, [2, 32, 60, 16, 16]> square_39_cast = square(x = sub_78_cast);
            tensor<int32, [3]> reduce_mean_119_axes_0 = const()[name = tensor<string, []>("reduce_mean_119_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_119_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_119_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_119_cast = reduce_mean(axes = reduce_mean_119_axes_0, keep_dims = reduce_mean_119_keep_dims_0, x = square_39_cast);
            tensor<fp16, []> add_78_y_0_to_fp16 = const()[name = tensor<string, []>("add_78_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_78_cast = add(x = reduce_mean_119_cast, y = add_78_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_39_cast = sqrt(x = add_78_cast);
            tensor<fp16, [2, 32, 60, 16, 16]> real_div_39_cast = real_div(x = sub_78_cast, y = sqrt_39_cast);
            tensor<int32, [4]> reshape_157_shape_0 = const()[name = tensor<string, []>("reshape_157_shape_0"), val = tensor<int32, [4]>([2, 1920, 16, 16])];
            tensor<fp16, [2, 1920, 16, 16]> reshape_157_cast = reshape(shape = reshape_157_shape_0, x = real_div_39_cast);
            tensor<fp16, [1920]> add_79_mean_0_to_fp16 = const()[name = tensor<string, []>("add_79_mean_0_to_fp16"), val = tensor<fp16, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1366346048)))];
            tensor<fp16, [1920]> add_79_variance_0_to_fp16 = const()[name = tensor<string, []>("add_79_variance_0_to_fp16"), val = tensor<fp16, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1366349952)))];
            tensor<fp16, [1920]> add_79_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_79_gamma_0_to_fp16"), val = tensor<fp16, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1366353856)))];
            tensor<fp16, [1920]> add_79_beta_0_to_fp16 = const()[name = tensor<string, []>("add_79_beta_0_to_fp16"), val = tensor<fp16, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1366357760)))];
            tensor<fp16, []> add_79_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_79_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1920, 16, 16]> add_79_cast = batch_norm(beta = add_79_beta_0_to_fp16, epsilon = add_79_epsilon_0_to_fp16, gamma = add_79_gamma_0_to_fp16, mean = add_79_mean_0_to_fp16, variance = add_79_variance_0_to_fp16, x = reshape_157_cast);
            tensor<fp16, [2, 1920, 16, 16]> input_341_cast = silu(x = add_79_cast);
            tensor<int32, [2]> var_7614 = const()[name = tensor<string, []>("op_7614"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7616 = const()[name = tensor<string, []>("op_7616"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_199_pad_type_0 = const()[name = tensor<string, []>("hidden_states_199_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_199_pad_0 = const()[name = tensor<string, []>("hidden_states_199_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1920, 3, 3]> up_blocks_1_resnets_2_conv1_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_2_conv1_weight_to_fp16"), val = tensor<fp16, [1280, 1920, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1366361664)))];
            tensor<fp16, [1280]> up_blocks_1_resnets_2_conv1_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_2_conv1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1410598528)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_199_cast = conv(bias = up_blocks_1_resnets_2_conv1_bias_to_fp16, dilations = var_7616, groups = var_5599, pad = hidden_states_199_pad_0, pad_type = hidden_states_199_pad_type_0, strides = var_7614, weight = up_blocks_1_resnets_2_conv1_weight_to_fp16, x = input_341_cast);
            tensor<int32, [2]> var_7622 = const()[name = tensor<string, []>("op_7622"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7624 = const()[name = tensor<string, []>("op_7624"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_31_pad_type_0 = const()[name = tensor<string, []>("temb_31_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_31_pad_0 = const()[name = tensor<string, []>("temb_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_1_resnets_2_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_2_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1410601152)))];
            tensor<fp16, [1280]> up_blocks_1_resnets_2_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_2_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1413878016)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_31_cast = conv(bias = up_blocks_1_resnets_2_time_emb_proj_bias_to_fp16, dilations = var_7624, groups = var_5599, pad = temb_31_pad_0, pad_type = temb_31_pad_type_0, strides = var_7622, weight = up_blocks_1_resnets_2_time_emb_proj_weight_to_fp16, x = input_15_cast);
            tensor<fp16, [2, 1280, 16, 16]> input_345_cast = add(x = hidden_states_199_cast, y = temb_31_cast);
            tensor<int32, [5]> reshape_160_shape_0 = const()[name = tensor<string, []>("reshape_160_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 16, 16])];
            tensor<fp16, [2, 32, 40, 16, 16]> reshape_160_cast = reshape(shape = reshape_160_shape_0, x = input_345_cast);
            tensor<int32, [3]> reduce_mean_120_axes_0 = const()[name = tensor<string, []>("reduce_mean_120_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_120_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_120_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_120_cast = reduce_mean(axes = reduce_mean_120_axes_0, keep_dims = reduce_mean_120_keep_dims_0, x = reshape_160_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> sub_80_cast = sub(x = reshape_160_cast, y = reduce_mean_120_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> square_40_cast = square(x = sub_80_cast);
            tensor<int32, [3]> reduce_mean_122_axes_0 = const()[name = tensor<string, []>("reduce_mean_122_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_122_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_122_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_122_cast = reduce_mean(axes = reduce_mean_122_axes_0, keep_dims = reduce_mean_122_keep_dims_0, x = square_40_cast);
            tensor<fp16, []> add_80_y_0_to_fp16 = const()[name = tensor<string, []>("add_80_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_80_cast = add(x = reduce_mean_122_cast, y = add_80_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_40_cast = sqrt(x = add_80_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> real_div_40_cast = real_div(x = sub_80_cast, y = sqrt_40_cast);
            tensor<int32, [4]> reshape_161_shape_0 = const()[name = tensor<string, []>("reshape_161_shape_0"), val = tensor<int32, [4]>([2, 1280, 16, 16])];
            tensor<fp16, [2, 1280, 16, 16]> reshape_161_cast = reshape(shape = reshape_161_shape_0, x = real_div_40_cast);
            tensor<fp16, [1280]> add_81_mean_0_to_fp16 = const()[name = tensor<string, []>("add_81_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1413880640)))];
            tensor<fp16, [1280]> add_81_variance_0_to_fp16 = const()[name = tensor<string, []>("add_81_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1413883264)))];
            tensor<fp16, [1280]> add_81_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_81_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1413885888)))];
            tensor<fp16, [1280]> add_81_beta_0_to_fp16 = const()[name = tensor<string, []>("add_81_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1413888512)))];
            tensor<fp16, []> add_81_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_81_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 16, 16]> add_81_cast = batch_norm(beta = add_81_beta_0_to_fp16, epsilon = add_81_epsilon_0_to_fp16, gamma = add_81_gamma_0_to_fp16, mean = add_81_mean_0_to_fp16, variance = add_81_variance_0_to_fp16, x = reshape_161_cast);
            tensor<fp16, [2, 1280, 16, 16]> input_349_cast = silu(x = add_81_cast);
            tensor<int32, [2]> var_7634 = const()[name = tensor<string, []>("op_7634"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7636 = const()[name = tensor<string, []>("op_7636"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_201_pad_type_0 = const()[name = tensor<string, []>("hidden_states_201_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_201_pad_0 = const()[name = tensor<string, []>("hidden_states_201_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> up_blocks_1_resnets_2_conv2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_2_conv2_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1413891136)))];
            tensor<fp16, [1280]> up_blocks_1_resnets_2_conv2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_2_conv2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1443382400)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_201_cast = conv(bias = up_blocks_1_resnets_2_conv2_bias_to_fp16, dilations = var_7636, groups = var_5599, pad = hidden_states_201_pad_0, pad_type = hidden_states_201_pad_type_0, strides = var_7634, weight = up_blocks_1_resnets_2_conv2_weight_to_fp16, x = input_349_cast);
            tensor<int32, [2]> var_7641 = const()[name = tensor<string, []>("op_7641"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7643 = const()[name = tensor<string, []>("op_7643"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> x_15_pad_type_0 = const()[name = tensor<string, []>("x_15_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> x_15_pad_0 = const()[name = tensor<string, []>("x_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1920, 1, 1]> up_blocks_1_resnets_2_conv_shortcut_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_2_conv_shortcut_weight_to_fp16"), val = tensor<fp16, [1280, 1920, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1443385024)))];
            tensor<fp16, [1280]> up_blocks_1_resnets_2_conv_shortcut_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_resnets_2_conv_shortcut_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1448300288)))];
            tensor<fp16, [2, 1280, 16, 16]> x_15_cast = conv(bias = up_blocks_1_resnets_2_conv_shortcut_bias_to_fp16, dilations = var_7643, groups = var_5599, pad = x_15_pad_0, pad_type = x_15_pad_type_0, strides = var_7641, weight = up_blocks_1_resnets_2_conv_shortcut_weight_to_fp16, x = input_337_cast);
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_203_cast = add(x = x_15_cast, y = hidden_states_201_cast);
            tensor<int32, [5]> reshape_164_shape_0 = const()[name = tensor<string, []>("reshape_164_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 16, 16])];
            tensor<fp16, [2, 32, 40, 16, 16]> reshape_164_cast = reshape(shape = reshape_164_shape_0, x = hidden_states_203_cast);
            tensor<int32, [3]> reduce_mean_123_axes_0 = const()[name = tensor<string, []>("reduce_mean_123_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_123_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_123_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_123_cast = reduce_mean(axes = reduce_mean_123_axes_0, keep_dims = reduce_mean_123_keep_dims_0, x = reshape_164_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> sub_82_cast = sub(x = reshape_164_cast, y = reduce_mean_123_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> square_41_cast = square(x = sub_82_cast);
            tensor<int32, [3]> reduce_mean_125_axes_0 = const()[name = tensor<string, []>("reduce_mean_125_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_125_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_125_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_125_cast = reduce_mean(axes = reduce_mean_125_axes_0, keep_dims = reduce_mean_125_keep_dims_0, x = square_41_cast);
            tensor<fp16, []> add_82_y_0_to_fp16 = const()[name = tensor<string, []>("add_82_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_82_cast = add(x = reduce_mean_125_cast, y = add_82_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_41_cast = sqrt(x = add_82_cast);
            tensor<fp16, [2, 32, 40, 16, 16]> real_div_41_cast = real_div(x = sub_82_cast, y = sqrt_41_cast);
            tensor<int32, [4]> reshape_165_shape_0 = const()[name = tensor<string, []>("reshape_165_shape_0"), val = tensor<int32, [4]>([2, 1280, 16, 16])];
            tensor<fp16, [2, 1280, 16, 16]> reshape_165_cast = reshape(shape = reshape_165_shape_0, x = real_div_41_cast);
            tensor<fp16, [1280]> add_83_mean_0_to_fp16 = const()[name = tensor<string, []>("add_83_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1448302912)))];
            tensor<fp16, [1280]> add_83_variance_0_to_fp16 = const()[name = tensor<string, []>("add_83_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1448305536)))];
            tensor<fp16, [1280]> add_83_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_83_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1448308160)))];
            tensor<fp16, [1280]> add_83_beta_0_to_fp16 = const()[name = tensor<string, []>("add_83_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1448310784)))];
            tensor<fp16, []> add_83_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_83_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 16, 16]> add_83_cast = batch_norm(beta = add_83_beta_0_to_fp16, epsilon = add_83_epsilon_0_to_fp16, gamma = add_83_gamma_0_to_fp16, mean = add_83_mean_0_to_fp16, variance = add_83_variance_0_to_fp16, x = reshape_165_cast);
            tensor<int32, [2]> var_7663 = const()[name = tensor<string, []>("op_7663"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7665 = const()[name = tensor<string, []>("op_7665"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_205_pad_type_0 = const()[name = tensor<string, []>("hidden_states_205_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_205_pad_0 = const()[name = tensor<string, []>("hidden_states_205_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_1_attentions_2_proj_in_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_proj_in_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1448313408)))];
            tensor<fp16, [1280]> up_blocks_1_attentions_2_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_proj_in_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1451590272)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_205_cast = conv(bias = up_blocks_1_attentions_2_proj_in_bias_to_fp16, dilations = var_7665, groups = var_5599, pad = hidden_states_205_pad_0, pad_type = hidden_states_205_pad_type_0, strides = var_7663, weight = up_blocks_1_attentions_2_proj_in_weight_to_fp16, x = add_83_cast);
            tensor<int32, [4]> var_7670 = const()[name = tensor<string, []>("op_7670"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<fp16, [2, 1280, 1, 256]> inputs_55_cast = reshape(shape = var_7670, x = hidden_states_205_cast);
            tensor<int32, [1]> var_7680 = const()[name = tensor<string, []>("op_7680"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_55_cast = reduce_mean(axes = var_7680, keep_dims = var_5594, x = inputs_55_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_55_cast = sub(x = inputs_55_cast, y = channels_mean_55_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_sq_55_cast = mul(x = zero_mean_55_cast, y = zero_mean_55_cast);
            tensor<int32, [1]> var_7684 = const()[name = tensor<string, []>("op_7684"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_7685_cast = reduce_mean(axes = var_7684, keep_dims = var_5594, x = zero_mean_sq_55_cast);
            tensor<fp16, []> var_7686_to_fp16 = const()[name = tensor<string, []>("op_7686_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_7687_cast = add(x = var_7685_cast, y = var_7686_to_fp16);
            tensor<fp16, []> denom_55_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_55_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 256]> denom_55_cast = rsqrt(epsilon = denom_55_epsilon_0_to_fp16, x = var_7687_cast);
            tensor<fp16, [2, 1280, 1, 256]> out_55_cast = mul(x = zero_mean_55_cast, y = denom_55_cast);
            tensor<fp16, [1, 1280, 1, 1]> var_7691_to_fp16 = const()[name = tensor<string, []>("op_7691_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1451592896)))];
            tensor<fp16, [2, 1280, 1, 256]> var_7692_cast = add(x = out_55_cast, y = var_7691_to_fp16);
            tensor<fp16, [1, 1280, 1, 1]> var_7694_to_fp16 = const()[name = tensor<string, []>("op_7694_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1451595520)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_207_cast = mul(x = var_7692_cast, y = var_7694_to_fp16);
            tensor<int32, [2]> var_7701 = const()[name = tensor<string, []>("op_7701"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7703 = const()[name = tensor<string, []>("op_7703"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_37_pad_type_0 = const()[name = tensor<string, []>("q_37_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_37_pad_0 = const()[name = tensor<string, []>("q_37_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1451598144)))];
            tensor<fp16, [2, 1280, 1, 256]> q_37_cast = conv(dilations = var_7703, groups = var_5599, pad = q_37_pad_0, pad_type = q_37_pad_type_0, strides = var_7701, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_207_cast);
            tensor<int32, [2]> var_7707 = const()[name = tensor<string, []>("op_7707"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7709 = const()[name = tensor<string, []>("op_7709"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_73_pad_type_0 = const()[name = tensor<string, []>("k_73_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_73_pad_0 = const()[name = tensor<string, []>("k_73_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1454875008)))];
            tensor<fp16, [2, 1280, 1, 256]> k_73_cast = conv(dilations = var_7709, groups = var_5599, pad = k_73_pad_0, pad_type = k_73_pad_type_0, strides = var_7707, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_207_cast);
            tensor<int32, [2]> var_7713 = const()[name = tensor<string, []>("op_7713"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7715 = const()[name = tensor<string, []>("op_7715"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_37_pad_type_0 = const()[name = tensor<string, []>("v_37_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_37_pad_0 = const()[name = tensor<string, []>("v_37_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1458151872)))];
            tensor<fp16, [2, 1280, 1, 256]> v_37_cast = conv(dilations = var_7715, groups = var_5599, pad = v_37_pad_0, pad_type = v_37_pad_type_0, strides = var_7713, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_207_cast);
            tensor<int32, [4]> var_7719_begin_0 = const()[name = tensor<string, []>("op_7719_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7719_end_0 = const()[name = tensor<string, []>("op_7719_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_7719_end_mask_0 = const()[name = tensor<string, []>("op_7719_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7719_cast = slice_by_index(begin = var_7719_begin_0, end = var_7719_end_0, end_mask = var_7719_end_mask_0, x = q_37_cast);
            tensor<int32, [4]> var_7723_begin_0 = const()[name = tensor<string, []>("op_7723_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_7723_end_0 = const()[name = tensor<string, []>("op_7723_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_7723_end_mask_0 = const()[name = tensor<string, []>("op_7723_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7723_cast = slice_by_index(begin = var_7723_begin_0, end = var_7723_end_0, end_mask = var_7723_end_mask_0, x = q_37_cast);
            tensor<int32, [4]> var_7727_begin_0 = const()[name = tensor<string, []>("op_7727_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_7727_end_0 = const()[name = tensor<string, []>("op_7727_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_7727_end_mask_0 = const()[name = tensor<string, []>("op_7727_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7727_cast = slice_by_index(begin = var_7727_begin_0, end = var_7727_end_0, end_mask = var_7727_end_mask_0, x = q_37_cast);
            tensor<int32, [4]> var_7731_begin_0 = const()[name = tensor<string, []>("op_7731_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_7731_end_0 = const()[name = tensor<string, []>("op_7731_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_7731_end_mask_0 = const()[name = tensor<string, []>("op_7731_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7731_cast = slice_by_index(begin = var_7731_begin_0, end = var_7731_end_0, end_mask = var_7731_end_mask_0, x = q_37_cast);
            tensor<int32, [4]> var_7735_begin_0 = const()[name = tensor<string, []>("op_7735_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_7735_end_0 = const()[name = tensor<string, []>("op_7735_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_7735_end_mask_0 = const()[name = tensor<string, []>("op_7735_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7735_cast = slice_by_index(begin = var_7735_begin_0, end = var_7735_end_0, end_mask = var_7735_end_mask_0, x = q_37_cast);
            tensor<int32, [4]> var_7739_begin_0 = const()[name = tensor<string, []>("op_7739_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_7739_end_0 = const()[name = tensor<string, []>("op_7739_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_7739_end_mask_0 = const()[name = tensor<string, []>("op_7739_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7739_cast = slice_by_index(begin = var_7739_begin_0, end = var_7739_end_0, end_mask = var_7739_end_mask_0, x = q_37_cast);
            tensor<int32, [4]> var_7743_begin_0 = const()[name = tensor<string, []>("op_7743_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_7743_end_0 = const()[name = tensor<string, []>("op_7743_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_7743_end_mask_0 = const()[name = tensor<string, []>("op_7743_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7743_cast = slice_by_index(begin = var_7743_begin_0, end = var_7743_end_0, end_mask = var_7743_end_mask_0, x = q_37_cast);
            tensor<int32, [4]> var_7747_begin_0 = const()[name = tensor<string, []>("op_7747_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_7747_end_0 = const()[name = tensor<string, []>("op_7747_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_7747_end_mask_0 = const()[name = tensor<string, []>("op_7747_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7747_cast = slice_by_index(begin = var_7747_begin_0, end = var_7747_end_0, end_mask = var_7747_end_mask_0, x = q_37_cast);
            tensor<int32, [4]> var_7751_begin_0 = const()[name = tensor<string, []>("op_7751_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_7751_end_0 = const()[name = tensor<string, []>("op_7751_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_7751_end_mask_0 = const()[name = tensor<string, []>("op_7751_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7751_cast = slice_by_index(begin = var_7751_begin_0, end = var_7751_end_0, end_mask = var_7751_end_mask_0, x = q_37_cast);
            tensor<int32, [4]> var_7755_begin_0 = const()[name = tensor<string, []>("op_7755_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_7755_end_0 = const()[name = tensor<string, []>("op_7755_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_7755_end_mask_0 = const()[name = tensor<string, []>("op_7755_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7755_cast = slice_by_index(begin = var_7755_begin_0, end = var_7755_end_0, end_mask = var_7755_end_mask_0, x = q_37_cast);
            tensor<int32, [4]> var_7759_begin_0 = const()[name = tensor<string, []>("op_7759_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_7759_end_0 = const()[name = tensor<string, []>("op_7759_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_7759_end_mask_0 = const()[name = tensor<string, []>("op_7759_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7759_cast = slice_by_index(begin = var_7759_begin_0, end = var_7759_end_0, end_mask = var_7759_end_mask_0, x = q_37_cast);
            tensor<int32, [4]> var_7763_begin_0 = const()[name = tensor<string, []>("op_7763_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_7763_end_0 = const()[name = tensor<string, []>("op_7763_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_7763_end_mask_0 = const()[name = tensor<string, []>("op_7763_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7763_cast = slice_by_index(begin = var_7763_begin_0, end = var_7763_end_0, end_mask = var_7763_end_mask_0, x = q_37_cast);
            tensor<int32, [4]> var_7767_begin_0 = const()[name = tensor<string, []>("op_7767_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_7767_end_0 = const()[name = tensor<string, []>("op_7767_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_7767_end_mask_0 = const()[name = tensor<string, []>("op_7767_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7767_cast = slice_by_index(begin = var_7767_begin_0, end = var_7767_end_0, end_mask = var_7767_end_mask_0, x = q_37_cast);
            tensor<int32, [4]> var_7771_begin_0 = const()[name = tensor<string, []>("op_7771_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_7771_end_0 = const()[name = tensor<string, []>("op_7771_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_7771_end_mask_0 = const()[name = tensor<string, []>("op_7771_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7771_cast = slice_by_index(begin = var_7771_begin_0, end = var_7771_end_0, end_mask = var_7771_end_mask_0, x = q_37_cast);
            tensor<int32, [4]> var_7775_begin_0 = const()[name = tensor<string, []>("op_7775_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_7775_end_0 = const()[name = tensor<string, []>("op_7775_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_7775_end_mask_0 = const()[name = tensor<string, []>("op_7775_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7775_cast = slice_by_index(begin = var_7775_begin_0, end = var_7775_end_0, end_mask = var_7775_end_mask_0, x = q_37_cast);
            tensor<int32, [4]> var_7779_begin_0 = const()[name = tensor<string, []>("op_7779_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_7779_end_0 = const()[name = tensor<string, []>("op_7779_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_7779_end_mask_0 = const()[name = tensor<string, []>("op_7779_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7779_cast = slice_by_index(begin = var_7779_begin_0, end = var_7779_end_0, end_mask = var_7779_end_mask_0, x = q_37_cast);
            tensor<int32, [4]> var_7783_begin_0 = const()[name = tensor<string, []>("op_7783_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_7783_end_0 = const()[name = tensor<string, []>("op_7783_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_7783_end_mask_0 = const()[name = tensor<string, []>("op_7783_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7783_cast = slice_by_index(begin = var_7783_begin_0, end = var_7783_end_0, end_mask = var_7783_end_mask_0, x = q_37_cast);
            tensor<int32, [4]> var_7787_begin_0 = const()[name = tensor<string, []>("op_7787_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_7787_end_0 = const()[name = tensor<string, []>("op_7787_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_7787_end_mask_0 = const()[name = tensor<string, []>("op_7787_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7787_cast = slice_by_index(begin = var_7787_begin_0, end = var_7787_end_0, end_mask = var_7787_end_mask_0, x = q_37_cast);
            tensor<int32, [4]> var_7791_begin_0 = const()[name = tensor<string, []>("op_7791_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_7791_end_0 = const()[name = tensor<string, []>("op_7791_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_7791_end_mask_0 = const()[name = tensor<string, []>("op_7791_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7791_cast = slice_by_index(begin = var_7791_begin_0, end = var_7791_end_0, end_mask = var_7791_end_mask_0, x = q_37_cast);
            tensor<int32, [4]> var_7795_begin_0 = const()[name = tensor<string, []>("op_7795_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_7795_end_0 = const()[name = tensor<string, []>("op_7795_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_7795_end_mask_0 = const()[name = tensor<string, []>("op_7795_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7795_cast = slice_by_index(begin = var_7795_begin_0, end = var_7795_end_0, end_mask = var_7795_end_mask_0, x = q_37_cast);
            tensor<int32, [4]> k_75_perm_0 = const()[name = tensor<string, []>("k_75_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_7802_begin_0 = const()[name = tensor<string, []>("op_7802_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7802_end_0 = const()[name = tensor<string, []>("op_7802_end_0"), val = tensor<int32, [4]>([2, 256, 1, 64])];
            tensor<bool, [4]> var_7802_end_mask_0 = const()[name = tensor<string, []>("op_7802_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 1280]> transpose_13 = transpose(perm = k_75_perm_0, x = k_73_cast);
            tensor<fp16, [2, 256, 1, 64]> var_7802_cast = slice_by_index(begin = var_7802_begin_0, end = var_7802_end_0, end_mask = var_7802_end_mask_0, x = transpose_13);
            tensor<int32, [4]> var_7806_begin_0 = const()[name = tensor<string, []>("op_7806_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_7806_end_0 = const()[name = tensor<string, []>("op_7806_end_0"), val = tensor<int32, [4]>([2, 256, 1, 128])];
            tensor<bool, [4]> var_7806_end_mask_0 = const()[name = tensor<string, []>("op_7806_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7806_cast = slice_by_index(begin = var_7806_begin_0, end = var_7806_end_0, end_mask = var_7806_end_mask_0, x = transpose_13);
            tensor<int32, [4]> var_7810_begin_0 = const()[name = tensor<string, []>("op_7810_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_7810_end_0 = const()[name = tensor<string, []>("op_7810_end_0"), val = tensor<int32, [4]>([2, 256, 1, 192])];
            tensor<bool, [4]> var_7810_end_mask_0 = const()[name = tensor<string, []>("op_7810_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7810_cast = slice_by_index(begin = var_7810_begin_0, end = var_7810_end_0, end_mask = var_7810_end_mask_0, x = transpose_13);
            tensor<int32, [4]> var_7814_begin_0 = const()[name = tensor<string, []>("op_7814_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_7814_end_0 = const()[name = tensor<string, []>("op_7814_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_7814_end_mask_0 = const()[name = tensor<string, []>("op_7814_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7814_cast = slice_by_index(begin = var_7814_begin_0, end = var_7814_end_0, end_mask = var_7814_end_mask_0, x = transpose_13);
            tensor<int32, [4]> var_7818_begin_0 = const()[name = tensor<string, []>("op_7818_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_7818_end_0 = const()[name = tensor<string, []>("op_7818_end_0"), val = tensor<int32, [4]>([2, 256, 1, 320])];
            tensor<bool, [4]> var_7818_end_mask_0 = const()[name = tensor<string, []>("op_7818_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7818_cast = slice_by_index(begin = var_7818_begin_0, end = var_7818_end_0, end_mask = var_7818_end_mask_0, x = transpose_13);
            tensor<int32, [4]> var_7822_begin_0 = const()[name = tensor<string, []>("op_7822_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_7822_end_0 = const()[name = tensor<string, []>("op_7822_end_0"), val = tensor<int32, [4]>([2, 256, 1, 384])];
            tensor<bool, [4]> var_7822_end_mask_0 = const()[name = tensor<string, []>("op_7822_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7822_cast = slice_by_index(begin = var_7822_begin_0, end = var_7822_end_0, end_mask = var_7822_end_mask_0, x = transpose_13);
            tensor<int32, [4]> var_7826_begin_0 = const()[name = tensor<string, []>("op_7826_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_7826_end_0 = const()[name = tensor<string, []>("op_7826_end_0"), val = tensor<int32, [4]>([2, 256, 1, 448])];
            tensor<bool, [4]> var_7826_end_mask_0 = const()[name = tensor<string, []>("op_7826_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7826_cast = slice_by_index(begin = var_7826_begin_0, end = var_7826_end_0, end_mask = var_7826_end_mask_0, x = transpose_13);
            tensor<int32, [4]> var_7830_begin_0 = const()[name = tensor<string, []>("op_7830_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_7830_end_0 = const()[name = tensor<string, []>("op_7830_end_0"), val = tensor<int32, [4]>([2, 256, 1, 512])];
            tensor<bool, [4]> var_7830_end_mask_0 = const()[name = tensor<string, []>("op_7830_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7830_cast = slice_by_index(begin = var_7830_begin_0, end = var_7830_end_0, end_mask = var_7830_end_mask_0, x = transpose_13);
            tensor<int32, [4]> var_7834_begin_0 = const()[name = tensor<string, []>("op_7834_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_7834_end_0 = const()[name = tensor<string, []>("op_7834_end_0"), val = tensor<int32, [4]>([2, 256, 1, 576])];
            tensor<bool, [4]> var_7834_end_mask_0 = const()[name = tensor<string, []>("op_7834_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7834_cast = slice_by_index(begin = var_7834_begin_0, end = var_7834_end_0, end_mask = var_7834_end_mask_0, x = transpose_13);
            tensor<int32, [4]> var_7838_begin_0 = const()[name = tensor<string, []>("op_7838_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_7838_end_0 = const()[name = tensor<string, []>("op_7838_end_0"), val = tensor<int32, [4]>([2, 256, 1, 640])];
            tensor<bool, [4]> var_7838_end_mask_0 = const()[name = tensor<string, []>("op_7838_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7838_cast = slice_by_index(begin = var_7838_begin_0, end = var_7838_end_0, end_mask = var_7838_end_mask_0, x = transpose_13);
            tensor<int32, [4]> var_7842_begin_0 = const()[name = tensor<string, []>("op_7842_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_7842_end_0 = const()[name = tensor<string, []>("op_7842_end_0"), val = tensor<int32, [4]>([2, 256, 1, 704])];
            tensor<bool, [4]> var_7842_end_mask_0 = const()[name = tensor<string, []>("op_7842_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7842_cast = slice_by_index(begin = var_7842_begin_0, end = var_7842_end_0, end_mask = var_7842_end_mask_0, x = transpose_13);
            tensor<int32, [4]> var_7846_begin_0 = const()[name = tensor<string, []>("op_7846_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_7846_end_0 = const()[name = tensor<string, []>("op_7846_end_0"), val = tensor<int32, [4]>([2, 256, 1, 768])];
            tensor<bool, [4]> var_7846_end_mask_0 = const()[name = tensor<string, []>("op_7846_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7846_cast = slice_by_index(begin = var_7846_begin_0, end = var_7846_end_0, end_mask = var_7846_end_mask_0, x = transpose_13);
            tensor<int32, [4]> var_7850_begin_0 = const()[name = tensor<string, []>("op_7850_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_7850_end_0 = const()[name = tensor<string, []>("op_7850_end_0"), val = tensor<int32, [4]>([2, 256, 1, 832])];
            tensor<bool, [4]> var_7850_end_mask_0 = const()[name = tensor<string, []>("op_7850_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7850_cast = slice_by_index(begin = var_7850_begin_0, end = var_7850_end_0, end_mask = var_7850_end_mask_0, x = transpose_13);
            tensor<int32, [4]> var_7854_begin_0 = const()[name = tensor<string, []>("op_7854_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_7854_end_0 = const()[name = tensor<string, []>("op_7854_end_0"), val = tensor<int32, [4]>([2, 256, 1, 896])];
            tensor<bool, [4]> var_7854_end_mask_0 = const()[name = tensor<string, []>("op_7854_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7854_cast = slice_by_index(begin = var_7854_begin_0, end = var_7854_end_0, end_mask = var_7854_end_mask_0, x = transpose_13);
            tensor<int32, [4]> var_7858_begin_0 = const()[name = tensor<string, []>("op_7858_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_7858_end_0 = const()[name = tensor<string, []>("op_7858_end_0"), val = tensor<int32, [4]>([2, 256, 1, 960])];
            tensor<bool, [4]> var_7858_end_mask_0 = const()[name = tensor<string, []>("op_7858_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7858_cast = slice_by_index(begin = var_7858_begin_0, end = var_7858_end_0, end_mask = var_7858_end_mask_0, x = transpose_13);
            tensor<int32, [4]> var_7862_begin_0 = const()[name = tensor<string, []>("op_7862_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_7862_end_0 = const()[name = tensor<string, []>("op_7862_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_7862_end_mask_0 = const()[name = tensor<string, []>("op_7862_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7862_cast = slice_by_index(begin = var_7862_begin_0, end = var_7862_end_0, end_mask = var_7862_end_mask_0, x = transpose_13);
            tensor<int32, [4]> var_7866_begin_0 = const()[name = tensor<string, []>("op_7866_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_7866_end_0 = const()[name = tensor<string, []>("op_7866_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1088])];
            tensor<bool, [4]> var_7866_end_mask_0 = const()[name = tensor<string, []>("op_7866_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7866_cast = slice_by_index(begin = var_7866_begin_0, end = var_7866_end_0, end_mask = var_7866_end_mask_0, x = transpose_13);
            tensor<int32, [4]> var_7870_begin_0 = const()[name = tensor<string, []>("op_7870_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_7870_end_0 = const()[name = tensor<string, []>("op_7870_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1152])];
            tensor<bool, [4]> var_7870_end_mask_0 = const()[name = tensor<string, []>("op_7870_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7870_cast = slice_by_index(begin = var_7870_begin_0, end = var_7870_end_0, end_mask = var_7870_end_mask_0, x = transpose_13);
            tensor<int32, [4]> var_7874_begin_0 = const()[name = tensor<string, []>("op_7874_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_7874_end_0 = const()[name = tensor<string, []>("op_7874_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1216])];
            tensor<bool, [4]> var_7874_end_mask_0 = const()[name = tensor<string, []>("op_7874_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7874_cast = slice_by_index(begin = var_7874_begin_0, end = var_7874_end_0, end_mask = var_7874_end_mask_0, x = transpose_13);
            tensor<int32, [4]> var_7878_begin_0 = const()[name = tensor<string, []>("op_7878_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_7878_end_0 = const()[name = tensor<string, []>("op_7878_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1280])];
            tensor<bool, [4]> var_7878_end_mask_0 = const()[name = tensor<string, []>("op_7878_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7878_cast = slice_by_index(begin = var_7878_begin_0, end = var_7878_end_0, end_mask = var_7878_end_mask_0, x = transpose_13);
            tensor<int32, [4]> var_7880_begin_0 = const()[name = tensor<string, []>("op_7880_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7880_end_0 = const()[name = tensor<string, []>("op_7880_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_7880_end_mask_0 = const()[name = tensor<string, []>("op_7880_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7880_cast = slice_by_index(begin = var_7880_begin_0, end = var_7880_end_0, end_mask = var_7880_end_mask_0, x = v_37_cast);
            tensor<int32, [4]> var_7884_begin_0 = const()[name = tensor<string, []>("op_7884_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_7884_end_0 = const()[name = tensor<string, []>("op_7884_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_7884_end_mask_0 = const()[name = tensor<string, []>("op_7884_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7884_cast = slice_by_index(begin = var_7884_begin_0, end = var_7884_end_0, end_mask = var_7884_end_mask_0, x = v_37_cast);
            tensor<int32, [4]> var_7888_begin_0 = const()[name = tensor<string, []>("op_7888_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_7888_end_0 = const()[name = tensor<string, []>("op_7888_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_7888_end_mask_0 = const()[name = tensor<string, []>("op_7888_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7888_cast = slice_by_index(begin = var_7888_begin_0, end = var_7888_end_0, end_mask = var_7888_end_mask_0, x = v_37_cast);
            tensor<int32, [4]> var_7892_begin_0 = const()[name = tensor<string, []>("op_7892_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_7892_end_0 = const()[name = tensor<string, []>("op_7892_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_7892_end_mask_0 = const()[name = tensor<string, []>("op_7892_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7892_cast = slice_by_index(begin = var_7892_begin_0, end = var_7892_end_0, end_mask = var_7892_end_mask_0, x = v_37_cast);
            tensor<int32, [4]> var_7896_begin_0 = const()[name = tensor<string, []>("op_7896_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_7896_end_0 = const()[name = tensor<string, []>("op_7896_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_7896_end_mask_0 = const()[name = tensor<string, []>("op_7896_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7896_cast = slice_by_index(begin = var_7896_begin_0, end = var_7896_end_0, end_mask = var_7896_end_mask_0, x = v_37_cast);
            tensor<int32, [4]> var_7900_begin_0 = const()[name = tensor<string, []>("op_7900_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_7900_end_0 = const()[name = tensor<string, []>("op_7900_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_7900_end_mask_0 = const()[name = tensor<string, []>("op_7900_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7900_cast = slice_by_index(begin = var_7900_begin_0, end = var_7900_end_0, end_mask = var_7900_end_mask_0, x = v_37_cast);
            tensor<int32, [4]> var_7904_begin_0 = const()[name = tensor<string, []>("op_7904_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_7904_end_0 = const()[name = tensor<string, []>("op_7904_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_7904_end_mask_0 = const()[name = tensor<string, []>("op_7904_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7904_cast = slice_by_index(begin = var_7904_begin_0, end = var_7904_end_0, end_mask = var_7904_end_mask_0, x = v_37_cast);
            tensor<int32, [4]> var_7908_begin_0 = const()[name = tensor<string, []>("op_7908_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_7908_end_0 = const()[name = tensor<string, []>("op_7908_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_7908_end_mask_0 = const()[name = tensor<string, []>("op_7908_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7908_cast = slice_by_index(begin = var_7908_begin_0, end = var_7908_end_0, end_mask = var_7908_end_mask_0, x = v_37_cast);
            tensor<int32, [4]> var_7912_begin_0 = const()[name = tensor<string, []>("op_7912_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_7912_end_0 = const()[name = tensor<string, []>("op_7912_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_7912_end_mask_0 = const()[name = tensor<string, []>("op_7912_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7912_cast = slice_by_index(begin = var_7912_begin_0, end = var_7912_end_0, end_mask = var_7912_end_mask_0, x = v_37_cast);
            tensor<int32, [4]> var_7916_begin_0 = const()[name = tensor<string, []>("op_7916_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_7916_end_0 = const()[name = tensor<string, []>("op_7916_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_7916_end_mask_0 = const()[name = tensor<string, []>("op_7916_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7916_cast = slice_by_index(begin = var_7916_begin_0, end = var_7916_end_0, end_mask = var_7916_end_mask_0, x = v_37_cast);
            tensor<int32, [4]> var_7920_begin_0 = const()[name = tensor<string, []>("op_7920_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_7920_end_0 = const()[name = tensor<string, []>("op_7920_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_7920_end_mask_0 = const()[name = tensor<string, []>("op_7920_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7920_cast = slice_by_index(begin = var_7920_begin_0, end = var_7920_end_0, end_mask = var_7920_end_mask_0, x = v_37_cast);
            tensor<int32, [4]> var_7924_begin_0 = const()[name = tensor<string, []>("op_7924_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_7924_end_0 = const()[name = tensor<string, []>("op_7924_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_7924_end_mask_0 = const()[name = tensor<string, []>("op_7924_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7924_cast = slice_by_index(begin = var_7924_begin_0, end = var_7924_end_0, end_mask = var_7924_end_mask_0, x = v_37_cast);
            tensor<int32, [4]> var_7928_begin_0 = const()[name = tensor<string, []>("op_7928_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_7928_end_0 = const()[name = tensor<string, []>("op_7928_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_7928_end_mask_0 = const()[name = tensor<string, []>("op_7928_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7928_cast = slice_by_index(begin = var_7928_begin_0, end = var_7928_end_0, end_mask = var_7928_end_mask_0, x = v_37_cast);
            tensor<int32, [4]> var_7932_begin_0 = const()[name = tensor<string, []>("op_7932_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_7932_end_0 = const()[name = tensor<string, []>("op_7932_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_7932_end_mask_0 = const()[name = tensor<string, []>("op_7932_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7932_cast = slice_by_index(begin = var_7932_begin_0, end = var_7932_end_0, end_mask = var_7932_end_mask_0, x = v_37_cast);
            tensor<int32, [4]> var_7936_begin_0 = const()[name = tensor<string, []>("op_7936_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_7936_end_0 = const()[name = tensor<string, []>("op_7936_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_7936_end_mask_0 = const()[name = tensor<string, []>("op_7936_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7936_cast = slice_by_index(begin = var_7936_begin_0, end = var_7936_end_0, end_mask = var_7936_end_mask_0, x = v_37_cast);
            tensor<int32, [4]> var_7940_begin_0 = const()[name = tensor<string, []>("op_7940_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_7940_end_0 = const()[name = tensor<string, []>("op_7940_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_7940_end_mask_0 = const()[name = tensor<string, []>("op_7940_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7940_cast = slice_by_index(begin = var_7940_begin_0, end = var_7940_end_0, end_mask = var_7940_end_mask_0, x = v_37_cast);
            tensor<int32, [4]> var_7944_begin_0 = const()[name = tensor<string, []>("op_7944_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_7944_end_0 = const()[name = tensor<string, []>("op_7944_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_7944_end_mask_0 = const()[name = tensor<string, []>("op_7944_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7944_cast = slice_by_index(begin = var_7944_begin_0, end = var_7944_end_0, end_mask = var_7944_end_mask_0, x = v_37_cast);
            tensor<int32, [4]> var_7948_begin_0 = const()[name = tensor<string, []>("op_7948_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_7948_end_0 = const()[name = tensor<string, []>("op_7948_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_7948_end_mask_0 = const()[name = tensor<string, []>("op_7948_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7948_cast = slice_by_index(begin = var_7948_begin_0, end = var_7948_end_0, end_mask = var_7948_end_mask_0, x = v_37_cast);
            tensor<int32, [4]> var_7952_begin_0 = const()[name = tensor<string, []>("op_7952_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_7952_end_0 = const()[name = tensor<string, []>("op_7952_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_7952_end_mask_0 = const()[name = tensor<string, []>("op_7952_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7952_cast = slice_by_index(begin = var_7952_begin_0, end = var_7952_end_0, end_mask = var_7952_end_mask_0, x = v_37_cast);
            tensor<int32, [4]> var_7956_begin_0 = const()[name = tensor<string, []>("op_7956_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_7956_end_0 = const()[name = tensor<string, []>("op_7956_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_7956_end_mask_0 = const()[name = tensor<string, []>("op_7956_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7956_cast = slice_by_index(begin = var_7956_begin_0, end = var_7956_end_0, end_mask = var_7956_end_mask_0, x = v_37_cast);
            tensor<string, []> var_7960_equation_0 = const()[name = tensor<string, []>("op_7960_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7960_cast = einsum(equation = var_7960_equation_0, values = (var_7802_cast, var_7719_cast));
            tensor<fp16, []> var_7961_to_fp16 = const()[name = tensor<string, []>("op_7961_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_521_cast = mul(x = var_7960_cast, y = var_7961_to_fp16);
            tensor<string, []> var_7964_equation_0 = const()[name = tensor<string, []>("op_7964_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7964_cast = einsum(equation = var_7964_equation_0, values = (var_7806_cast, var_7723_cast));
            tensor<fp16, []> var_7965_to_fp16 = const()[name = tensor<string, []>("op_7965_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_523_cast = mul(x = var_7964_cast, y = var_7965_to_fp16);
            tensor<string, []> var_7968_equation_0 = const()[name = tensor<string, []>("op_7968_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7968_cast = einsum(equation = var_7968_equation_0, values = (var_7810_cast, var_7727_cast));
            tensor<fp16, []> var_7969_to_fp16 = const()[name = tensor<string, []>("op_7969_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_525_cast = mul(x = var_7968_cast, y = var_7969_to_fp16);
            tensor<string, []> var_7972_equation_0 = const()[name = tensor<string, []>("op_7972_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7972_cast = einsum(equation = var_7972_equation_0, values = (var_7814_cast, var_7731_cast));
            tensor<fp16, []> var_7973_to_fp16 = const()[name = tensor<string, []>("op_7973_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_527_cast = mul(x = var_7972_cast, y = var_7973_to_fp16);
            tensor<string, []> var_7976_equation_0 = const()[name = tensor<string, []>("op_7976_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7976_cast = einsum(equation = var_7976_equation_0, values = (var_7818_cast, var_7735_cast));
            tensor<fp16, []> var_7977_to_fp16 = const()[name = tensor<string, []>("op_7977_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_529_cast = mul(x = var_7976_cast, y = var_7977_to_fp16);
            tensor<string, []> var_7980_equation_0 = const()[name = tensor<string, []>("op_7980_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7980_cast = einsum(equation = var_7980_equation_0, values = (var_7822_cast, var_7739_cast));
            tensor<fp16, []> var_7981_to_fp16 = const()[name = tensor<string, []>("op_7981_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_531_cast = mul(x = var_7980_cast, y = var_7981_to_fp16);
            tensor<string, []> var_7984_equation_0 = const()[name = tensor<string, []>("op_7984_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7984_cast = einsum(equation = var_7984_equation_0, values = (var_7826_cast, var_7743_cast));
            tensor<fp16, []> var_7985_to_fp16 = const()[name = tensor<string, []>("op_7985_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_533_cast = mul(x = var_7984_cast, y = var_7985_to_fp16);
            tensor<string, []> var_7988_equation_0 = const()[name = tensor<string, []>("op_7988_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7988_cast = einsum(equation = var_7988_equation_0, values = (var_7830_cast, var_7747_cast));
            tensor<fp16, []> var_7989_to_fp16 = const()[name = tensor<string, []>("op_7989_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_535_cast = mul(x = var_7988_cast, y = var_7989_to_fp16);
            tensor<string, []> var_7992_equation_0 = const()[name = tensor<string, []>("op_7992_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7992_cast = einsum(equation = var_7992_equation_0, values = (var_7834_cast, var_7751_cast));
            tensor<fp16, []> var_7993_to_fp16 = const()[name = tensor<string, []>("op_7993_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_537_cast = mul(x = var_7992_cast, y = var_7993_to_fp16);
            tensor<string, []> var_7996_equation_0 = const()[name = tensor<string, []>("op_7996_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7996_cast = einsum(equation = var_7996_equation_0, values = (var_7838_cast, var_7755_cast));
            tensor<fp16, []> var_7997_to_fp16 = const()[name = tensor<string, []>("op_7997_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_539_cast = mul(x = var_7996_cast, y = var_7997_to_fp16);
            tensor<string, []> var_8000_equation_0 = const()[name = tensor<string, []>("op_8000_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8000_cast = einsum(equation = var_8000_equation_0, values = (var_7842_cast, var_7759_cast));
            tensor<fp16, []> var_8001_to_fp16 = const()[name = tensor<string, []>("op_8001_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_541_cast = mul(x = var_8000_cast, y = var_8001_to_fp16);
            tensor<string, []> var_8004_equation_0 = const()[name = tensor<string, []>("op_8004_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8004_cast = einsum(equation = var_8004_equation_0, values = (var_7846_cast, var_7763_cast));
            tensor<fp16, []> var_8005_to_fp16 = const()[name = tensor<string, []>("op_8005_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_543_cast = mul(x = var_8004_cast, y = var_8005_to_fp16);
            tensor<string, []> var_8008_equation_0 = const()[name = tensor<string, []>("op_8008_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8008_cast = einsum(equation = var_8008_equation_0, values = (var_7850_cast, var_7767_cast));
            tensor<fp16, []> var_8009_to_fp16 = const()[name = tensor<string, []>("op_8009_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_545_cast = mul(x = var_8008_cast, y = var_8009_to_fp16);
            tensor<string, []> var_8012_equation_0 = const()[name = tensor<string, []>("op_8012_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8012_cast = einsum(equation = var_8012_equation_0, values = (var_7854_cast, var_7771_cast));
            tensor<fp16, []> var_8013_to_fp16 = const()[name = tensor<string, []>("op_8013_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_547_cast = mul(x = var_8012_cast, y = var_8013_to_fp16);
            tensor<string, []> var_8016_equation_0 = const()[name = tensor<string, []>("op_8016_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8016_cast = einsum(equation = var_8016_equation_0, values = (var_7858_cast, var_7775_cast));
            tensor<fp16, []> var_8017_to_fp16 = const()[name = tensor<string, []>("op_8017_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_549_cast = mul(x = var_8016_cast, y = var_8017_to_fp16);
            tensor<string, []> var_8020_equation_0 = const()[name = tensor<string, []>("op_8020_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8020_cast = einsum(equation = var_8020_equation_0, values = (var_7862_cast, var_7779_cast));
            tensor<fp16, []> var_8021_to_fp16 = const()[name = tensor<string, []>("op_8021_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_551_cast = mul(x = var_8020_cast, y = var_8021_to_fp16);
            tensor<string, []> var_8024_equation_0 = const()[name = tensor<string, []>("op_8024_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8024_cast = einsum(equation = var_8024_equation_0, values = (var_7866_cast, var_7783_cast));
            tensor<fp16, []> var_8025_to_fp16 = const()[name = tensor<string, []>("op_8025_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_553_cast = mul(x = var_8024_cast, y = var_8025_to_fp16);
            tensor<string, []> var_8028_equation_0 = const()[name = tensor<string, []>("op_8028_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8028_cast = einsum(equation = var_8028_equation_0, values = (var_7870_cast, var_7787_cast));
            tensor<fp16, []> var_8029_to_fp16 = const()[name = tensor<string, []>("op_8029_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_555_cast = mul(x = var_8028_cast, y = var_8029_to_fp16);
            tensor<string, []> var_8032_equation_0 = const()[name = tensor<string, []>("op_8032_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8032_cast = einsum(equation = var_8032_equation_0, values = (var_7874_cast, var_7791_cast));
            tensor<fp16, []> var_8033_to_fp16 = const()[name = tensor<string, []>("op_8033_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_557_cast = mul(x = var_8032_cast, y = var_8033_to_fp16);
            tensor<string, []> var_8036_equation_0 = const()[name = tensor<string, []>("op_8036_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8036_cast = einsum(equation = var_8036_equation_0, values = (var_7878_cast, var_7795_cast));
            tensor<fp16, []> var_8037_to_fp16 = const()[name = tensor<string, []>("op_8037_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_559_cast = mul(x = var_8036_cast, y = var_8037_to_fp16);
            tensor<fp16, [2, 256, 1, 256]> var_8039_cast = softmax(axis = var_5599, x = aw_521_cast);
            tensor<fp16, [2, 256, 1, 256]> var_8040_cast = softmax(axis = var_5599, x = aw_523_cast);
            tensor<fp16, [2, 256, 1, 256]> var_8041_cast = softmax(axis = var_5599, x = aw_525_cast);
            tensor<fp16, [2, 256, 1, 256]> var_8042_cast = softmax(axis = var_5599, x = aw_527_cast);
            tensor<fp16, [2, 256, 1, 256]> var_8043_cast = softmax(axis = var_5599, x = aw_529_cast);
            tensor<fp16, [2, 256, 1, 256]> var_8044_cast = softmax(axis = var_5599, x = aw_531_cast);
            tensor<fp16, [2, 256, 1, 256]> var_8045_cast = softmax(axis = var_5599, x = aw_533_cast);
            tensor<fp16, [2, 256, 1, 256]> var_8046_cast = softmax(axis = var_5599, x = aw_535_cast);
            tensor<fp16, [2, 256, 1, 256]> var_8047_cast = softmax(axis = var_5599, x = aw_537_cast);
            tensor<fp16, [2, 256, 1, 256]> var_8048_cast = softmax(axis = var_5599, x = aw_539_cast);
            tensor<fp16, [2, 256, 1, 256]> var_8049_cast = softmax(axis = var_5599, x = aw_541_cast);
            tensor<fp16, [2, 256, 1, 256]> var_8050_cast = softmax(axis = var_5599, x = aw_543_cast);
            tensor<fp16, [2, 256, 1, 256]> var_8051_cast = softmax(axis = var_5599, x = aw_545_cast);
            tensor<fp16, [2, 256, 1, 256]> var_8052_cast = softmax(axis = var_5599, x = aw_547_cast);
            tensor<fp16, [2, 256, 1, 256]> var_8053_cast = softmax(axis = var_5599, x = aw_549_cast);
            tensor<fp16, [2, 256, 1, 256]> var_8054_cast = softmax(axis = var_5599, x = aw_551_cast);
            tensor<fp16, [2, 256, 1, 256]> var_8055_cast = softmax(axis = var_5599, x = aw_553_cast);
            tensor<fp16, [2, 256, 1, 256]> var_8056_cast = softmax(axis = var_5599, x = aw_555_cast);
            tensor<fp16, [2, 256, 1, 256]> var_8057_cast = softmax(axis = var_5599, x = aw_557_cast);
            tensor<fp16, [2, 256, 1, 256]> var_8058_cast = softmax(axis = var_5599, x = aw_559_cast);
            tensor<string, []> var_8060_equation_0 = const()[name = tensor<string, []>("op_8060_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8060_cast = einsum(equation = var_8060_equation_0, values = (var_7880_cast, var_8039_cast));
            tensor<string, []> var_8062_equation_0 = const()[name = tensor<string, []>("op_8062_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8062_cast = einsum(equation = var_8062_equation_0, values = (var_7884_cast, var_8040_cast));
            tensor<string, []> var_8064_equation_0 = const()[name = tensor<string, []>("op_8064_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8064_cast = einsum(equation = var_8064_equation_0, values = (var_7888_cast, var_8041_cast));
            tensor<string, []> var_8066_equation_0 = const()[name = tensor<string, []>("op_8066_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8066_cast = einsum(equation = var_8066_equation_0, values = (var_7892_cast, var_8042_cast));
            tensor<string, []> var_8068_equation_0 = const()[name = tensor<string, []>("op_8068_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8068_cast = einsum(equation = var_8068_equation_0, values = (var_7896_cast, var_8043_cast));
            tensor<string, []> var_8070_equation_0 = const()[name = tensor<string, []>("op_8070_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8070_cast = einsum(equation = var_8070_equation_0, values = (var_7900_cast, var_8044_cast));
            tensor<string, []> var_8072_equation_0 = const()[name = tensor<string, []>("op_8072_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8072_cast = einsum(equation = var_8072_equation_0, values = (var_7904_cast, var_8045_cast));
            tensor<string, []> var_8074_equation_0 = const()[name = tensor<string, []>("op_8074_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8074_cast = einsum(equation = var_8074_equation_0, values = (var_7908_cast, var_8046_cast));
            tensor<string, []> var_8076_equation_0 = const()[name = tensor<string, []>("op_8076_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8076_cast = einsum(equation = var_8076_equation_0, values = (var_7912_cast, var_8047_cast));
            tensor<string, []> var_8078_equation_0 = const()[name = tensor<string, []>("op_8078_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8078_cast = einsum(equation = var_8078_equation_0, values = (var_7916_cast, var_8048_cast));
            tensor<string, []> var_8080_equation_0 = const()[name = tensor<string, []>("op_8080_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8080_cast = einsum(equation = var_8080_equation_0, values = (var_7920_cast, var_8049_cast));
            tensor<string, []> var_8082_equation_0 = const()[name = tensor<string, []>("op_8082_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8082_cast = einsum(equation = var_8082_equation_0, values = (var_7924_cast, var_8050_cast));
            tensor<string, []> var_8084_equation_0 = const()[name = tensor<string, []>("op_8084_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8084_cast = einsum(equation = var_8084_equation_0, values = (var_7928_cast, var_8051_cast));
            tensor<string, []> var_8086_equation_0 = const()[name = tensor<string, []>("op_8086_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8086_cast = einsum(equation = var_8086_equation_0, values = (var_7932_cast, var_8052_cast));
            tensor<string, []> var_8088_equation_0 = const()[name = tensor<string, []>("op_8088_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8088_cast = einsum(equation = var_8088_equation_0, values = (var_7936_cast, var_8053_cast));
            tensor<string, []> var_8090_equation_0 = const()[name = tensor<string, []>("op_8090_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8090_cast = einsum(equation = var_8090_equation_0, values = (var_7940_cast, var_8054_cast));
            tensor<string, []> var_8092_equation_0 = const()[name = tensor<string, []>("op_8092_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8092_cast = einsum(equation = var_8092_equation_0, values = (var_7944_cast, var_8055_cast));
            tensor<string, []> var_8094_equation_0 = const()[name = tensor<string, []>("op_8094_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8094_cast = einsum(equation = var_8094_equation_0, values = (var_7948_cast, var_8056_cast));
            tensor<string, []> var_8096_equation_0 = const()[name = tensor<string, []>("op_8096_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8096_cast = einsum(equation = var_8096_equation_0, values = (var_7952_cast, var_8057_cast));
            tensor<string, []> var_8098_equation_0 = const()[name = tensor<string, []>("op_8098_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8098_cast = einsum(equation = var_8098_equation_0, values = (var_7956_cast, var_8058_cast));
            tensor<bool, []> input_353_interleave_0 = const()[name = tensor<string, []>("input_353_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 256]> input_353_cast = concat(axis = var_5599, interleave = input_353_interleave_0, values = (var_8060_cast, var_8062_cast, var_8064_cast, var_8066_cast, var_8068_cast, var_8070_cast, var_8072_cast, var_8074_cast, var_8076_cast, var_8078_cast, var_8080_cast, var_8082_cast, var_8084_cast, var_8086_cast, var_8088_cast, var_8090_cast, var_8092_cast, var_8094_cast, var_8096_cast, var_8098_cast));
            tensor<int32, [2]> var_8104 = const()[name = tensor<string, []>("op_8104"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8106 = const()[name = tensor<string, []>("op_8106"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_8108_pad_type_0 = const()[name = tensor<string, []>("op_8108_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_8108_pad_0 = const()[name = tensor<string, []>("op_8108_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1461428736)))];
            tensor<fp16, [1280]> up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1464705600)))];
            tensor<fp16, [2, 1280, 1, 256]> var_8108_cast = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_8106, groups = var_5599, pad = var_8108_pad_0, pad_type = var_8108_pad_type_0, strides = var_8104, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_353_cast);
            tensor<fp16, [2, 1280, 1, 256]> inputs_57_cast = add(x = var_8108_cast, y = inputs_55_cast);
            tensor<int32, [1]> var_8112 = const()[name = tensor<string, []>("op_8112"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_57_cast = reduce_mean(axes = var_8112, keep_dims = var_5594, x = inputs_57_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_57_cast = sub(x = inputs_57_cast, y = channels_mean_57_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_sq_57_cast = mul(x = zero_mean_57_cast, y = zero_mean_57_cast);
            tensor<int32, [1]> var_8116 = const()[name = tensor<string, []>("op_8116"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_8117_cast = reduce_mean(axes = var_8116, keep_dims = var_5594, x = zero_mean_sq_57_cast);
            tensor<fp16, []> var_8118_to_fp16 = const()[name = tensor<string, []>("op_8118_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_8119_cast = add(x = var_8117_cast, y = var_8118_to_fp16);
            tensor<fp16, []> denom_57_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_57_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 256]> denom_57_cast = rsqrt(epsilon = denom_57_epsilon_0_to_fp16, x = var_8119_cast);
            tensor<fp16, [2, 1280, 1, 256]> out_57_cast = mul(x = zero_mean_57_cast, y = denom_57_cast);
            tensor<fp16, [1, 1280, 1, 1]> var_8123_to_fp16 = const()[name = tensor<string, []>("op_8123_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1464708224)))];
            tensor<fp16, [2, 1280, 1, 256]> var_8124_cast = add(x = out_57_cast, y = var_8123_to_fp16);
            tensor<fp16, [1, 1280, 1, 1]> var_8126_to_fp16 = const()[name = tensor<string, []>("op_8126_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1464710848)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_209_cast = mul(x = var_8124_cast, y = var_8126_to_fp16);
            tensor<int32, [2]> var_8133 = const()[name = tensor<string, []>("op_8133"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8135 = const()[name = tensor<string, []>("op_8135"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_39_pad_type_0 = const()[name = tensor<string, []>("q_39_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_39_pad_0 = const()[name = tensor<string, []>("q_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1464713472)))];
            tensor<fp16, [2, 1280, 1, 256]> q_39_cast = conv(dilations = var_8135, groups = var_5599, pad = q_39_pad_0, pad_type = q_39_pad_type_0, strides = var_8133, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_209_cast);
            tensor<int32, [2]> var_8139 = const()[name = tensor<string, []>("op_8139"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8141 = const()[name = tensor<string, []>("op_8141"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_77_pad_type_0 = const()[name = tensor<string, []>("k_77_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_77_pad_0 = const()[name = tensor<string, []>("k_77_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1024, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [1280, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1467990336)))];
            tensor<fp16, [2, 1280, 1, 77]> k_77_cast = conv(dilations = var_8141, groups = var_5599, pad = k_77_pad_0, pad_type = k_77_pad_type_0, strides = var_8139, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [2]> var_8145 = const()[name = tensor<string, []>("op_8145"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8147 = const()[name = tensor<string, []>("op_8147"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_39_pad_type_0 = const()[name = tensor<string, []>("v_39_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_39_pad_0 = const()[name = tensor<string, []>("v_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1024, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [1280, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1470611840)))];
            tensor<fp16, [2, 1280, 1, 77]> v_39_cast = conv(dilations = var_8147, groups = var_5599, pad = v_39_pad_0, pad_type = v_39_pad_type_0, strides = var_8145, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [4]> var_8151_begin_0 = const()[name = tensor<string, []>("op_8151_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8151_end_0 = const()[name = tensor<string, []>("op_8151_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_8151_end_mask_0 = const()[name = tensor<string, []>("op_8151_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8151_cast = slice_by_index(begin = var_8151_begin_0, end = var_8151_end_0, end_mask = var_8151_end_mask_0, x = q_39_cast);
            tensor<int32, [4]> var_8155_begin_0 = const()[name = tensor<string, []>("op_8155_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_8155_end_0 = const()[name = tensor<string, []>("op_8155_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_8155_end_mask_0 = const()[name = tensor<string, []>("op_8155_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8155_cast = slice_by_index(begin = var_8155_begin_0, end = var_8155_end_0, end_mask = var_8155_end_mask_0, x = q_39_cast);
            tensor<int32, [4]> var_8159_begin_0 = const()[name = tensor<string, []>("op_8159_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_8159_end_0 = const()[name = tensor<string, []>("op_8159_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_8159_end_mask_0 = const()[name = tensor<string, []>("op_8159_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8159_cast = slice_by_index(begin = var_8159_begin_0, end = var_8159_end_0, end_mask = var_8159_end_mask_0, x = q_39_cast);
            tensor<int32, [4]> var_8163_begin_0 = const()[name = tensor<string, []>("op_8163_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_8163_end_0 = const()[name = tensor<string, []>("op_8163_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_8163_end_mask_0 = const()[name = tensor<string, []>("op_8163_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8163_cast = slice_by_index(begin = var_8163_begin_0, end = var_8163_end_0, end_mask = var_8163_end_mask_0, x = q_39_cast);
            tensor<int32, [4]> var_8167_begin_0 = const()[name = tensor<string, []>("op_8167_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_8167_end_0 = const()[name = tensor<string, []>("op_8167_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_8167_end_mask_0 = const()[name = tensor<string, []>("op_8167_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8167_cast = slice_by_index(begin = var_8167_begin_0, end = var_8167_end_0, end_mask = var_8167_end_mask_0, x = q_39_cast);
            tensor<int32, [4]> var_8171_begin_0 = const()[name = tensor<string, []>("op_8171_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_8171_end_0 = const()[name = tensor<string, []>("op_8171_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_8171_end_mask_0 = const()[name = tensor<string, []>("op_8171_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8171_cast = slice_by_index(begin = var_8171_begin_0, end = var_8171_end_0, end_mask = var_8171_end_mask_0, x = q_39_cast);
            tensor<int32, [4]> var_8175_begin_0 = const()[name = tensor<string, []>("op_8175_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_8175_end_0 = const()[name = tensor<string, []>("op_8175_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_8175_end_mask_0 = const()[name = tensor<string, []>("op_8175_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8175_cast = slice_by_index(begin = var_8175_begin_0, end = var_8175_end_0, end_mask = var_8175_end_mask_0, x = q_39_cast);
            tensor<int32, [4]> var_8179_begin_0 = const()[name = tensor<string, []>("op_8179_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_8179_end_0 = const()[name = tensor<string, []>("op_8179_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_8179_end_mask_0 = const()[name = tensor<string, []>("op_8179_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8179_cast = slice_by_index(begin = var_8179_begin_0, end = var_8179_end_0, end_mask = var_8179_end_mask_0, x = q_39_cast);
            tensor<int32, [4]> var_8183_begin_0 = const()[name = tensor<string, []>("op_8183_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_8183_end_0 = const()[name = tensor<string, []>("op_8183_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_8183_end_mask_0 = const()[name = tensor<string, []>("op_8183_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8183_cast = slice_by_index(begin = var_8183_begin_0, end = var_8183_end_0, end_mask = var_8183_end_mask_0, x = q_39_cast);
            tensor<int32, [4]> var_8187_begin_0 = const()[name = tensor<string, []>("op_8187_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_8187_end_0 = const()[name = tensor<string, []>("op_8187_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_8187_end_mask_0 = const()[name = tensor<string, []>("op_8187_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8187_cast = slice_by_index(begin = var_8187_begin_0, end = var_8187_end_0, end_mask = var_8187_end_mask_0, x = q_39_cast);
            tensor<int32, [4]> var_8191_begin_0 = const()[name = tensor<string, []>("op_8191_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_8191_end_0 = const()[name = tensor<string, []>("op_8191_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_8191_end_mask_0 = const()[name = tensor<string, []>("op_8191_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8191_cast = slice_by_index(begin = var_8191_begin_0, end = var_8191_end_0, end_mask = var_8191_end_mask_0, x = q_39_cast);
            tensor<int32, [4]> var_8195_begin_0 = const()[name = tensor<string, []>("op_8195_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_8195_end_0 = const()[name = tensor<string, []>("op_8195_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_8195_end_mask_0 = const()[name = tensor<string, []>("op_8195_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8195_cast = slice_by_index(begin = var_8195_begin_0, end = var_8195_end_0, end_mask = var_8195_end_mask_0, x = q_39_cast);
            tensor<int32, [4]> var_8199_begin_0 = const()[name = tensor<string, []>("op_8199_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_8199_end_0 = const()[name = tensor<string, []>("op_8199_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_8199_end_mask_0 = const()[name = tensor<string, []>("op_8199_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8199_cast = slice_by_index(begin = var_8199_begin_0, end = var_8199_end_0, end_mask = var_8199_end_mask_0, x = q_39_cast);
            tensor<int32, [4]> var_8203_begin_0 = const()[name = tensor<string, []>("op_8203_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_8203_end_0 = const()[name = tensor<string, []>("op_8203_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_8203_end_mask_0 = const()[name = tensor<string, []>("op_8203_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8203_cast = slice_by_index(begin = var_8203_begin_0, end = var_8203_end_0, end_mask = var_8203_end_mask_0, x = q_39_cast);
            tensor<int32, [4]> var_8207_begin_0 = const()[name = tensor<string, []>("op_8207_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_8207_end_0 = const()[name = tensor<string, []>("op_8207_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_8207_end_mask_0 = const()[name = tensor<string, []>("op_8207_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8207_cast = slice_by_index(begin = var_8207_begin_0, end = var_8207_end_0, end_mask = var_8207_end_mask_0, x = q_39_cast);
            tensor<int32, [4]> var_8211_begin_0 = const()[name = tensor<string, []>("op_8211_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_8211_end_0 = const()[name = tensor<string, []>("op_8211_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_8211_end_mask_0 = const()[name = tensor<string, []>("op_8211_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8211_cast = slice_by_index(begin = var_8211_begin_0, end = var_8211_end_0, end_mask = var_8211_end_mask_0, x = q_39_cast);
            tensor<int32, [4]> var_8215_begin_0 = const()[name = tensor<string, []>("op_8215_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_8215_end_0 = const()[name = tensor<string, []>("op_8215_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_8215_end_mask_0 = const()[name = tensor<string, []>("op_8215_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8215_cast = slice_by_index(begin = var_8215_begin_0, end = var_8215_end_0, end_mask = var_8215_end_mask_0, x = q_39_cast);
            tensor<int32, [4]> var_8219_begin_0 = const()[name = tensor<string, []>("op_8219_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_8219_end_0 = const()[name = tensor<string, []>("op_8219_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_8219_end_mask_0 = const()[name = tensor<string, []>("op_8219_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8219_cast = slice_by_index(begin = var_8219_begin_0, end = var_8219_end_0, end_mask = var_8219_end_mask_0, x = q_39_cast);
            tensor<int32, [4]> var_8223_begin_0 = const()[name = tensor<string, []>("op_8223_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_8223_end_0 = const()[name = tensor<string, []>("op_8223_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_8223_end_mask_0 = const()[name = tensor<string, []>("op_8223_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8223_cast = slice_by_index(begin = var_8223_begin_0, end = var_8223_end_0, end_mask = var_8223_end_mask_0, x = q_39_cast);
            tensor<int32, [4]> var_8227_begin_0 = const()[name = tensor<string, []>("op_8227_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_8227_end_0 = const()[name = tensor<string, []>("op_8227_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_8227_end_mask_0 = const()[name = tensor<string, []>("op_8227_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8227_cast = slice_by_index(begin = var_8227_begin_0, end = var_8227_end_0, end_mask = var_8227_end_mask_0, x = q_39_cast);
            tensor<int32, [4]> k_79_perm_0 = const()[name = tensor<string, []>("k_79_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_8234_begin_0 = const()[name = tensor<string, []>("op_8234_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8234_end_0 = const()[name = tensor<string, []>("op_8234_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_8234_end_mask_0 = const()[name = tensor<string, []>("op_8234_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 1280]> transpose_12 = transpose(perm = k_79_perm_0, x = k_77_cast);
            tensor<fp16, [2, 77, 1, 64]> var_8234_cast = slice_by_index(begin = var_8234_begin_0, end = var_8234_end_0, end_mask = var_8234_end_mask_0, x = transpose_12);
            tensor<int32, [4]> var_8238_begin_0 = const()[name = tensor<string, []>("op_8238_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_8238_end_0 = const()[name = tensor<string, []>("op_8238_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_8238_end_mask_0 = const()[name = tensor<string, []>("op_8238_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8238_cast = slice_by_index(begin = var_8238_begin_0, end = var_8238_end_0, end_mask = var_8238_end_mask_0, x = transpose_12);
            tensor<int32, [4]> var_8242_begin_0 = const()[name = tensor<string, []>("op_8242_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_8242_end_0 = const()[name = tensor<string, []>("op_8242_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_8242_end_mask_0 = const()[name = tensor<string, []>("op_8242_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8242_cast = slice_by_index(begin = var_8242_begin_0, end = var_8242_end_0, end_mask = var_8242_end_mask_0, x = transpose_12);
            tensor<int32, [4]> var_8246_begin_0 = const()[name = tensor<string, []>("op_8246_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_8246_end_0 = const()[name = tensor<string, []>("op_8246_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_8246_end_mask_0 = const()[name = tensor<string, []>("op_8246_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8246_cast = slice_by_index(begin = var_8246_begin_0, end = var_8246_end_0, end_mask = var_8246_end_mask_0, x = transpose_12);
            tensor<int32, [4]> var_8250_begin_0 = const()[name = tensor<string, []>("op_8250_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_8250_end_0 = const()[name = tensor<string, []>("op_8250_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_8250_end_mask_0 = const()[name = tensor<string, []>("op_8250_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8250_cast = slice_by_index(begin = var_8250_begin_0, end = var_8250_end_0, end_mask = var_8250_end_mask_0, x = transpose_12);
            tensor<int32, [4]> var_8254_begin_0 = const()[name = tensor<string, []>("op_8254_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_8254_end_0 = const()[name = tensor<string, []>("op_8254_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_8254_end_mask_0 = const()[name = tensor<string, []>("op_8254_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8254_cast = slice_by_index(begin = var_8254_begin_0, end = var_8254_end_0, end_mask = var_8254_end_mask_0, x = transpose_12);
            tensor<int32, [4]> var_8258_begin_0 = const()[name = tensor<string, []>("op_8258_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_8258_end_0 = const()[name = tensor<string, []>("op_8258_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_8258_end_mask_0 = const()[name = tensor<string, []>("op_8258_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8258_cast = slice_by_index(begin = var_8258_begin_0, end = var_8258_end_0, end_mask = var_8258_end_mask_0, x = transpose_12);
            tensor<int32, [4]> var_8262_begin_0 = const()[name = tensor<string, []>("op_8262_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_8262_end_0 = const()[name = tensor<string, []>("op_8262_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_8262_end_mask_0 = const()[name = tensor<string, []>("op_8262_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8262_cast = slice_by_index(begin = var_8262_begin_0, end = var_8262_end_0, end_mask = var_8262_end_mask_0, x = transpose_12);
            tensor<int32, [4]> var_8266_begin_0 = const()[name = tensor<string, []>("op_8266_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_8266_end_0 = const()[name = tensor<string, []>("op_8266_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_8266_end_mask_0 = const()[name = tensor<string, []>("op_8266_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8266_cast = slice_by_index(begin = var_8266_begin_0, end = var_8266_end_0, end_mask = var_8266_end_mask_0, x = transpose_12);
            tensor<int32, [4]> var_8270_begin_0 = const()[name = tensor<string, []>("op_8270_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_8270_end_0 = const()[name = tensor<string, []>("op_8270_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_8270_end_mask_0 = const()[name = tensor<string, []>("op_8270_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8270_cast = slice_by_index(begin = var_8270_begin_0, end = var_8270_end_0, end_mask = var_8270_end_mask_0, x = transpose_12);
            tensor<int32, [4]> var_8274_begin_0 = const()[name = tensor<string, []>("op_8274_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_8274_end_0 = const()[name = tensor<string, []>("op_8274_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_8274_end_mask_0 = const()[name = tensor<string, []>("op_8274_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8274_cast = slice_by_index(begin = var_8274_begin_0, end = var_8274_end_0, end_mask = var_8274_end_mask_0, x = transpose_12);
            tensor<int32, [4]> var_8278_begin_0 = const()[name = tensor<string, []>("op_8278_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_8278_end_0 = const()[name = tensor<string, []>("op_8278_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_8278_end_mask_0 = const()[name = tensor<string, []>("op_8278_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8278_cast = slice_by_index(begin = var_8278_begin_0, end = var_8278_end_0, end_mask = var_8278_end_mask_0, x = transpose_12);
            tensor<int32, [4]> var_8282_begin_0 = const()[name = tensor<string, []>("op_8282_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_8282_end_0 = const()[name = tensor<string, []>("op_8282_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_8282_end_mask_0 = const()[name = tensor<string, []>("op_8282_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8282_cast = slice_by_index(begin = var_8282_begin_0, end = var_8282_end_0, end_mask = var_8282_end_mask_0, x = transpose_12);
            tensor<int32, [4]> var_8286_begin_0 = const()[name = tensor<string, []>("op_8286_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_8286_end_0 = const()[name = tensor<string, []>("op_8286_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_8286_end_mask_0 = const()[name = tensor<string, []>("op_8286_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8286_cast = slice_by_index(begin = var_8286_begin_0, end = var_8286_end_0, end_mask = var_8286_end_mask_0, x = transpose_12);
            tensor<int32, [4]> var_8290_begin_0 = const()[name = tensor<string, []>("op_8290_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_8290_end_0 = const()[name = tensor<string, []>("op_8290_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_8290_end_mask_0 = const()[name = tensor<string, []>("op_8290_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8290_cast = slice_by_index(begin = var_8290_begin_0, end = var_8290_end_0, end_mask = var_8290_end_mask_0, x = transpose_12);
            tensor<int32, [4]> var_8294_begin_0 = const()[name = tensor<string, []>("op_8294_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_8294_end_0 = const()[name = tensor<string, []>("op_8294_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_8294_end_mask_0 = const()[name = tensor<string, []>("op_8294_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8294_cast = slice_by_index(begin = var_8294_begin_0, end = var_8294_end_0, end_mask = var_8294_end_mask_0, x = transpose_12);
            tensor<int32, [4]> var_8298_begin_0 = const()[name = tensor<string, []>("op_8298_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_8298_end_0 = const()[name = tensor<string, []>("op_8298_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_8298_end_mask_0 = const()[name = tensor<string, []>("op_8298_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8298_cast = slice_by_index(begin = var_8298_begin_0, end = var_8298_end_0, end_mask = var_8298_end_mask_0, x = transpose_12);
            tensor<int32, [4]> var_8302_begin_0 = const()[name = tensor<string, []>("op_8302_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_8302_end_0 = const()[name = tensor<string, []>("op_8302_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_8302_end_mask_0 = const()[name = tensor<string, []>("op_8302_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8302_cast = slice_by_index(begin = var_8302_begin_0, end = var_8302_end_0, end_mask = var_8302_end_mask_0, x = transpose_12);
            tensor<int32, [4]> var_8306_begin_0 = const()[name = tensor<string, []>("op_8306_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_8306_end_0 = const()[name = tensor<string, []>("op_8306_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_8306_end_mask_0 = const()[name = tensor<string, []>("op_8306_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8306_cast = slice_by_index(begin = var_8306_begin_0, end = var_8306_end_0, end_mask = var_8306_end_mask_0, x = transpose_12);
            tensor<int32, [4]> var_8310_begin_0 = const()[name = tensor<string, []>("op_8310_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_8310_end_0 = const()[name = tensor<string, []>("op_8310_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_8310_end_mask_0 = const()[name = tensor<string, []>("op_8310_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8310_cast = slice_by_index(begin = var_8310_begin_0, end = var_8310_end_0, end_mask = var_8310_end_mask_0, x = transpose_12);
            tensor<int32, [4]> var_8312_begin_0 = const()[name = tensor<string, []>("op_8312_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8312_end_0 = const()[name = tensor<string, []>("op_8312_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_8312_end_mask_0 = const()[name = tensor<string, []>("op_8312_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8312_cast = slice_by_index(begin = var_8312_begin_0, end = var_8312_end_0, end_mask = var_8312_end_mask_0, x = v_39_cast);
            tensor<int32, [4]> var_8316_begin_0 = const()[name = tensor<string, []>("op_8316_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_8316_end_0 = const()[name = tensor<string, []>("op_8316_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_8316_end_mask_0 = const()[name = tensor<string, []>("op_8316_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8316_cast = slice_by_index(begin = var_8316_begin_0, end = var_8316_end_0, end_mask = var_8316_end_mask_0, x = v_39_cast);
            tensor<int32, [4]> var_8320_begin_0 = const()[name = tensor<string, []>("op_8320_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_8320_end_0 = const()[name = tensor<string, []>("op_8320_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_8320_end_mask_0 = const()[name = tensor<string, []>("op_8320_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8320_cast = slice_by_index(begin = var_8320_begin_0, end = var_8320_end_0, end_mask = var_8320_end_mask_0, x = v_39_cast);
            tensor<int32, [4]> var_8324_begin_0 = const()[name = tensor<string, []>("op_8324_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_8324_end_0 = const()[name = tensor<string, []>("op_8324_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_8324_end_mask_0 = const()[name = tensor<string, []>("op_8324_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8324_cast = slice_by_index(begin = var_8324_begin_0, end = var_8324_end_0, end_mask = var_8324_end_mask_0, x = v_39_cast);
            tensor<int32, [4]> var_8328_begin_0 = const()[name = tensor<string, []>("op_8328_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_8328_end_0 = const()[name = tensor<string, []>("op_8328_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_8328_end_mask_0 = const()[name = tensor<string, []>("op_8328_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8328_cast = slice_by_index(begin = var_8328_begin_0, end = var_8328_end_0, end_mask = var_8328_end_mask_0, x = v_39_cast);
            tensor<int32, [4]> var_8332_begin_0 = const()[name = tensor<string, []>("op_8332_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_8332_end_0 = const()[name = tensor<string, []>("op_8332_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_8332_end_mask_0 = const()[name = tensor<string, []>("op_8332_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8332_cast = slice_by_index(begin = var_8332_begin_0, end = var_8332_end_0, end_mask = var_8332_end_mask_0, x = v_39_cast);
            tensor<int32, [4]> var_8336_begin_0 = const()[name = tensor<string, []>("op_8336_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_8336_end_0 = const()[name = tensor<string, []>("op_8336_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_8336_end_mask_0 = const()[name = tensor<string, []>("op_8336_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8336_cast = slice_by_index(begin = var_8336_begin_0, end = var_8336_end_0, end_mask = var_8336_end_mask_0, x = v_39_cast);
            tensor<int32, [4]> var_8340_begin_0 = const()[name = tensor<string, []>("op_8340_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_8340_end_0 = const()[name = tensor<string, []>("op_8340_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_8340_end_mask_0 = const()[name = tensor<string, []>("op_8340_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8340_cast = slice_by_index(begin = var_8340_begin_0, end = var_8340_end_0, end_mask = var_8340_end_mask_0, x = v_39_cast);
            tensor<int32, [4]> var_8344_begin_0 = const()[name = tensor<string, []>("op_8344_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_8344_end_0 = const()[name = tensor<string, []>("op_8344_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_8344_end_mask_0 = const()[name = tensor<string, []>("op_8344_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8344_cast = slice_by_index(begin = var_8344_begin_0, end = var_8344_end_0, end_mask = var_8344_end_mask_0, x = v_39_cast);
            tensor<int32, [4]> var_8348_begin_0 = const()[name = tensor<string, []>("op_8348_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_8348_end_0 = const()[name = tensor<string, []>("op_8348_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_8348_end_mask_0 = const()[name = tensor<string, []>("op_8348_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8348_cast = slice_by_index(begin = var_8348_begin_0, end = var_8348_end_0, end_mask = var_8348_end_mask_0, x = v_39_cast);
            tensor<int32, [4]> var_8352_begin_0 = const()[name = tensor<string, []>("op_8352_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_8352_end_0 = const()[name = tensor<string, []>("op_8352_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_8352_end_mask_0 = const()[name = tensor<string, []>("op_8352_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8352_cast = slice_by_index(begin = var_8352_begin_0, end = var_8352_end_0, end_mask = var_8352_end_mask_0, x = v_39_cast);
            tensor<int32, [4]> var_8356_begin_0 = const()[name = tensor<string, []>("op_8356_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_8356_end_0 = const()[name = tensor<string, []>("op_8356_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_8356_end_mask_0 = const()[name = tensor<string, []>("op_8356_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8356_cast = slice_by_index(begin = var_8356_begin_0, end = var_8356_end_0, end_mask = var_8356_end_mask_0, x = v_39_cast);
            tensor<int32, [4]> var_8360_begin_0 = const()[name = tensor<string, []>("op_8360_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_8360_end_0 = const()[name = tensor<string, []>("op_8360_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_8360_end_mask_0 = const()[name = tensor<string, []>("op_8360_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8360_cast = slice_by_index(begin = var_8360_begin_0, end = var_8360_end_0, end_mask = var_8360_end_mask_0, x = v_39_cast);
            tensor<int32, [4]> var_8364_begin_0 = const()[name = tensor<string, []>("op_8364_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_8364_end_0 = const()[name = tensor<string, []>("op_8364_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_8364_end_mask_0 = const()[name = tensor<string, []>("op_8364_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8364_cast = slice_by_index(begin = var_8364_begin_0, end = var_8364_end_0, end_mask = var_8364_end_mask_0, x = v_39_cast);
            tensor<int32, [4]> var_8368_begin_0 = const()[name = tensor<string, []>("op_8368_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_8368_end_0 = const()[name = tensor<string, []>("op_8368_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_8368_end_mask_0 = const()[name = tensor<string, []>("op_8368_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8368_cast = slice_by_index(begin = var_8368_begin_0, end = var_8368_end_0, end_mask = var_8368_end_mask_0, x = v_39_cast);
            tensor<int32, [4]> var_8372_begin_0 = const()[name = tensor<string, []>("op_8372_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_8372_end_0 = const()[name = tensor<string, []>("op_8372_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_8372_end_mask_0 = const()[name = tensor<string, []>("op_8372_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8372_cast = slice_by_index(begin = var_8372_begin_0, end = var_8372_end_0, end_mask = var_8372_end_mask_0, x = v_39_cast);
            tensor<int32, [4]> var_8376_begin_0 = const()[name = tensor<string, []>("op_8376_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_8376_end_0 = const()[name = tensor<string, []>("op_8376_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_8376_end_mask_0 = const()[name = tensor<string, []>("op_8376_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8376_cast = slice_by_index(begin = var_8376_begin_0, end = var_8376_end_0, end_mask = var_8376_end_mask_0, x = v_39_cast);
            tensor<int32, [4]> var_8380_begin_0 = const()[name = tensor<string, []>("op_8380_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_8380_end_0 = const()[name = tensor<string, []>("op_8380_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_8380_end_mask_0 = const()[name = tensor<string, []>("op_8380_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8380_cast = slice_by_index(begin = var_8380_begin_0, end = var_8380_end_0, end_mask = var_8380_end_mask_0, x = v_39_cast);
            tensor<int32, [4]> var_8384_begin_0 = const()[name = tensor<string, []>("op_8384_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_8384_end_0 = const()[name = tensor<string, []>("op_8384_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_8384_end_mask_0 = const()[name = tensor<string, []>("op_8384_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8384_cast = slice_by_index(begin = var_8384_begin_0, end = var_8384_end_0, end_mask = var_8384_end_mask_0, x = v_39_cast);
            tensor<int32, [4]> var_8388_begin_0 = const()[name = tensor<string, []>("op_8388_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_8388_end_0 = const()[name = tensor<string, []>("op_8388_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_8388_end_mask_0 = const()[name = tensor<string, []>("op_8388_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8388_cast = slice_by_index(begin = var_8388_begin_0, end = var_8388_end_0, end_mask = var_8388_end_mask_0, x = v_39_cast);
            tensor<string, []> var_8392_equation_0 = const()[name = tensor<string, []>("op_8392_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8392_cast = einsum(equation = var_8392_equation_0, values = (var_8234_cast, var_8151_cast));
            tensor<fp16, []> var_8393_to_fp16 = const()[name = tensor<string, []>("op_8393_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_561_cast = mul(x = var_8392_cast, y = var_8393_to_fp16);
            tensor<string, []> var_8396_equation_0 = const()[name = tensor<string, []>("op_8396_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8396_cast = einsum(equation = var_8396_equation_0, values = (var_8238_cast, var_8155_cast));
            tensor<fp16, []> var_8397_to_fp16 = const()[name = tensor<string, []>("op_8397_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_563_cast = mul(x = var_8396_cast, y = var_8397_to_fp16);
            tensor<string, []> var_8400_equation_0 = const()[name = tensor<string, []>("op_8400_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8400_cast = einsum(equation = var_8400_equation_0, values = (var_8242_cast, var_8159_cast));
            tensor<fp16, []> var_8401_to_fp16 = const()[name = tensor<string, []>("op_8401_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_565_cast = mul(x = var_8400_cast, y = var_8401_to_fp16);
            tensor<string, []> var_8404_equation_0 = const()[name = tensor<string, []>("op_8404_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8404_cast = einsum(equation = var_8404_equation_0, values = (var_8246_cast, var_8163_cast));
            tensor<fp16, []> var_8405_to_fp16 = const()[name = tensor<string, []>("op_8405_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_567_cast = mul(x = var_8404_cast, y = var_8405_to_fp16);
            tensor<string, []> var_8408_equation_0 = const()[name = tensor<string, []>("op_8408_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8408_cast = einsum(equation = var_8408_equation_0, values = (var_8250_cast, var_8167_cast));
            tensor<fp16, []> var_8409_to_fp16 = const()[name = tensor<string, []>("op_8409_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_569_cast = mul(x = var_8408_cast, y = var_8409_to_fp16);
            tensor<string, []> var_8412_equation_0 = const()[name = tensor<string, []>("op_8412_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8412_cast = einsum(equation = var_8412_equation_0, values = (var_8254_cast, var_8171_cast));
            tensor<fp16, []> var_8413_to_fp16 = const()[name = tensor<string, []>("op_8413_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_571_cast = mul(x = var_8412_cast, y = var_8413_to_fp16);
            tensor<string, []> var_8416_equation_0 = const()[name = tensor<string, []>("op_8416_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8416_cast = einsum(equation = var_8416_equation_0, values = (var_8258_cast, var_8175_cast));
            tensor<fp16, []> var_8417_to_fp16 = const()[name = tensor<string, []>("op_8417_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_573_cast = mul(x = var_8416_cast, y = var_8417_to_fp16);
            tensor<string, []> var_8420_equation_0 = const()[name = tensor<string, []>("op_8420_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8420_cast = einsum(equation = var_8420_equation_0, values = (var_8262_cast, var_8179_cast));
            tensor<fp16, []> var_8421_to_fp16 = const()[name = tensor<string, []>("op_8421_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_575_cast = mul(x = var_8420_cast, y = var_8421_to_fp16);
            tensor<string, []> var_8424_equation_0 = const()[name = tensor<string, []>("op_8424_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8424_cast = einsum(equation = var_8424_equation_0, values = (var_8266_cast, var_8183_cast));
            tensor<fp16, []> var_8425_to_fp16 = const()[name = tensor<string, []>("op_8425_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_577_cast = mul(x = var_8424_cast, y = var_8425_to_fp16);
            tensor<string, []> var_8428_equation_0 = const()[name = tensor<string, []>("op_8428_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8428_cast = einsum(equation = var_8428_equation_0, values = (var_8270_cast, var_8187_cast));
            tensor<fp16, []> var_8429_to_fp16 = const()[name = tensor<string, []>("op_8429_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_579_cast = mul(x = var_8428_cast, y = var_8429_to_fp16);
            tensor<string, []> var_8432_equation_0 = const()[name = tensor<string, []>("op_8432_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8432_cast = einsum(equation = var_8432_equation_0, values = (var_8274_cast, var_8191_cast));
            tensor<fp16, []> var_8433_to_fp16 = const()[name = tensor<string, []>("op_8433_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_581_cast = mul(x = var_8432_cast, y = var_8433_to_fp16);
            tensor<string, []> var_8436_equation_0 = const()[name = tensor<string, []>("op_8436_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8436_cast = einsum(equation = var_8436_equation_0, values = (var_8278_cast, var_8195_cast));
            tensor<fp16, []> var_8437_to_fp16 = const()[name = tensor<string, []>("op_8437_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_583_cast = mul(x = var_8436_cast, y = var_8437_to_fp16);
            tensor<string, []> var_8440_equation_0 = const()[name = tensor<string, []>("op_8440_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8440_cast = einsum(equation = var_8440_equation_0, values = (var_8282_cast, var_8199_cast));
            tensor<fp16, []> var_8441_to_fp16 = const()[name = tensor<string, []>("op_8441_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_585_cast = mul(x = var_8440_cast, y = var_8441_to_fp16);
            tensor<string, []> var_8444_equation_0 = const()[name = tensor<string, []>("op_8444_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8444_cast = einsum(equation = var_8444_equation_0, values = (var_8286_cast, var_8203_cast));
            tensor<fp16, []> var_8445_to_fp16 = const()[name = tensor<string, []>("op_8445_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_587_cast = mul(x = var_8444_cast, y = var_8445_to_fp16);
            tensor<string, []> var_8448_equation_0 = const()[name = tensor<string, []>("op_8448_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8448_cast = einsum(equation = var_8448_equation_0, values = (var_8290_cast, var_8207_cast));
            tensor<fp16, []> var_8449_to_fp16 = const()[name = tensor<string, []>("op_8449_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_589_cast = mul(x = var_8448_cast, y = var_8449_to_fp16);
            tensor<string, []> var_8452_equation_0 = const()[name = tensor<string, []>("op_8452_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8452_cast = einsum(equation = var_8452_equation_0, values = (var_8294_cast, var_8211_cast));
            tensor<fp16, []> var_8453_to_fp16 = const()[name = tensor<string, []>("op_8453_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_591_cast = mul(x = var_8452_cast, y = var_8453_to_fp16);
            tensor<string, []> var_8456_equation_0 = const()[name = tensor<string, []>("op_8456_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8456_cast = einsum(equation = var_8456_equation_0, values = (var_8298_cast, var_8215_cast));
            tensor<fp16, []> var_8457_to_fp16 = const()[name = tensor<string, []>("op_8457_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_593_cast = mul(x = var_8456_cast, y = var_8457_to_fp16);
            tensor<string, []> var_8460_equation_0 = const()[name = tensor<string, []>("op_8460_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8460_cast = einsum(equation = var_8460_equation_0, values = (var_8302_cast, var_8219_cast));
            tensor<fp16, []> var_8461_to_fp16 = const()[name = tensor<string, []>("op_8461_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_595_cast = mul(x = var_8460_cast, y = var_8461_to_fp16);
            tensor<string, []> var_8464_equation_0 = const()[name = tensor<string, []>("op_8464_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8464_cast = einsum(equation = var_8464_equation_0, values = (var_8306_cast, var_8223_cast));
            tensor<fp16, []> var_8465_to_fp16 = const()[name = tensor<string, []>("op_8465_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_597_cast = mul(x = var_8464_cast, y = var_8465_to_fp16);
            tensor<string, []> var_8468_equation_0 = const()[name = tensor<string, []>("op_8468_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8468_cast = einsum(equation = var_8468_equation_0, values = (var_8310_cast, var_8227_cast));
            tensor<fp16, []> var_8469_to_fp16 = const()[name = tensor<string, []>("op_8469_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_599_cast = mul(x = var_8468_cast, y = var_8469_to_fp16);
            tensor<fp16, [2, 77, 1, 256]> var_8471_cast = softmax(axis = var_5599, x = aw_561_cast);
            tensor<fp16, [2, 77, 1, 256]> var_8472_cast = softmax(axis = var_5599, x = aw_563_cast);
            tensor<fp16, [2, 77, 1, 256]> var_8473_cast = softmax(axis = var_5599, x = aw_565_cast);
            tensor<fp16, [2, 77, 1, 256]> var_8474_cast = softmax(axis = var_5599, x = aw_567_cast);
            tensor<fp16, [2, 77, 1, 256]> var_8475_cast = softmax(axis = var_5599, x = aw_569_cast);
            tensor<fp16, [2, 77, 1, 256]> var_8476_cast = softmax(axis = var_5599, x = aw_571_cast);
            tensor<fp16, [2, 77, 1, 256]> var_8477_cast = softmax(axis = var_5599, x = aw_573_cast);
            tensor<fp16, [2, 77, 1, 256]> var_8478_cast = softmax(axis = var_5599, x = aw_575_cast);
            tensor<fp16, [2, 77, 1, 256]> var_8479_cast = softmax(axis = var_5599, x = aw_577_cast);
            tensor<fp16, [2, 77, 1, 256]> var_8480_cast = softmax(axis = var_5599, x = aw_579_cast);
            tensor<fp16, [2, 77, 1, 256]> var_8481_cast = softmax(axis = var_5599, x = aw_581_cast);
            tensor<fp16, [2, 77, 1, 256]> var_8482_cast = softmax(axis = var_5599, x = aw_583_cast);
            tensor<fp16, [2, 77, 1, 256]> var_8483_cast = softmax(axis = var_5599, x = aw_585_cast);
            tensor<fp16, [2, 77, 1, 256]> var_8484_cast = softmax(axis = var_5599, x = aw_587_cast);
            tensor<fp16, [2, 77, 1, 256]> var_8485_cast = softmax(axis = var_5599, x = aw_589_cast);
            tensor<fp16, [2, 77, 1, 256]> var_8486_cast = softmax(axis = var_5599, x = aw_591_cast);
            tensor<fp16, [2, 77, 1, 256]> var_8487_cast = softmax(axis = var_5599, x = aw_593_cast);
            tensor<fp16, [2, 77, 1, 256]> var_8488_cast = softmax(axis = var_5599, x = aw_595_cast);
            tensor<fp16, [2, 77, 1, 256]> var_8489_cast = softmax(axis = var_5599, x = aw_597_cast);
            tensor<fp16, [2, 77, 1, 256]> var_8490_cast = softmax(axis = var_5599, x = aw_599_cast);
            tensor<string, []> var_8492_equation_0 = const()[name = tensor<string, []>("op_8492_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8492_cast = einsum(equation = var_8492_equation_0, values = (var_8312_cast, var_8471_cast));
            tensor<string, []> var_8494_equation_0 = const()[name = tensor<string, []>("op_8494_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8494_cast = einsum(equation = var_8494_equation_0, values = (var_8316_cast, var_8472_cast));
            tensor<string, []> var_8496_equation_0 = const()[name = tensor<string, []>("op_8496_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8496_cast = einsum(equation = var_8496_equation_0, values = (var_8320_cast, var_8473_cast));
            tensor<string, []> var_8498_equation_0 = const()[name = tensor<string, []>("op_8498_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8498_cast = einsum(equation = var_8498_equation_0, values = (var_8324_cast, var_8474_cast));
            tensor<string, []> var_8500_equation_0 = const()[name = tensor<string, []>("op_8500_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8500_cast = einsum(equation = var_8500_equation_0, values = (var_8328_cast, var_8475_cast));
            tensor<string, []> var_8502_equation_0 = const()[name = tensor<string, []>("op_8502_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8502_cast = einsum(equation = var_8502_equation_0, values = (var_8332_cast, var_8476_cast));
            tensor<string, []> var_8504_equation_0 = const()[name = tensor<string, []>("op_8504_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8504_cast = einsum(equation = var_8504_equation_0, values = (var_8336_cast, var_8477_cast));
            tensor<string, []> var_8506_equation_0 = const()[name = tensor<string, []>("op_8506_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8506_cast = einsum(equation = var_8506_equation_0, values = (var_8340_cast, var_8478_cast));
            tensor<string, []> var_8508_equation_0 = const()[name = tensor<string, []>("op_8508_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8508_cast = einsum(equation = var_8508_equation_0, values = (var_8344_cast, var_8479_cast));
            tensor<string, []> var_8510_equation_0 = const()[name = tensor<string, []>("op_8510_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8510_cast = einsum(equation = var_8510_equation_0, values = (var_8348_cast, var_8480_cast));
            tensor<string, []> var_8512_equation_0 = const()[name = tensor<string, []>("op_8512_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8512_cast = einsum(equation = var_8512_equation_0, values = (var_8352_cast, var_8481_cast));
            tensor<string, []> var_8514_equation_0 = const()[name = tensor<string, []>("op_8514_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8514_cast = einsum(equation = var_8514_equation_0, values = (var_8356_cast, var_8482_cast));
            tensor<string, []> var_8516_equation_0 = const()[name = tensor<string, []>("op_8516_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8516_cast = einsum(equation = var_8516_equation_0, values = (var_8360_cast, var_8483_cast));
            tensor<string, []> var_8518_equation_0 = const()[name = tensor<string, []>("op_8518_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8518_cast = einsum(equation = var_8518_equation_0, values = (var_8364_cast, var_8484_cast));
            tensor<string, []> var_8520_equation_0 = const()[name = tensor<string, []>("op_8520_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8520_cast = einsum(equation = var_8520_equation_0, values = (var_8368_cast, var_8485_cast));
            tensor<string, []> var_8522_equation_0 = const()[name = tensor<string, []>("op_8522_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8522_cast = einsum(equation = var_8522_equation_0, values = (var_8372_cast, var_8486_cast));
            tensor<string, []> var_8524_equation_0 = const()[name = tensor<string, []>("op_8524_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8524_cast = einsum(equation = var_8524_equation_0, values = (var_8376_cast, var_8487_cast));
            tensor<string, []> var_8526_equation_0 = const()[name = tensor<string, []>("op_8526_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8526_cast = einsum(equation = var_8526_equation_0, values = (var_8380_cast, var_8488_cast));
            tensor<string, []> var_8528_equation_0 = const()[name = tensor<string, []>("op_8528_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8528_cast = einsum(equation = var_8528_equation_0, values = (var_8384_cast, var_8489_cast));
            tensor<string, []> var_8530_equation_0 = const()[name = tensor<string, []>("op_8530_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8530_cast = einsum(equation = var_8530_equation_0, values = (var_8388_cast, var_8490_cast));
            tensor<bool, []> input_355_interleave_0 = const()[name = tensor<string, []>("input_355_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 1, 256]> input_355_cast = concat(axis = var_5599, interleave = input_355_interleave_0, values = (var_8492_cast, var_8494_cast, var_8496_cast, var_8498_cast, var_8500_cast, var_8502_cast, var_8504_cast, var_8506_cast, var_8508_cast, var_8510_cast, var_8512_cast, var_8514_cast, var_8516_cast, var_8518_cast, var_8520_cast, var_8522_cast, var_8524_cast, var_8526_cast, var_8528_cast, var_8530_cast));
            tensor<int32, [2]> var_8536 = const()[name = tensor<string, []>("op_8536"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8538 = const()[name = tensor<string, []>("op_8538"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_8540_pad_type_0 = const()[name = tensor<string, []>("op_8540_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_8540_pad_0 = const()[name = tensor<string, []>("op_8540_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1473233344)))];
            tensor<fp16, [1280]> up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1476510208)))];
            tensor<fp16, [2, 1280, 1, 256]> var_8540_cast = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_8538, groups = var_5599, pad = var_8540_pad_0, pad_type = var_8540_pad_type_0, strides = var_8536, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_355_cast);
            tensor<fp16, [2, 1280, 1, 256]> inputs_59_cast = add(x = var_8540_cast, y = inputs_57_cast);
            tensor<int32, [1]> var_8544 = const()[name = tensor<string, []>("op_8544"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_59_cast = reduce_mean(axes = var_8544, keep_dims = var_5594, x = inputs_59_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_59_cast = sub(x = inputs_59_cast, y = channels_mean_59_cast);
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_sq_59_cast = mul(x = zero_mean_59_cast, y = zero_mean_59_cast);
            tensor<int32, [1]> var_8548 = const()[name = tensor<string, []>("op_8548"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_8549_cast = reduce_mean(axes = var_8548, keep_dims = var_5594, x = zero_mean_sq_59_cast);
            tensor<fp16, []> var_8550_to_fp16 = const()[name = tensor<string, []>("op_8550_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_8551_cast = add(x = var_8549_cast, y = var_8550_to_fp16);
            tensor<fp16, []> denom_59_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_59_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 256]> denom_59_cast = rsqrt(epsilon = denom_59_epsilon_0_to_fp16, x = var_8551_cast);
            tensor<fp16, [2, 1280, 1, 256]> out_59_cast = mul(x = zero_mean_59_cast, y = denom_59_cast);
            tensor<fp16, [1, 1280, 1, 1]> var_8555_to_fp16 = const()[name = tensor<string, []>("op_8555_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1476512832)))];
            tensor<fp16, [2, 1280, 1, 256]> var_8556_cast = add(x = out_59_cast, y = var_8555_to_fp16);
            tensor<fp16, [1, 1280, 1, 1]> var_8558_to_fp16 = const()[name = tensor<string, []>("op_8558_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1476515456)))];
            tensor<fp16, [2, 1280, 1, 256]> input_357_cast = mul(x = var_8556_cast, y = var_8558_to_fp16);
            tensor<int32, [2]> var_8566 = const()[name = tensor<string, []>("op_8566"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8568 = const()[name = tensor<string, []>("op_8568"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_8570_pad_type_0 = const()[name = tensor<string, []>("op_8570_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_8570_pad_0 = const()[name = tensor<string, []>("op_8570_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [10240, 1280, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [10240, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1476518080)))];
            tensor<fp16, [10240]> up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [10240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1502732544)))];
            tensor<fp16, [2, 10240, 1, 256]> var_8570_cast = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_8568, groups = var_5599, pad = var_8570_pad_0, pad_type = var_8570_pad_type_0, strides = var_8566, weight = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_357_cast);
            tensor<int32, [2]> var_8571_split_sizes_0 = const()[name = tensor<string, []>("op_8571_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_8571_axis_0 = const()[name = tensor<string, []>("op_8571_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 256]> var_8571_cast_0, tensor<fp16, [2, 5120, 1, 256]> var_8571_cast_1 = split(axis = var_8571_axis_0, split_sizes = var_8571_split_sizes_0, x = var_8570_cast);
            tensor<string, []> var_8573_mode_0 = const()[name = tensor<string, []>("op_8573_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 256]> var_8573_cast = gelu(mode = var_8573_mode_0, x = var_8571_cast_1);
            tensor<fp16, [2, 5120, 1, 256]> input_359_cast = mul(x = var_8571_cast_0, y = var_8573_cast);
            tensor<int32, [2]> var_8577 = const()[name = tensor<string, []>("op_8577"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8579 = const()[name = tensor<string, []>("op_8579"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_8581_pad_type_0 = const()[name = tensor<string, []>("op_8581_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_8581_pad_0 = const()[name = tensor<string, []>("op_8581_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 5120, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1502753088)))];
            tensor<fp16, [1280]> up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1515860352)))];
            tensor<fp16, [2, 1280, 1, 256]> var_8581_cast = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_8579, groups = var_5599, pad = var_8581_pad_0, pad_type = var_8581_pad_type_0, strides = var_8577, weight = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_359_cast);
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_213_cast = add(x = var_8581_cast, y = inputs_59_cast);
            tensor<int32, [4]> var_8583 = const()[name = tensor<string, []>("op_8583"), val = tensor<int32, [4]>([2, 1280, 16, 16])];
            tensor<fp16, [2, 1280, 16, 16]> input_361_cast = reshape(shape = var_8583, x = hidden_states_213_cast);
            tensor<int32, [2]> var_8587 = const()[name = tensor<string, []>("op_8587"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8589 = const()[name = tensor<string, []>("op_8589"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_215_pad_type_0 = const()[name = tensor<string, []>("hidden_states_215_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_215_pad_0 = const()[name = tensor<string, []>("hidden_states_215_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1280, 1, 1]> up_blocks_1_attentions_2_proj_out_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_proj_out_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1515862976)))];
            tensor<fp16, [1280]> up_blocks_1_attentions_2_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_attentions_2_proj_out_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1519139840)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_215_cast = conv(bias = up_blocks_1_attentions_2_proj_out_bias_to_fp16, dilations = var_8589, groups = var_5599, pad = hidden_states_215_pad_0, pad_type = hidden_states_215_pad_type_0, strides = var_8587, weight = up_blocks_1_attentions_2_proj_out_weight_to_fp16, x = input_361_cast);
            tensor<fp16, [2, 1280, 16, 16]> input_363_cast = add(x = hidden_states_215_cast, y = hidden_states_203_cast);
            tensor<fp32, []> input_365_scale_factor_height_0 = const()[name = tensor<string, []>("input_365_scale_factor_height_0"), val = tensor<fp32, []>(0x1p+1)];
            tensor<fp32, []> input_365_scale_factor_width_0 = const()[name = tensor<string, []>("input_365_scale_factor_width_0"), val = tensor<fp32, []>(0x1p+1)];
            tensor<fp16, [2, 1280, 32, 32]> input_365_cast = upsample_nearest_neighbor(scale_factor_height = input_365_scale_factor_height_0, scale_factor_width = input_365_scale_factor_width_0, x = input_363_cast);
            tensor<int32, [2]> var_8598 = const()[name = tensor<string, []>("op_8598"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8600 = const()[name = tensor<string, []>("op_8600"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_217_pad_type_0 = const()[name = tensor<string, []>("hidden_states_217_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_217_pad_0 = const()[name = tensor<string, []>("hidden_states_217_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [1280, 1280, 3, 3]> up_blocks_1_upsamplers_0_conv_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_upsamplers_0_conv_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1519142464)))];
            tensor<fp16, [1280]> up_blocks_1_upsamplers_0_conv_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_1_upsamplers_0_conv_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1548633728)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_217_cast = conv(bias = up_blocks_1_upsamplers_0_conv_bias_to_fp16, dilations = var_8600, groups = var_5599, pad = hidden_states_217_pad_0, pad_type = hidden_states_217_pad_type_0, strides = var_8598, weight = up_blocks_1_upsamplers_0_conv_weight_to_fp16, x = input_365_cast);
            tensor<bool, []> var_8622 = const()[name = tensor<string, []>("op_8622"), val = tensor<bool, []>(true)];
            tensor<int32, []> var_8627 = const()[name = tensor<string, []>("op_8627"), val = tensor<int32, []>(1)];
            tensor<bool, []> input_367_interleave_0 = const()[name = tensor<string, []>("input_367_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1920, 32, 32]> input_367_cast = concat(axis = var_8627, interleave = input_367_interleave_0, values = (hidden_states_217_cast, input_115_cast));
            tensor<int32, [5]> reshape_168_shape_0 = const()[name = tensor<string, []>("reshape_168_shape_0"), val = tensor<int32, [5]>([2, 32, 60, 32, 32])];
            tensor<fp16, [2, 32, 60, 32, 32]> reshape_168_cast = reshape(shape = reshape_168_shape_0, x = input_367_cast);
            tensor<int32, [3]> reduce_mean_126_axes_0 = const()[name = tensor<string, []>("reduce_mean_126_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_126_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_126_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_126_cast = reduce_mean(axes = reduce_mean_126_axes_0, keep_dims = reduce_mean_126_keep_dims_0, x = reshape_168_cast);
            tensor<fp16, [2, 32, 60, 32, 32]> sub_84_cast = sub(x = reshape_168_cast, y = reduce_mean_126_cast);
            tensor<fp16, [2, 32, 60, 32, 32]> square_42_cast = square(x = sub_84_cast);
            tensor<int32, [3]> reduce_mean_128_axes_0 = const()[name = tensor<string, []>("reduce_mean_128_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_128_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_128_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_128_cast = reduce_mean(axes = reduce_mean_128_axes_0, keep_dims = reduce_mean_128_keep_dims_0, x = square_42_cast);
            tensor<fp16, []> add_84_y_0_to_fp16 = const()[name = tensor<string, []>("add_84_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_84_cast = add(x = reduce_mean_128_cast, y = add_84_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_42_cast = sqrt(x = add_84_cast);
            tensor<fp16, [2, 32, 60, 32, 32]> real_div_42_cast = real_div(x = sub_84_cast, y = sqrt_42_cast);
            tensor<int32, [4]> reshape_169_shape_0 = const()[name = tensor<string, []>("reshape_169_shape_0"), val = tensor<int32, [4]>([2, 1920, 32, 32])];
            tensor<fp16, [2, 1920, 32, 32]> reshape_169_cast = reshape(shape = reshape_169_shape_0, x = real_div_42_cast);
            tensor<fp16, [1920]> add_85_mean_0_to_fp16 = const()[name = tensor<string, []>("add_85_mean_0_to_fp16"), val = tensor<fp16, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1548636352)))];
            tensor<fp16, [1920]> add_85_variance_0_to_fp16 = const()[name = tensor<string, []>("add_85_variance_0_to_fp16"), val = tensor<fp16, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1548640256)))];
            tensor<fp16, [1920]> add_85_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_85_gamma_0_to_fp16"), val = tensor<fp16, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1548644160)))];
            tensor<fp16, [1920]> add_85_beta_0_to_fp16 = const()[name = tensor<string, []>("add_85_beta_0_to_fp16"), val = tensor<fp16, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1548648064)))];
            tensor<fp16, []> add_85_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_85_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1920, 32, 32]> add_85_cast = batch_norm(beta = add_85_beta_0_to_fp16, epsilon = add_85_epsilon_0_to_fp16, gamma = add_85_gamma_0_to_fp16, mean = add_85_mean_0_to_fp16, variance = add_85_variance_0_to_fp16, x = reshape_169_cast);
            tensor<fp16, [2, 1920, 32, 32]> input_371_cast = silu(x = add_85_cast);
            tensor<int32, [2]> var_8656 = const()[name = tensor<string, []>("op_8656"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8658 = const()[name = tensor<string, []>("op_8658"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_219_pad_type_0 = const()[name = tensor<string, []>("hidden_states_219_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_219_pad_0 = const()[name = tensor<string, []>("hidden_states_219_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [640, 1920, 3, 3]> up_blocks_2_resnets_0_conv1_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_0_conv1_weight_to_fp16"), val = tensor<fp16, [640, 1920, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1548651968)))];
            tensor<fp16, [640]> up_blocks_2_resnets_0_conv1_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_0_conv1_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1570770432)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_219_cast = conv(bias = up_blocks_2_resnets_0_conv1_bias_to_fp16, dilations = var_8658, groups = var_8627, pad = hidden_states_219_pad_0, pad_type = hidden_states_219_pad_type_0, strides = var_8656, weight = up_blocks_2_resnets_0_conv1_weight_to_fp16, x = input_371_cast);
            tensor<int32, [2]> var_8664 = const()[name = tensor<string, []>("op_8664"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8666 = const()[name = tensor<string, []>("op_8666"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_33_pad_type_0 = const()[name = tensor<string, []>("temb_33_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_33_pad_0 = const()[name = tensor<string, []>("temb_33_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1280, 1, 1]> up_blocks_2_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [640, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1570771776)))];
            tensor<fp16, [640]> up_blocks_2_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1572410240)))];
            tensor<fp16, [2, 640, 1, 1]> temb_33_cast = conv(bias = up_blocks_2_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_8666, groups = var_8627, pad = temb_33_pad_0, pad_type = temb_33_pad_type_0, strides = var_8664, weight = up_blocks_2_resnets_0_time_emb_proj_weight_to_fp16, x = input_15_cast);
            tensor<fp16, [2, 640, 32, 32]> input_375_cast = add(x = hidden_states_219_cast, y = temb_33_cast);
            tensor<int32, [5]> reshape_172_shape_0 = const()[name = tensor<string, []>("reshape_172_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 32, 32])];
            tensor<fp16, [2, 32, 20, 32, 32]> reshape_172_cast = reshape(shape = reshape_172_shape_0, x = input_375_cast);
            tensor<int32, [3]> reduce_mean_129_axes_0 = const()[name = tensor<string, []>("reduce_mean_129_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_129_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_129_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_129_cast = reduce_mean(axes = reduce_mean_129_axes_0, keep_dims = reduce_mean_129_keep_dims_0, x = reshape_172_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> sub_86_cast = sub(x = reshape_172_cast, y = reduce_mean_129_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> square_43_cast = square(x = sub_86_cast);
            tensor<int32, [3]> reduce_mean_131_axes_0 = const()[name = tensor<string, []>("reduce_mean_131_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_131_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_131_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_131_cast = reduce_mean(axes = reduce_mean_131_axes_0, keep_dims = reduce_mean_131_keep_dims_0, x = square_43_cast);
            tensor<fp16, []> add_86_y_0_to_fp16 = const()[name = tensor<string, []>("add_86_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_86_cast = add(x = reduce_mean_131_cast, y = add_86_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_43_cast = sqrt(x = add_86_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> real_div_43_cast = real_div(x = sub_86_cast, y = sqrt_43_cast);
            tensor<int32, [4]> reshape_173_shape_0 = const()[name = tensor<string, []>("reshape_173_shape_0"), val = tensor<int32, [4]>([2, 640, 32, 32])];
            tensor<fp16, [2, 640, 32, 32]> reshape_173_cast = reshape(shape = reshape_173_shape_0, x = real_div_43_cast);
            tensor<fp16, [640]> add_87_mean_0_to_fp16 = const()[name = tensor<string, []>("add_87_mean_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1572411584)))];
            tensor<fp16, [640]> add_87_variance_0_to_fp16 = const()[name = tensor<string, []>("add_87_variance_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1572412928)))];
            tensor<fp16, [640]> add_87_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_87_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1572414272)))];
            tensor<fp16, [640]> add_87_beta_0_to_fp16 = const()[name = tensor<string, []>("add_87_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1572415616)))];
            tensor<fp16, []> add_87_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_87_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 32, 32]> add_87_cast = batch_norm(beta = add_87_beta_0_to_fp16, epsilon = add_87_epsilon_0_to_fp16, gamma = add_87_gamma_0_to_fp16, mean = add_87_mean_0_to_fp16, variance = add_87_variance_0_to_fp16, x = reshape_173_cast);
            tensor<fp16, [2, 640, 32, 32]> input_379_cast = silu(x = add_87_cast);
            tensor<int32, [2]> var_8676 = const()[name = tensor<string, []>("op_8676"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8678 = const()[name = tensor<string, []>("op_8678"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_221_pad_type_0 = const()[name = tensor<string, []>("hidden_states_221_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_221_pad_0 = const()[name = tensor<string, []>("hidden_states_221_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [640, 640, 3, 3]> up_blocks_2_resnets_0_conv2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_0_conv2_weight_to_fp16"), val = tensor<fp16, [640, 640, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1572416960)))];
            tensor<fp16, [640]> up_blocks_2_resnets_0_conv2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_0_conv2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1579789824)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_221_cast = conv(bias = up_blocks_2_resnets_0_conv2_bias_to_fp16, dilations = var_8678, groups = var_8627, pad = hidden_states_221_pad_0, pad_type = hidden_states_221_pad_type_0, strides = var_8676, weight = up_blocks_2_resnets_0_conv2_weight_to_fp16, x = input_379_cast);
            tensor<int32, [2]> var_8683 = const()[name = tensor<string, []>("op_8683"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8685 = const()[name = tensor<string, []>("op_8685"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> x_17_pad_type_0 = const()[name = tensor<string, []>("x_17_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> x_17_pad_0 = const()[name = tensor<string, []>("x_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1920, 1, 1]> up_blocks_2_resnets_0_conv_shortcut_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_0_conv_shortcut_weight_to_fp16"), val = tensor<fp16, [640, 1920, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1579791168)))];
            tensor<fp16, [640]> up_blocks_2_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1582248832)))];
            tensor<fp16, [2, 640, 32, 32]> x_17_cast = conv(bias = up_blocks_2_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_8685, groups = var_8627, pad = x_17_pad_0, pad_type = x_17_pad_type_0, strides = var_8683, weight = up_blocks_2_resnets_0_conv_shortcut_weight_to_fp16, x = input_367_cast);
            tensor<fp16, [2, 640, 32, 32]> hidden_states_223_cast = add(x = x_17_cast, y = hidden_states_221_cast);
            tensor<int32, [5]> reshape_176_shape_0 = const()[name = tensor<string, []>("reshape_176_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 32, 32])];
            tensor<fp16, [2, 32, 20, 32, 32]> reshape_176_cast = reshape(shape = reshape_176_shape_0, x = hidden_states_223_cast);
            tensor<int32, [3]> reduce_mean_132_axes_0 = const()[name = tensor<string, []>("reduce_mean_132_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_132_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_132_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_132_cast = reduce_mean(axes = reduce_mean_132_axes_0, keep_dims = reduce_mean_132_keep_dims_0, x = reshape_176_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> sub_88_cast = sub(x = reshape_176_cast, y = reduce_mean_132_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> square_44_cast = square(x = sub_88_cast);
            tensor<int32, [3]> reduce_mean_134_axes_0 = const()[name = tensor<string, []>("reduce_mean_134_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_134_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_134_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_134_cast = reduce_mean(axes = reduce_mean_134_axes_0, keep_dims = reduce_mean_134_keep_dims_0, x = square_44_cast);
            tensor<fp16, []> add_88_y_0_to_fp16 = const()[name = tensor<string, []>("add_88_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_88_cast = add(x = reduce_mean_134_cast, y = add_88_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_44_cast = sqrt(x = add_88_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> real_div_44_cast = real_div(x = sub_88_cast, y = sqrt_44_cast);
            tensor<int32, [4]> reshape_177_shape_0 = const()[name = tensor<string, []>("reshape_177_shape_0"), val = tensor<int32, [4]>([2, 640, 32, 32])];
            tensor<fp16, [2, 640, 32, 32]> reshape_177_cast = reshape(shape = reshape_177_shape_0, x = real_div_44_cast);
            tensor<fp16, [640]> add_89_mean_0_to_fp16 = const()[name = tensor<string, []>("add_89_mean_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1582250176)))];
            tensor<fp16, [640]> add_89_variance_0_to_fp16 = const()[name = tensor<string, []>("add_89_variance_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1582251520)))];
            tensor<fp16, [640]> add_89_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_89_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1582252864)))];
            tensor<fp16, [640]> add_89_beta_0_to_fp16 = const()[name = tensor<string, []>("add_89_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1582254208)))];
            tensor<fp16, []> add_89_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_89_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 32, 32]> add_89_cast = batch_norm(beta = add_89_beta_0_to_fp16, epsilon = add_89_epsilon_0_to_fp16, gamma = add_89_gamma_0_to_fp16, mean = add_89_mean_0_to_fp16, variance = add_89_variance_0_to_fp16, x = reshape_177_cast);
            tensor<int32, [2]> var_8705 = const()[name = tensor<string, []>("op_8705"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8707 = const()[name = tensor<string, []>("op_8707"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_225_pad_type_0 = const()[name = tensor<string, []>("hidden_states_225_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_225_pad_0 = const()[name = tensor<string, []>("hidden_states_225_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_2_attentions_0_proj_in_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_0_proj_in_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1582255552)))];
            tensor<fp16, [640]> up_blocks_2_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_0_proj_in_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1583074816)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_225_cast = conv(bias = up_blocks_2_attentions_0_proj_in_bias_to_fp16, dilations = var_8707, groups = var_8627, pad = hidden_states_225_pad_0, pad_type = hidden_states_225_pad_type_0, strides = var_8705, weight = up_blocks_2_attentions_0_proj_in_weight_to_fp16, x = add_89_cast);
            tensor<int32, [4]> var_8712 = const()[name = tensor<string, []>("op_8712"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<fp16, [2, 640, 1, 1024]> inputs_61_cast = reshape(shape = var_8712, x = hidden_states_225_cast);
            tensor<int32, [1]> var_8722 = const()[name = tensor<string, []>("op_8722"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_61_cast = reduce_mean(axes = var_8722, keep_dims = var_8622, x = inputs_61_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_61_cast = sub(x = inputs_61_cast, y = channels_mean_61_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_sq_61_cast = mul(x = zero_mean_61_cast, y = zero_mean_61_cast);
            tensor<int32, [1]> var_8726 = const()[name = tensor<string, []>("op_8726"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_8727_cast = reduce_mean(axes = var_8726, keep_dims = var_8622, x = zero_mean_sq_61_cast);
            tensor<fp16, []> var_8728_to_fp16 = const()[name = tensor<string, []>("op_8728_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_8729_cast = add(x = var_8727_cast, y = var_8728_to_fp16);
            tensor<fp16, []> denom_61_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_61_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_61_cast = rsqrt(epsilon = denom_61_epsilon_0_to_fp16, x = var_8729_cast);
            tensor<fp16, [2, 640, 1, 1024]> out_61_cast = mul(x = zero_mean_61_cast, y = denom_61_cast);
            tensor<fp16, [1, 640, 1, 1]> var_8733_to_fp16 = const()[name = tensor<string, []>("op_8733_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1583076160)))];
            tensor<fp16, [2, 640, 1, 1024]> var_8734_cast = add(x = out_61_cast, y = var_8733_to_fp16);
            tensor<fp16, [1, 640, 1, 1]> var_8736_to_fp16 = const()[name = tensor<string, []>("op_8736_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1583077504)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_227_cast = mul(x = var_8734_cast, y = var_8736_to_fp16);
            tensor<int32, [2]> var_8743 = const()[name = tensor<string, []>("op_8743"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8745 = const()[name = tensor<string, []>("op_8745"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_41_pad_type_0 = const()[name = tensor<string, []>("q_41_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_41_pad_0 = const()[name = tensor<string, []>("q_41_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1583078848)))];
            tensor<fp16, [2, 640, 1, 1024]> q_41_cast = conv(dilations = var_8745, groups = var_8627, pad = q_41_pad_0, pad_type = q_41_pad_type_0, strides = var_8743, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_227_cast);
            tensor<int32, [2]> var_8749 = const()[name = tensor<string, []>("op_8749"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8751 = const()[name = tensor<string, []>("op_8751"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_81_pad_type_0 = const()[name = tensor<string, []>("k_81_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_81_pad_0 = const()[name = tensor<string, []>("k_81_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1583898112)))];
            tensor<fp16, [2, 640, 1, 1024]> k_81_cast = conv(dilations = var_8751, groups = var_8627, pad = k_81_pad_0, pad_type = k_81_pad_type_0, strides = var_8749, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_227_cast);
            tensor<int32, [2]> var_8755 = const()[name = tensor<string, []>("op_8755"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8757 = const()[name = tensor<string, []>("op_8757"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_41_pad_type_0 = const()[name = tensor<string, []>("v_41_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_41_pad_0 = const()[name = tensor<string, []>("v_41_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1584717376)))];
            tensor<fp16, [2, 640, 1, 1024]> v_41_cast = conv(dilations = var_8757, groups = var_8627, pad = v_41_pad_0, pad_type = v_41_pad_type_0, strides = var_8755, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_227_cast);
            tensor<int32, [4]> var_8761_begin_0 = const()[name = tensor<string, []>("op_8761_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8761_end_0 = const()[name = tensor<string, []>("op_8761_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_8761_end_mask_0 = const()[name = tensor<string, []>("op_8761_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8761_cast = slice_by_index(begin = var_8761_begin_0, end = var_8761_end_0, end_mask = var_8761_end_mask_0, x = q_41_cast);
            tensor<int32, [4]> var_8765_begin_0 = const()[name = tensor<string, []>("op_8765_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_8765_end_0 = const()[name = tensor<string, []>("op_8765_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_8765_end_mask_0 = const()[name = tensor<string, []>("op_8765_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8765_cast = slice_by_index(begin = var_8765_begin_0, end = var_8765_end_0, end_mask = var_8765_end_mask_0, x = q_41_cast);
            tensor<int32, [4]> var_8769_begin_0 = const()[name = tensor<string, []>("op_8769_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_8769_end_0 = const()[name = tensor<string, []>("op_8769_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_8769_end_mask_0 = const()[name = tensor<string, []>("op_8769_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8769_cast = slice_by_index(begin = var_8769_begin_0, end = var_8769_end_0, end_mask = var_8769_end_mask_0, x = q_41_cast);
            tensor<int32, [4]> var_8773_begin_0 = const()[name = tensor<string, []>("op_8773_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_8773_end_0 = const()[name = tensor<string, []>("op_8773_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_8773_end_mask_0 = const()[name = tensor<string, []>("op_8773_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8773_cast = slice_by_index(begin = var_8773_begin_0, end = var_8773_end_0, end_mask = var_8773_end_mask_0, x = q_41_cast);
            tensor<int32, [4]> var_8777_begin_0 = const()[name = tensor<string, []>("op_8777_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_8777_end_0 = const()[name = tensor<string, []>("op_8777_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_8777_end_mask_0 = const()[name = tensor<string, []>("op_8777_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8777_cast = slice_by_index(begin = var_8777_begin_0, end = var_8777_end_0, end_mask = var_8777_end_mask_0, x = q_41_cast);
            tensor<int32, [4]> var_8781_begin_0 = const()[name = tensor<string, []>("op_8781_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_8781_end_0 = const()[name = tensor<string, []>("op_8781_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_8781_end_mask_0 = const()[name = tensor<string, []>("op_8781_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8781_cast = slice_by_index(begin = var_8781_begin_0, end = var_8781_end_0, end_mask = var_8781_end_mask_0, x = q_41_cast);
            tensor<int32, [4]> var_8785_begin_0 = const()[name = tensor<string, []>("op_8785_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_8785_end_0 = const()[name = tensor<string, []>("op_8785_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_8785_end_mask_0 = const()[name = tensor<string, []>("op_8785_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8785_cast = slice_by_index(begin = var_8785_begin_0, end = var_8785_end_0, end_mask = var_8785_end_mask_0, x = q_41_cast);
            tensor<int32, [4]> var_8789_begin_0 = const()[name = tensor<string, []>("op_8789_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_8789_end_0 = const()[name = tensor<string, []>("op_8789_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_8789_end_mask_0 = const()[name = tensor<string, []>("op_8789_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8789_cast = slice_by_index(begin = var_8789_begin_0, end = var_8789_end_0, end_mask = var_8789_end_mask_0, x = q_41_cast);
            tensor<int32, [4]> var_8793_begin_0 = const()[name = tensor<string, []>("op_8793_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_8793_end_0 = const()[name = tensor<string, []>("op_8793_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_8793_end_mask_0 = const()[name = tensor<string, []>("op_8793_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8793_cast = slice_by_index(begin = var_8793_begin_0, end = var_8793_end_0, end_mask = var_8793_end_mask_0, x = q_41_cast);
            tensor<int32, [4]> var_8797_begin_0 = const()[name = tensor<string, []>("op_8797_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_8797_end_0 = const()[name = tensor<string, []>("op_8797_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_8797_end_mask_0 = const()[name = tensor<string, []>("op_8797_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8797_cast = slice_by_index(begin = var_8797_begin_0, end = var_8797_end_0, end_mask = var_8797_end_mask_0, x = q_41_cast);
            tensor<int32, [4]> k_83_perm_0 = const()[name = tensor<string, []>("k_83_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_8804_begin_0 = const()[name = tensor<string, []>("op_8804_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8804_end_0 = const()[name = tensor<string, []>("op_8804_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_8804_end_mask_0 = const()[name = tensor<string, []>("op_8804_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 640]> transpose_11 = transpose(perm = k_83_perm_0, x = k_81_cast);
            tensor<fp16, [2, 1024, 1, 64]> var_8804_cast = slice_by_index(begin = var_8804_begin_0, end = var_8804_end_0, end_mask = var_8804_end_mask_0, x = transpose_11);
            tensor<int32, [4]> var_8808_begin_0 = const()[name = tensor<string, []>("op_8808_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_8808_end_0 = const()[name = tensor<string, []>("op_8808_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_8808_end_mask_0 = const()[name = tensor<string, []>("op_8808_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_8808_cast = slice_by_index(begin = var_8808_begin_0, end = var_8808_end_0, end_mask = var_8808_end_mask_0, x = transpose_11);
            tensor<int32, [4]> var_8812_begin_0 = const()[name = tensor<string, []>("op_8812_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_8812_end_0 = const()[name = tensor<string, []>("op_8812_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_8812_end_mask_0 = const()[name = tensor<string, []>("op_8812_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_8812_cast = slice_by_index(begin = var_8812_begin_0, end = var_8812_end_0, end_mask = var_8812_end_mask_0, x = transpose_11);
            tensor<int32, [4]> var_8816_begin_0 = const()[name = tensor<string, []>("op_8816_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_8816_end_0 = const()[name = tensor<string, []>("op_8816_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_8816_end_mask_0 = const()[name = tensor<string, []>("op_8816_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_8816_cast = slice_by_index(begin = var_8816_begin_0, end = var_8816_end_0, end_mask = var_8816_end_mask_0, x = transpose_11);
            tensor<int32, [4]> var_8820_begin_0 = const()[name = tensor<string, []>("op_8820_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_8820_end_0 = const()[name = tensor<string, []>("op_8820_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_8820_end_mask_0 = const()[name = tensor<string, []>("op_8820_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_8820_cast = slice_by_index(begin = var_8820_begin_0, end = var_8820_end_0, end_mask = var_8820_end_mask_0, x = transpose_11);
            tensor<int32, [4]> var_8824_begin_0 = const()[name = tensor<string, []>("op_8824_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_8824_end_0 = const()[name = tensor<string, []>("op_8824_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_8824_end_mask_0 = const()[name = tensor<string, []>("op_8824_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_8824_cast = slice_by_index(begin = var_8824_begin_0, end = var_8824_end_0, end_mask = var_8824_end_mask_0, x = transpose_11);
            tensor<int32, [4]> var_8828_begin_0 = const()[name = tensor<string, []>("op_8828_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_8828_end_0 = const()[name = tensor<string, []>("op_8828_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_8828_end_mask_0 = const()[name = tensor<string, []>("op_8828_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_8828_cast = slice_by_index(begin = var_8828_begin_0, end = var_8828_end_0, end_mask = var_8828_end_mask_0, x = transpose_11);
            tensor<int32, [4]> var_8832_begin_0 = const()[name = tensor<string, []>("op_8832_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_8832_end_0 = const()[name = tensor<string, []>("op_8832_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_8832_end_mask_0 = const()[name = tensor<string, []>("op_8832_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_8832_cast = slice_by_index(begin = var_8832_begin_0, end = var_8832_end_0, end_mask = var_8832_end_mask_0, x = transpose_11);
            tensor<int32, [4]> var_8836_begin_0 = const()[name = tensor<string, []>("op_8836_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_8836_end_0 = const()[name = tensor<string, []>("op_8836_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_8836_end_mask_0 = const()[name = tensor<string, []>("op_8836_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_8836_cast = slice_by_index(begin = var_8836_begin_0, end = var_8836_end_0, end_mask = var_8836_end_mask_0, x = transpose_11);
            tensor<int32, [4]> var_8840_begin_0 = const()[name = tensor<string, []>("op_8840_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_8840_end_0 = const()[name = tensor<string, []>("op_8840_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_8840_end_mask_0 = const()[name = tensor<string, []>("op_8840_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_8840_cast = slice_by_index(begin = var_8840_begin_0, end = var_8840_end_0, end_mask = var_8840_end_mask_0, x = transpose_11);
            tensor<int32, [4]> var_8842_begin_0 = const()[name = tensor<string, []>("op_8842_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8842_end_0 = const()[name = tensor<string, []>("op_8842_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_8842_end_mask_0 = const()[name = tensor<string, []>("op_8842_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8842_cast = slice_by_index(begin = var_8842_begin_0, end = var_8842_end_0, end_mask = var_8842_end_mask_0, x = v_41_cast);
            tensor<int32, [4]> var_8846_begin_0 = const()[name = tensor<string, []>("op_8846_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_8846_end_0 = const()[name = tensor<string, []>("op_8846_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_8846_end_mask_0 = const()[name = tensor<string, []>("op_8846_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8846_cast = slice_by_index(begin = var_8846_begin_0, end = var_8846_end_0, end_mask = var_8846_end_mask_0, x = v_41_cast);
            tensor<int32, [4]> var_8850_begin_0 = const()[name = tensor<string, []>("op_8850_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_8850_end_0 = const()[name = tensor<string, []>("op_8850_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_8850_end_mask_0 = const()[name = tensor<string, []>("op_8850_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8850_cast = slice_by_index(begin = var_8850_begin_0, end = var_8850_end_0, end_mask = var_8850_end_mask_0, x = v_41_cast);
            tensor<int32, [4]> var_8854_begin_0 = const()[name = tensor<string, []>("op_8854_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_8854_end_0 = const()[name = tensor<string, []>("op_8854_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_8854_end_mask_0 = const()[name = tensor<string, []>("op_8854_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8854_cast = slice_by_index(begin = var_8854_begin_0, end = var_8854_end_0, end_mask = var_8854_end_mask_0, x = v_41_cast);
            tensor<int32, [4]> var_8858_begin_0 = const()[name = tensor<string, []>("op_8858_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_8858_end_0 = const()[name = tensor<string, []>("op_8858_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_8858_end_mask_0 = const()[name = tensor<string, []>("op_8858_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8858_cast = slice_by_index(begin = var_8858_begin_0, end = var_8858_end_0, end_mask = var_8858_end_mask_0, x = v_41_cast);
            tensor<int32, [4]> var_8862_begin_0 = const()[name = tensor<string, []>("op_8862_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_8862_end_0 = const()[name = tensor<string, []>("op_8862_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_8862_end_mask_0 = const()[name = tensor<string, []>("op_8862_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8862_cast = slice_by_index(begin = var_8862_begin_0, end = var_8862_end_0, end_mask = var_8862_end_mask_0, x = v_41_cast);
            tensor<int32, [4]> var_8866_begin_0 = const()[name = tensor<string, []>("op_8866_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_8866_end_0 = const()[name = tensor<string, []>("op_8866_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_8866_end_mask_0 = const()[name = tensor<string, []>("op_8866_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8866_cast = slice_by_index(begin = var_8866_begin_0, end = var_8866_end_0, end_mask = var_8866_end_mask_0, x = v_41_cast);
            tensor<int32, [4]> var_8870_begin_0 = const()[name = tensor<string, []>("op_8870_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_8870_end_0 = const()[name = tensor<string, []>("op_8870_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_8870_end_mask_0 = const()[name = tensor<string, []>("op_8870_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8870_cast = slice_by_index(begin = var_8870_begin_0, end = var_8870_end_0, end_mask = var_8870_end_mask_0, x = v_41_cast);
            tensor<int32, [4]> var_8874_begin_0 = const()[name = tensor<string, []>("op_8874_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_8874_end_0 = const()[name = tensor<string, []>("op_8874_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_8874_end_mask_0 = const()[name = tensor<string, []>("op_8874_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8874_cast = slice_by_index(begin = var_8874_begin_0, end = var_8874_end_0, end_mask = var_8874_end_mask_0, x = v_41_cast);
            tensor<int32, [4]> var_8878_begin_0 = const()[name = tensor<string, []>("op_8878_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_8878_end_0 = const()[name = tensor<string, []>("op_8878_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_8878_end_mask_0 = const()[name = tensor<string, []>("op_8878_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_8878_cast = slice_by_index(begin = var_8878_begin_0, end = var_8878_end_0, end_mask = var_8878_end_mask_0, x = v_41_cast);
            tensor<string, []> var_8882_equation_0 = const()[name = tensor<string, []>("op_8882_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8882_cast = einsum(equation = var_8882_equation_0, values = (var_8804_cast, var_8761_cast));
            tensor<fp16, []> var_8883_to_fp16 = const()[name = tensor<string, []>("op_8883_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_601_cast = mul(x = var_8882_cast, y = var_8883_to_fp16);
            tensor<string, []> var_8886_equation_0 = const()[name = tensor<string, []>("op_8886_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8886_cast = einsum(equation = var_8886_equation_0, values = (var_8808_cast, var_8765_cast));
            tensor<fp16, []> var_8887_to_fp16 = const()[name = tensor<string, []>("op_8887_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_603_cast = mul(x = var_8886_cast, y = var_8887_to_fp16);
            tensor<string, []> var_8890_equation_0 = const()[name = tensor<string, []>("op_8890_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8890_cast = einsum(equation = var_8890_equation_0, values = (var_8812_cast, var_8769_cast));
            tensor<fp16, []> var_8891_to_fp16 = const()[name = tensor<string, []>("op_8891_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_605_cast = mul(x = var_8890_cast, y = var_8891_to_fp16);
            tensor<string, []> var_8894_equation_0 = const()[name = tensor<string, []>("op_8894_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8894_cast = einsum(equation = var_8894_equation_0, values = (var_8816_cast, var_8773_cast));
            tensor<fp16, []> var_8895_to_fp16 = const()[name = tensor<string, []>("op_8895_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_607_cast = mul(x = var_8894_cast, y = var_8895_to_fp16);
            tensor<string, []> var_8898_equation_0 = const()[name = tensor<string, []>("op_8898_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8898_cast = einsum(equation = var_8898_equation_0, values = (var_8820_cast, var_8777_cast));
            tensor<fp16, []> var_8899_to_fp16 = const()[name = tensor<string, []>("op_8899_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_609_cast = mul(x = var_8898_cast, y = var_8899_to_fp16);
            tensor<string, []> var_8902_equation_0 = const()[name = tensor<string, []>("op_8902_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8902_cast = einsum(equation = var_8902_equation_0, values = (var_8824_cast, var_8781_cast));
            tensor<fp16, []> var_8903_to_fp16 = const()[name = tensor<string, []>("op_8903_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_611_cast = mul(x = var_8902_cast, y = var_8903_to_fp16);
            tensor<string, []> var_8906_equation_0 = const()[name = tensor<string, []>("op_8906_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8906_cast = einsum(equation = var_8906_equation_0, values = (var_8828_cast, var_8785_cast));
            tensor<fp16, []> var_8907_to_fp16 = const()[name = tensor<string, []>("op_8907_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_613_cast = mul(x = var_8906_cast, y = var_8907_to_fp16);
            tensor<string, []> var_8910_equation_0 = const()[name = tensor<string, []>("op_8910_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8910_cast = einsum(equation = var_8910_equation_0, values = (var_8832_cast, var_8789_cast));
            tensor<fp16, []> var_8911_to_fp16 = const()[name = tensor<string, []>("op_8911_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_615_cast = mul(x = var_8910_cast, y = var_8911_to_fp16);
            tensor<string, []> var_8914_equation_0 = const()[name = tensor<string, []>("op_8914_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8914_cast = einsum(equation = var_8914_equation_0, values = (var_8836_cast, var_8793_cast));
            tensor<fp16, []> var_8915_to_fp16 = const()[name = tensor<string, []>("op_8915_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_617_cast = mul(x = var_8914_cast, y = var_8915_to_fp16);
            tensor<string, []> var_8918_equation_0 = const()[name = tensor<string, []>("op_8918_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_8918_cast = einsum(equation = var_8918_equation_0, values = (var_8840_cast, var_8797_cast));
            tensor<fp16, []> var_8919_to_fp16 = const()[name = tensor<string, []>("op_8919_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_619_cast = mul(x = var_8918_cast, y = var_8919_to_fp16);
            tensor<fp16, [2, 1024, 1, 1024]> var_8921_cast = softmax(axis = var_8627, x = aw_601_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_8922_cast = softmax(axis = var_8627, x = aw_603_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_8923_cast = softmax(axis = var_8627, x = aw_605_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_8924_cast = softmax(axis = var_8627, x = aw_607_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_8925_cast = softmax(axis = var_8627, x = aw_609_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_8926_cast = softmax(axis = var_8627, x = aw_611_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_8927_cast = softmax(axis = var_8627, x = aw_613_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_8928_cast = softmax(axis = var_8627, x = aw_615_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_8929_cast = softmax(axis = var_8627, x = aw_617_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_8930_cast = softmax(axis = var_8627, x = aw_619_cast);
            tensor<string, []> var_8932_equation_0 = const()[name = tensor<string, []>("op_8932_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8932_cast = einsum(equation = var_8932_equation_0, values = (var_8842_cast, var_8921_cast));
            tensor<string, []> var_8934_equation_0 = const()[name = tensor<string, []>("op_8934_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8934_cast = einsum(equation = var_8934_equation_0, values = (var_8846_cast, var_8922_cast));
            tensor<string, []> var_8936_equation_0 = const()[name = tensor<string, []>("op_8936_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8936_cast = einsum(equation = var_8936_equation_0, values = (var_8850_cast, var_8923_cast));
            tensor<string, []> var_8938_equation_0 = const()[name = tensor<string, []>("op_8938_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8938_cast = einsum(equation = var_8938_equation_0, values = (var_8854_cast, var_8924_cast));
            tensor<string, []> var_8940_equation_0 = const()[name = tensor<string, []>("op_8940_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8940_cast = einsum(equation = var_8940_equation_0, values = (var_8858_cast, var_8925_cast));
            tensor<string, []> var_8942_equation_0 = const()[name = tensor<string, []>("op_8942_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8942_cast = einsum(equation = var_8942_equation_0, values = (var_8862_cast, var_8926_cast));
            tensor<string, []> var_8944_equation_0 = const()[name = tensor<string, []>("op_8944_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8944_cast = einsum(equation = var_8944_equation_0, values = (var_8866_cast, var_8927_cast));
            tensor<string, []> var_8946_equation_0 = const()[name = tensor<string, []>("op_8946_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8946_cast = einsum(equation = var_8946_equation_0, values = (var_8870_cast, var_8928_cast));
            tensor<string, []> var_8948_equation_0 = const()[name = tensor<string, []>("op_8948_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8948_cast = einsum(equation = var_8948_equation_0, values = (var_8874_cast, var_8929_cast));
            tensor<string, []> var_8950_equation_0 = const()[name = tensor<string, []>("op_8950_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_8950_cast = einsum(equation = var_8950_equation_0, values = (var_8878_cast, var_8930_cast));
            tensor<bool, []> input_383_interleave_0 = const()[name = tensor<string, []>("input_383_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 1024]> input_383_cast = concat(axis = var_8627, interleave = input_383_interleave_0, values = (var_8932_cast, var_8934_cast, var_8936_cast, var_8938_cast, var_8940_cast, var_8942_cast, var_8944_cast, var_8946_cast, var_8948_cast, var_8950_cast));
            tensor<int32, [2]> var_8956 = const()[name = tensor<string, []>("op_8956"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8958 = const()[name = tensor<string, []>("op_8958"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_8960_pad_type_0 = const()[name = tensor<string, []>("op_8960_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_8960_pad_0 = const()[name = tensor<string, []>("op_8960_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1585536640)))];
            tensor<fp16, [640]> up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1586355904)))];
            tensor<fp16, [2, 640, 1, 1024]> var_8960_cast = conv(bias = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_8958, groups = var_8627, pad = var_8960_pad_0, pad_type = var_8960_pad_type_0, strides = var_8956, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_383_cast);
            tensor<fp16, [2, 640, 1, 1024]> inputs_63_cast = add(x = var_8960_cast, y = inputs_61_cast);
            tensor<int32, [1]> var_8964 = const()[name = tensor<string, []>("op_8964"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_63_cast = reduce_mean(axes = var_8964, keep_dims = var_8622, x = inputs_63_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_63_cast = sub(x = inputs_63_cast, y = channels_mean_63_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_sq_63_cast = mul(x = zero_mean_63_cast, y = zero_mean_63_cast);
            tensor<int32, [1]> var_8968 = const()[name = tensor<string, []>("op_8968"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_8969_cast = reduce_mean(axes = var_8968, keep_dims = var_8622, x = zero_mean_sq_63_cast);
            tensor<fp16, []> var_8970_to_fp16 = const()[name = tensor<string, []>("op_8970_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_8971_cast = add(x = var_8969_cast, y = var_8970_to_fp16);
            tensor<fp16, []> denom_63_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_63_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_63_cast = rsqrt(epsilon = denom_63_epsilon_0_to_fp16, x = var_8971_cast);
            tensor<fp16, [2, 640, 1, 1024]> out_63_cast = mul(x = zero_mean_63_cast, y = denom_63_cast);
            tensor<fp16, [1, 640, 1, 1]> var_8975_to_fp16 = const()[name = tensor<string, []>("op_8975_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1586357248)))];
            tensor<fp16, [2, 640, 1, 1024]> var_8976_cast = add(x = out_63_cast, y = var_8975_to_fp16);
            tensor<fp16, [1, 640, 1, 1]> var_8978_to_fp16 = const()[name = tensor<string, []>("op_8978_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1586358592)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_229_cast = mul(x = var_8976_cast, y = var_8978_to_fp16);
            tensor<int32, [2]> var_8985 = const()[name = tensor<string, []>("op_8985"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8987 = const()[name = tensor<string, []>("op_8987"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_43_pad_type_0 = const()[name = tensor<string, []>("q_43_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_43_pad_0 = const()[name = tensor<string, []>("q_43_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1586359936)))];
            tensor<fp16, [2, 640, 1, 1024]> q_43_cast = conv(dilations = var_8987, groups = var_8627, pad = q_43_pad_0, pad_type = q_43_pad_type_0, strides = var_8985, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_229_cast);
            tensor<int32, [2]> var_8991 = const()[name = tensor<string, []>("op_8991"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8993 = const()[name = tensor<string, []>("op_8993"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_85_pad_type_0 = const()[name = tensor<string, []>("k_85_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_85_pad_0 = const()[name = tensor<string, []>("k_85_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1024, 1, 1]> up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [640, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1587179200)))];
            tensor<fp16, [2, 640, 1, 77]> k_85_cast = conv(dilations = var_8993, groups = var_8627, pad = k_85_pad_0, pad_type = k_85_pad_type_0, strides = var_8991, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [2]> var_8997 = const()[name = tensor<string, []>("op_8997"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8999 = const()[name = tensor<string, []>("op_8999"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_43_pad_type_0 = const()[name = tensor<string, []>("v_43_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_43_pad_0 = const()[name = tensor<string, []>("v_43_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1024, 1, 1]> up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [640, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1588489984)))];
            tensor<fp16, [2, 640, 1, 77]> v_43_cast = conv(dilations = var_8999, groups = var_8627, pad = v_43_pad_0, pad_type = v_43_pad_type_0, strides = var_8997, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [4]> var_9003_begin_0 = const()[name = tensor<string, []>("op_9003_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9003_end_0 = const()[name = tensor<string, []>("op_9003_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_9003_end_mask_0 = const()[name = tensor<string, []>("op_9003_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9003_cast = slice_by_index(begin = var_9003_begin_0, end = var_9003_end_0, end_mask = var_9003_end_mask_0, x = q_43_cast);
            tensor<int32, [4]> var_9007_begin_0 = const()[name = tensor<string, []>("op_9007_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_9007_end_0 = const()[name = tensor<string, []>("op_9007_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_9007_end_mask_0 = const()[name = tensor<string, []>("op_9007_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9007_cast = slice_by_index(begin = var_9007_begin_0, end = var_9007_end_0, end_mask = var_9007_end_mask_0, x = q_43_cast);
            tensor<int32, [4]> var_9011_begin_0 = const()[name = tensor<string, []>("op_9011_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_9011_end_0 = const()[name = tensor<string, []>("op_9011_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_9011_end_mask_0 = const()[name = tensor<string, []>("op_9011_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9011_cast = slice_by_index(begin = var_9011_begin_0, end = var_9011_end_0, end_mask = var_9011_end_mask_0, x = q_43_cast);
            tensor<int32, [4]> var_9015_begin_0 = const()[name = tensor<string, []>("op_9015_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_9015_end_0 = const()[name = tensor<string, []>("op_9015_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_9015_end_mask_0 = const()[name = tensor<string, []>("op_9015_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9015_cast = slice_by_index(begin = var_9015_begin_0, end = var_9015_end_0, end_mask = var_9015_end_mask_0, x = q_43_cast);
            tensor<int32, [4]> var_9019_begin_0 = const()[name = tensor<string, []>("op_9019_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_9019_end_0 = const()[name = tensor<string, []>("op_9019_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_9019_end_mask_0 = const()[name = tensor<string, []>("op_9019_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9019_cast = slice_by_index(begin = var_9019_begin_0, end = var_9019_end_0, end_mask = var_9019_end_mask_0, x = q_43_cast);
            tensor<int32, [4]> var_9023_begin_0 = const()[name = tensor<string, []>("op_9023_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_9023_end_0 = const()[name = tensor<string, []>("op_9023_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_9023_end_mask_0 = const()[name = tensor<string, []>("op_9023_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9023_cast = slice_by_index(begin = var_9023_begin_0, end = var_9023_end_0, end_mask = var_9023_end_mask_0, x = q_43_cast);
            tensor<int32, [4]> var_9027_begin_0 = const()[name = tensor<string, []>("op_9027_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_9027_end_0 = const()[name = tensor<string, []>("op_9027_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_9027_end_mask_0 = const()[name = tensor<string, []>("op_9027_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9027_cast = slice_by_index(begin = var_9027_begin_0, end = var_9027_end_0, end_mask = var_9027_end_mask_0, x = q_43_cast);
            tensor<int32, [4]> var_9031_begin_0 = const()[name = tensor<string, []>("op_9031_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_9031_end_0 = const()[name = tensor<string, []>("op_9031_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_9031_end_mask_0 = const()[name = tensor<string, []>("op_9031_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9031_cast = slice_by_index(begin = var_9031_begin_0, end = var_9031_end_0, end_mask = var_9031_end_mask_0, x = q_43_cast);
            tensor<int32, [4]> var_9035_begin_0 = const()[name = tensor<string, []>("op_9035_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_9035_end_0 = const()[name = tensor<string, []>("op_9035_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_9035_end_mask_0 = const()[name = tensor<string, []>("op_9035_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9035_cast = slice_by_index(begin = var_9035_begin_0, end = var_9035_end_0, end_mask = var_9035_end_mask_0, x = q_43_cast);
            tensor<int32, [4]> var_9039_begin_0 = const()[name = tensor<string, []>("op_9039_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_9039_end_0 = const()[name = tensor<string, []>("op_9039_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_9039_end_mask_0 = const()[name = tensor<string, []>("op_9039_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9039_cast = slice_by_index(begin = var_9039_begin_0, end = var_9039_end_0, end_mask = var_9039_end_mask_0, x = q_43_cast);
            tensor<int32, [4]> k_87_perm_0 = const()[name = tensor<string, []>("k_87_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_9046_begin_0 = const()[name = tensor<string, []>("op_9046_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9046_end_0 = const()[name = tensor<string, []>("op_9046_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_9046_end_mask_0 = const()[name = tensor<string, []>("op_9046_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 640]> transpose_10 = transpose(perm = k_87_perm_0, x = k_85_cast);
            tensor<fp16, [2, 77, 1, 64]> var_9046_cast = slice_by_index(begin = var_9046_begin_0, end = var_9046_end_0, end_mask = var_9046_end_mask_0, x = transpose_10);
            tensor<int32, [4]> var_9050_begin_0 = const()[name = tensor<string, []>("op_9050_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_9050_end_0 = const()[name = tensor<string, []>("op_9050_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_9050_end_mask_0 = const()[name = tensor<string, []>("op_9050_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9050_cast = slice_by_index(begin = var_9050_begin_0, end = var_9050_end_0, end_mask = var_9050_end_mask_0, x = transpose_10);
            tensor<int32, [4]> var_9054_begin_0 = const()[name = tensor<string, []>("op_9054_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_9054_end_0 = const()[name = tensor<string, []>("op_9054_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_9054_end_mask_0 = const()[name = tensor<string, []>("op_9054_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9054_cast = slice_by_index(begin = var_9054_begin_0, end = var_9054_end_0, end_mask = var_9054_end_mask_0, x = transpose_10);
            tensor<int32, [4]> var_9058_begin_0 = const()[name = tensor<string, []>("op_9058_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_9058_end_0 = const()[name = tensor<string, []>("op_9058_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_9058_end_mask_0 = const()[name = tensor<string, []>("op_9058_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9058_cast = slice_by_index(begin = var_9058_begin_0, end = var_9058_end_0, end_mask = var_9058_end_mask_0, x = transpose_10);
            tensor<int32, [4]> var_9062_begin_0 = const()[name = tensor<string, []>("op_9062_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_9062_end_0 = const()[name = tensor<string, []>("op_9062_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_9062_end_mask_0 = const()[name = tensor<string, []>("op_9062_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9062_cast = slice_by_index(begin = var_9062_begin_0, end = var_9062_end_0, end_mask = var_9062_end_mask_0, x = transpose_10);
            tensor<int32, [4]> var_9066_begin_0 = const()[name = tensor<string, []>("op_9066_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_9066_end_0 = const()[name = tensor<string, []>("op_9066_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_9066_end_mask_0 = const()[name = tensor<string, []>("op_9066_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9066_cast = slice_by_index(begin = var_9066_begin_0, end = var_9066_end_0, end_mask = var_9066_end_mask_0, x = transpose_10);
            tensor<int32, [4]> var_9070_begin_0 = const()[name = tensor<string, []>("op_9070_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_9070_end_0 = const()[name = tensor<string, []>("op_9070_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_9070_end_mask_0 = const()[name = tensor<string, []>("op_9070_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9070_cast = slice_by_index(begin = var_9070_begin_0, end = var_9070_end_0, end_mask = var_9070_end_mask_0, x = transpose_10);
            tensor<int32, [4]> var_9074_begin_0 = const()[name = tensor<string, []>("op_9074_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_9074_end_0 = const()[name = tensor<string, []>("op_9074_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_9074_end_mask_0 = const()[name = tensor<string, []>("op_9074_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9074_cast = slice_by_index(begin = var_9074_begin_0, end = var_9074_end_0, end_mask = var_9074_end_mask_0, x = transpose_10);
            tensor<int32, [4]> var_9078_begin_0 = const()[name = tensor<string, []>("op_9078_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9078_end_0 = const()[name = tensor<string, []>("op_9078_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_9078_end_mask_0 = const()[name = tensor<string, []>("op_9078_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9078_cast = slice_by_index(begin = var_9078_begin_0, end = var_9078_end_0, end_mask = var_9078_end_mask_0, x = transpose_10);
            tensor<int32, [4]> var_9082_begin_0 = const()[name = tensor<string, []>("op_9082_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_9082_end_0 = const()[name = tensor<string, []>("op_9082_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_9082_end_mask_0 = const()[name = tensor<string, []>("op_9082_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9082_cast = slice_by_index(begin = var_9082_begin_0, end = var_9082_end_0, end_mask = var_9082_end_mask_0, x = transpose_10);
            tensor<int32, [4]> var_9084_begin_0 = const()[name = tensor<string, []>("op_9084_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9084_end_0 = const()[name = tensor<string, []>("op_9084_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_9084_end_mask_0 = const()[name = tensor<string, []>("op_9084_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9084_cast = slice_by_index(begin = var_9084_begin_0, end = var_9084_end_0, end_mask = var_9084_end_mask_0, x = v_43_cast);
            tensor<int32, [4]> var_9088_begin_0 = const()[name = tensor<string, []>("op_9088_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_9088_end_0 = const()[name = tensor<string, []>("op_9088_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_9088_end_mask_0 = const()[name = tensor<string, []>("op_9088_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9088_cast = slice_by_index(begin = var_9088_begin_0, end = var_9088_end_0, end_mask = var_9088_end_mask_0, x = v_43_cast);
            tensor<int32, [4]> var_9092_begin_0 = const()[name = tensor<string, []>("op_9092_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_9092_end_0 = const()[name = tensor<string, []>("op_9092_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_9092_end_mask_0 = const()[name = tensor<string, []>("op_9092_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9092_cast = slice_by_index(begin = var_9092_begin_0, end = var_9092_end_0, end_mask = var_9092_end_mask_0, x = v_43_cast);
            tensor<int32, [4]> var_9096_begin_0 = const()[name = tensor<string, []>("op_9096_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_9096_end_0 = const()[name = tensor<string, []>("op_9096_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_9096_end_mask_0 = const()[name = tensor<string, []>("op_9096_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9096_cast = slice_by_index(begin = var_9096_begin_0, end = var_9096_end_0, end_mask = var_9096_end_mask_0, x = v_43_cast);
            tensor<int32, [4]> var_9100_begin_0 = const()[name = tensor<string, []>("op_9100_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_9100_end_0 = const()[name = tensor<string, []>("op_9100_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_9100_end_mask_0 = const()[name = tensor<string, []>("op_9100_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9100_cast = slice_by_index(begin = var_9100_begin_0, end = var_9100_end_0, end_mask = var_9100_end_mask_0, x = v_43_cast);
            tensor<int32, [4]> var_9104_begin_0 = const()[name = tensor<string, []>("op_9104_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_9104_end_0 = const()[name = tensor<string, []>("op_9104_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_9104_end_mask_0 = const()[name = tensor<string, []>("op_9104_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9104_cast = slice_by_index(begin = var_9104_begin_0, end = var_9104_end_0, end_mask = var_9104_end_mask_0, x = v_43_cast);
            tensor<int32, [4]> var_9108_begin_0 = const()[name = tensor<string, []>("op_9108_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_9108_end_0 = const()[name = tensor<string, []>("op_9108_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_9108_end_mask_0 = const()[name = tensor<string, []>("op_9108_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9108_cast = slice_by_index(begin = var_9108_begin_0, end = var_9108_end_0, end_mask = var_9108_end_mask_0, x = v_43_cast);
            tensor<int32, [4]> var_9112_begin_0 = const()[name = tensor<string, []>("op_9112_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_9112_end_0 = const()[name = tensor<string, []>("op_9112_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_9112_end_mask_0 = const()[name = tensor<string, []>("op_9112_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9112_cast = slice_by_index(begin = var_9112_begin_0, end = var_9112_end_0, end_mask = var_9112_end_mask_0, x = v_43_cast);
            tensor<int32, [4]> var_9116_begin_0 = const()[name = tensor<string, []>("op_9116_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_9116_end_0 = const()[name = tensor<string, []>("op_9116_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_9116_end_mask_0 = const()[name = tensor<string, []>("op_9116_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9116_cast = slice_by_index(begin = var_9116_begin_0, end = var_9116_end_0, end_mask = var_9116_end_mask_0, x = v_43_cast);
            tensor<int32, [4]> var_9120_begin_0 = const()[name = tensor<string, []>("op_9120_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_9120_end_0 = const()[name = tensor<string, []>("op_9120_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_9120_end_mask_0 = const()[name = tensor<string, []>("op_9120_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9120_cast = slice_by_index(begin = var_9120_begin_0, end = var_9120_end_0, end_mask = var_9120_end_mask_0, x = v_43_cast);
            tensor<string, []> var_9124_equation_0 = const()[name = tensor<string, []>("op_9124_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9124_cast = einsum(equation = var_9124_equation_0, values = (var_9046_cast, var_9003_cast));
            tensor<fp16, []> var_9125_to_fp16 = const()[name = tensor<string, []>("op_9125_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_621_cast = mul(x = var_9124_cast, y = var_9125_to_fp16);
            tensor<string, []> var_9128_equation_0 = const()[name = tensor<string, []>("op_9128_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9128_cast = einsum(equation = var_9128_equation_0, values = (var_9050_cast, var_9007_cast));
            tensor<fp16, []> var_9129_to_fp16 = const()[name = tensor<string, []>("op_9129_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_623_cast = mul(x = var_9128_cast, y = var_9129_to_fp16);
            tensor<string, []> var_9132_equation_0 = const()[name = tensor<string, []>("op_9132_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9132_cast = einsum(equation = var_9132_equation_0, values = (var_9054_cast, var_9011_cast));
            tensor<fp16, []> var_9133_to_fp16 = const()[name = tensor<string, []>("op_9133_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_625_cast = mul(x = var_9132_cast, y = var_9133_to_fp16);
            tensor<string, []> var_9136_equation_0 = const()[name = tensor<string, []>("op_9136_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9136_cast = einsum(equation = var_9136_equation_0, values = (var_9058_cast, var_9015_cast));
            tensor<fp16, []> var_9137_to_fp16 = const()[name = tensor<string, []>("op_9137_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_627_cast = mul(x = var_9136_cast, y = var_9137_to_fp16);
            tensor<string, []> var_9140_equation_0 = const()[name = tensor<string, []>("op_9140_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9140_cast = einsum(equation = var_9140_equation_0, values = (var_9062_cast, var_9019_cast));
            tensor<fp16, []> var_9141_to_fp16 = const()[name = tensor<string, []>("op_9141_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_629_cast = mul(x = var_9140_cast, y = var_9141_to_fp16);
            tensor<string, []> var_9144_equation_0 = const()[name = tensor<string, []>("op_9144_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9144_cast = einsum(equation = var_9144_equation_0, values = (var_9066_cast, var_9023_cast));
            tensor<fp16, []> var_9145_to_fp16 = const()[name = tensor<string, []>("op_9145_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_631_cast = mul(x = var_9144_cast, y = var_9145_to_fp16);
            tensor<string, []> var_9148_equation_0 = const()[name = tensor<string, []>("op_9148_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9148_cast = einsum(equation = var_9148_equation_0, values = (var_9070_cast, var_9027_cast));
            tensor<fp16, []> var_9149_to_fp16 = const()[name = tensor<string, []>("op_9149_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_633_cast = mul(x = var_9148_cast, y = var_9149_to_fp16);
            tensor<string, []> var_9152_equation_0 = const()[name = tensor<string, []>("op_9152_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9152_cast = einsum(equation = var_9152_equation_0, values = (var_9074_cast, var_9031_cast));
            tensor<fp16, []> var_9153_to_fp16 = const()[name = tensor<string, []>("op_9153_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_635_cast = mul(x = var_9152_cast, y = var_9153_to_fp16);
            tensor<string, []> var_9156_equation_0 = const()[name = tensor<string, []>("op_9156_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9156_cast = einsum(equation = var_9156_equation_0, values = (var_9078_cast, var_9035_cast));
            tensor<fp16, []> var_9157_to_fp16 = const()[name = tensor<string, []>("op_9157_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_637_cast = mul(x = var_9156_cast, y = var_9157_to_fp16);
            tensor<string, []> var_9160_equation_0 = const()[name = tensor<string, []>("op_9160_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9160_cast = einsum(equation = var_9160_equation_0, values = (var_9082_cast, var_9039_cast));
            tensor<fp16, []> var_9161_to_fp16 = const()[name = tensor<string, []>("op_9161_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_639_cast = mul(x = var_9160_cast, y = var_9161_to_fp16);
            tensor<fp16, [2, 77, 1, 1024]> var_9163_cast = softmax(axis = var_8627, x = aw_621_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_9164_cast = softmax(axis = var_8627, x = aw_623_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_9165_cast = softmax(axis = var_8627, x = aw_625_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_9166_cast = softmax(axis = var_8627, x = aw_627_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_9167_cast = softmax(axis = var_8627, x = aw_629_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_9168_cast = softmax(axis = var_8627, x = aw_631_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_9169_cast = softmax(axis = var_8627, x = aw_633_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_9170_cast = softmax(axis = var_8627, x = aw_635_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_9171_cast = softmax(axis = var_8627, x = aw_637_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_9172_cast = softmax(axis = var_8627, x = aw_639_cast);
            tensor<string, []> var_9174_equation_0 = const()[name = tensor<string, []>("op_9174_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9174_cast = einsum(equation = var_9174_equation_0, values = (var_9084_cast, var_9163_cast));
            tensor<string, []> var_9176_equation_0 = const()[name = tensor<string, []>("op_9176_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9176_cast = einsum(equation = var_9176_equation_0, values = (var_9088_cast, var_9164_cast));
            tensor<string, []> var_9178_equation_0 = const()[name = tensor<string, []>("op_9178_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9178_cast = einsum(equation = var_9178_equation_0, values = (var_9092_cast, var_9165_cast));
            tensor<string, []> var_9180_equation_0 = const()[name = tensor<string, []>("op_9180_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9180_cast = einsum(equation = var_9180_equation_0, values = (var_9096_cast, var_9166_cast));
            tensor<string, []> var_9182_equation_0 = const()[name = tensor<string, []>("op_9182_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9182_cast = einsum(equation = var_9182_equation_0, values = (var_9100_cast, var_9167_cast));
            tensor<string, []> var_9184_equation_0 = const()[name = tensor<string, []>("op_9184_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9184_cast = einsum(equation = var_9184_equation_0, values = (var_9104_cast, var_9168_cast));
            tensor<string, []> var_9186_equation_0 = const()[name = tensor<string, []>("op_9186_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9186_cast = einsum(equation = var_9186_equation_0, values = (var_9108_cast, var_9169_cast));
            tensor<string, []> var_9188_equation_0 = const()[name = tensor<string, []>("op_9188_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9188_cast = einsum(equation = var_9188_equation_0, values = (var_9112_cast, var_9170_cast));
            tensor<string, []> var_9190_equation_0 = const()[name = tensor<string, []>("op_9190_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9190_cast = einsum(equation = var_9190_equation_0, values = (var_9116_cast, var_9171_cast));
            tensor<string, []> var_9192_equation_0 = const()[name = tensor<string, []>("op_9192_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9192_cast = einsum(equation = var_9192_equation_0, values = (var_9120_cast, var_9172_cast));
            tensor<bool, []> input_385_interleave_0 = const()[name = tensor<string, []>("input_385_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 1024]> input_385_cast = concat(axis = var_8627, interleave = input_385_interleave_0, values = (var_9174_cast, var_9176_cast, var_9178_cast, var_9180_cast, var_9182_cast, var_9184_cast, var_9186_cast, var_9188_cast, var_9190_cast, var_9192_cast));
            tensor<int32, [2]> var_9198 = const()[name = tensor<string, []>("op_9198"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9200 = const()[name = tensor<string, []>("op_9200"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_9202_pad_type_0 = const()[name = tensor<string, []>("op_9202_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_9202_pad_0 = const()[name = tensor<string, []>("op_9202_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1589800768)))];
            tensor<fp16, [640]> up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1590620032)))];
            tensor<fp16, [2, 640, 1, 1024]> var_9202_cast = conv(bias = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_9200, groups = var_8627, pad = var_9202_pad_0, pad_type = var_9202_pad_type_0, strides = var_9198, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_385_cast);
            tensor<fp16, [2, 640, 1, 1024]> inputs_65_cast = add(x = var_9202_cast, y = inputs_63_cast);
            tensor<int32, [1]> var_9206 = const()[name = tensor<string, []>("op_9206"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_65_cast = reduce_mean(axes = var_9206, keep_dims = var_8622, x = inputs_65_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_65_cast = sub(x = inputs_65_cast, y = channels_mean_65_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_sq_65_cast = mul(x = zero_mean_65_cast, y = zero_mean_65_cast);
            tensor<int32, [1]> var_9210 = const()[name = tensor<string, []>("op_9210"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_9211_cast = reduce_mean(axes = var_9210, keep_dims = var_8622, x = zero_mean_sq_65_cast);
            tensor<fp16, []> var_9212_to_fp16 = const()[name = tensor<string, []>("op_9212_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_9213_cast = add(x = var_9211_cast, y = var_9212_to_fp16);
            tensor<fp16, []> denom_65_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_65_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_65_cast = rsqrt(epsilon = denom_65_epsilon_0_to_fp16, x = var_9213_cast);
            tensor<fp16, [2, 640, 1, 1024]> out_65_cast = mul(x = zero_mean_65_cast, y = denom_65_cast);
            tensor<fp16, [1, 640, 1, 1]> var_9217_to_fp16 = const()[name = tensor<string, []>("op_9217_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1590621376)))];
            tensor<fp16, [2, 640, 1, 1024]> var_9218_cast = add(x = out_65_cast, y = var_9217_to_fp16);
            tensor<fp16, [1, 640, 1, 1]> var_9220_to_fp16 = const()[name = tensor<string, []>("op_9220_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1590622720)))];
            tensor<fp16, [2, 640, 1, 1024]> input_387_cast = mul(x = var_9218_cast, y = var_9220_to_fp16);
            tensor<int32, [2]> var_9228 = const()[name = tensor<string, []>("op_9228"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9230 = const()[name = tensor<string, []>("op_9230"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_9232_pad_type_0 = const()[name = tensor<string, []>("op_9232_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_9232_pad_0 = const()[name = tensor<string, []>("op_9232_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [5120, 640, 1, 1]> up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [5120, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1590624064)))];
            tensor<fp16, [5120]> up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1597177728)))];
            tensor<fp16, [2, 5120, 1, 1024]> var_9232_cast = conv(bias = up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_9230, groups = var_8627, pad = var_9232_pad_0, pad_type = var_9232_pad_type_0, strides = var_9228, weight = up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_387_cast);
            tensor<int32, [2]> var_9233_split_sizes_0 = const()[name = tensor<string, []>("op_9233_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_9233_axis_0 = const()[name = tensor<string, []>("op_9233_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 1024]> var_9233_cast_0, tensor<fp16, [2, 2560, 1, 1024]> var_9233_cast_1 = split(axis = var_9233_axis_0, split_sizes = var_9233_split_sizes_0, x = var_9232_cast);
            tensor<string, []> var_9235_mode_0 = const()[name = tensor<string, []>("op_9235_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 1024]> var_9235_cast = gelu(mode = var_9235_mode_0, x = var_9233_cast_1);
            tensor<fp16, [2, 2560, 1, 1024]> input_389_cast = mul(x = var_9233_cast_0, y = var_9235_cast);
            tensor<int32, [2]> var_9239 = const()[name = tensor<string, []>("op_9239"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9241 = const()[name = tensor<string, []>("op_9241"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_9243_pad_type_0 = const()[name = tensor<string, []>("op_9243_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_9243_pad_0 = const()[name = tensor<string, []>("op_9243_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2560, 1, 1]> up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [640, 2560, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1597188032)))];
            tensor<fp16, [640]> up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1600464896)))];
            tensor<fp16, [2, 640, 1, 1024]> var_9243_cast = conv(bias = up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_9241, groups = var_8627, pad = var_9243_pad_0, pad_type = var_9243_pad_type_0, strides = var_9239, weight = up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_389_cast);
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_233_cast = add(x = var_9243_cast, y = inputs_65_cast);
            tensor<int32, [4]> var_9245 = const()[name = tensor<string, []>("op_9245"), val = tensor<int32, [4]>([2, 640, 32, 32])];
            tensor<fp16, [2, 640, 32, 32]> input_391_cast = reshape(shape = var_9245, x = hidden_states_233_cast);
            tensor<int32, [2]> var_9249 = const()[name = tensor<string, []>("op_9249"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9251 = const()[name = tensor<string, []>("op_9251"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_235_pad_type_0 = const()[name = tensor<string, []>("hidden_states_235_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_235_pad_0 = const()[name = tensor<string, []>("hidden_states_235_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_2_attentions_0_proj_out_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_0_proj_out_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1600466240)))];
            tensor<fp16, [640]> up_blocks_2_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_0_proj_out_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1601285504)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_235_cast = conv(bias = up_blocks_2_attentions_0_proj_out_bias_to_fp16, dilations = var_9251, groups = var_8627, pad = hidden_states_235_pad_0, pad_type = hidden_states_235_pad_type_0, strides = var_9249, weight = up_blocks_2_attentions_0_proj_out_weight_to_fp16, x = input_391_cast);
            tensor<fp16, [2, 640, 32, 32]> hidden_states_237_cast = add(x = hidden_states_235_cast, y = hidden_states_223_cast);
            tensor<bool, []> input_393_interleave_0 = const()[name = tensor<string, []>("input_393_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 1280, 32, 32]> input_393_cast = concat(axis = var_8627, interleave = input_393_interleave_0, values = (hidden_states_237_cast, input_89_cast));
            tensor<int32, [5]> reshape_180_shape_0 = const()[name = tensor<string, []>("reshape_180_shape_0"), val = tensor<int32, [5]>([2, 32, 40, 32, 32])];
            tensor<fp16, [2, 32, 40, 32, 32]> reshape_180_cast = reshape(shape = reshape_180_shape_0, x = input_393_cast);
            tensor<int32, [3]> reduce_mean_135_axes_0 = const()[name = tensor<string, []>("reduce_mean_135_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_135_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_135_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_135_cast = reduce_mean(axes = reduce_mean_135_axes_0, keep_dims = reduce_mean_135_keep_dims_0, x = reshape_180_cast);
            tensor<fp16, [2, 32, 40, 32, 32]> sub_90_cast = sub(x = reshape_180_cast, y = reduce_mean_135_cast);
            tensor<fp16, [2, 32, 40, 32, 32]> square_45_cast = square(x = sub_90_cast);
            tensor<int32, [3]> reduce_mean_137_axes_0 = const()[name = tensor<string, []>("reduce_mean_137_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_137_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_137_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_137_cast = reduce_mean(axes = reduce_mean_137_axes_0, keep_dims = reduce_mean_137_keep_dims_0, x = square_45_cast);
            tensor<fp16, []> add_90_y_0_to_fp16 = const()[name = tensor<string, []>("add_90_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_90_cast = add(x = reduce_mean_137_cast, y = add_90_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_45_cast = sqrt(x = add_90_cast);
            tensor<fp16, [2, 32, 40, 32, 32]> real_div_45_cast = real_div(x = sub_90_cast, y = sqrt_45_cast);
            tensor<int32, [4]> reshape_181_shape_0 = const()[name = tensor<string, []>("reshape_181_shape_0"), val = tensor<int32, [4]>([2, 1280, 32, 32])];
            tensor<fp16, [2, 1280, 32, 32]> reshape_181_cast = reshape(shape = reshape_181_shape_0, x = real_div_45_cast);
            tensor<fp16, [1280]> add_91_mean_0_to_fp16 = const()[name = tensor<string, []>("add_91_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1601286848)))];
            tensor<fp16, [1280]> add_91_variance_0_to_fp16 = const()[name = tensor<string, []>("add_91_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1601289472)))];
            tensor<fp16, [1280]> add_91_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_91_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1601292096)))];
            tensor<fp16, [1280]> add_91_beta_0_to_fp16 = const()[name = tensor<string, []>("add_91_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1601294720)))];
            tensor<fp16, []> add_91_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_91_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1280, 32, 32]> add_91_cast = batch_norm(beta = add_91_beta_0_to_fp16, epsilon = add_91_epsilon_0_to_fp16, gamma = add_91_gamma_0_to_fp16, mean = add_91_mean_0_to_fp16, variance = add_91_variance_0_to_fp16, x = reshape_181_cast);
            tensor<fp16, [2, 1280, 32, 32]> input_397_cast = silu(x = add_91_cast);
            tensor<int32, [2]> var_9269 = const()[name = tensor<string, []>("op_9269"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9271 = const()[name = tensor<string, []>("op_9271"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_239_pad_type_0 = const()[name = tensor<string, []>("hidden_states_239_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_239_pad_0 = const()[name = tensor<string, []>("hidden_states_239_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [640, 1280, 3, 3]> up_blocks_2_resnets_1_conv1_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_1_conv1_weight_to_fp16"), val = tensor<fp16, [640, 1280, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1601297344)))];
            tensor<fp16, [640]> up_blocks_2_resnets_1_conv1_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_1_conv1_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1616043008)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_239_cast = conv(bias = up_blocks_2_resnets_1_conv1_bias_to_fp16, dilations = var_9271, groups = var_8627, pad = hidden_states_239_pad_0, pad_type = hidden_states_239_pad_type_0, strides = var_9269, weight = up_blocks_2_resnets_1_conv1_weight_to_fp16, x = input_397_cast);
            tensor<int32, [2]> var_9277 = const()[name = tensor<string, []>("op_9277"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9279 = const()[name = tensor<string, []>("op_9279"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_35_pad_type_0 = const()[name = tensor<string, []>("temb_35_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_35_pad_0 = const()[name = tensor<string, []>("temb_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1280, 1, 1]> up_blocks_2_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [640, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1616044352)))];
            tensor<fp16, [640]> up_blocks_2_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1617682816)))];
            tensor<fp16, [2, 640, 1, 1]> temb_35_cast = conv(bias = up_blocks_2_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_9279, groups = var_8627, pad = temb_35_pad_0, pad_type = temb_35_pad_type_0, strides = var_9277, weight = up_blocks_2_resnets_1_time_emb_proj_weight_to_fp16, x = input_15_cast);
            tensor<fp16, [2, 640, 32, 32]> input_401_cast = add(x = hidden_states_239_cast, y = temb_35_cast);
            tensor<int32, [5]> reshape_184_shape_0 = const()[name = tensor<string, []>("reshape_184_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 32, 32])];
            tensor<fp16, [2, 32, 20, 32, 32]> reshape_184_cast = reshape(shape = reshape_184_shape_0, x = input_401_cast);
            tensor<int32, [3]> reduce_mean_138_axes_0 = const()[name = tensor<string, []>("reduce_mean_138_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_138_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_138_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_138_cast = reduce_mean(axes = reduce_mean_138_axes_0, keep_dims = reduce_mean_138_keep_dims_0, x = reshape_184_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> sub_92_cast = sub(x = reshape_184_cast, y = reduce_mean_138_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> square_46_cast = square(x = sub_92_cast);
            tensor<int32, [3]> reduce_mean_140_axes_0 = const()[name = tensor<string, []>("reduce_mean_140_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_140_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_140_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_140_cast = reduce_mean(axes = reduce_mean_140_axes_0, keep_dims = reduce_mean_140_keep_dims_0, x = square_46_cast);
            tensor<fp16, []> add_92_y_0_to_fp16 = const()[name = tensor<string, []>("add_92_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_92_cast = add(x = reduce_mean_140_cast, y = add_92_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_46_cast = sqrt(x = add_92_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> real_div_46_cast = real_div(x = sub_92_cast, y = sqrt_46_cast);
            tensor<int32, [4]> reshape_185_shape_0 = const()[name = tensor<string, []>("reshape_185_shape_0"), val = tensor<int32, [4]>([2, 640, 32, 32])];
            tensor<fp16, [2, 640, 32, 32]> reshape_185_cast = reshape(shape = reshape_185_shape_0, x = real_div_46_cast);
            tensor<fp16, [640]> add_93_mean_0_to_fp16 = const()[name = tensor<string, []>("add_93_mean_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1617684160)))];
            tensor<fp16, [640]> add_93_variance_0_to_fp16 = const()[name = tensor<string, []>("add_93_variance_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1617685504)))];
            tensor<fp16, [640]> add_93_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_93_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1617686848)))];
            tensor<fp16, [640]> add_93_beta_0_to_fp16 = const()[name = tensor<string, []>("add_93_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1617688192)))];
            tensor<fp16, []> add_93_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_93_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 32, 32]> add_93_cast = batch_norm(beta = add_93_beta_0_to_fp16, epsilon = add_93_epsilon_0_to_fp16, gamma = add_93_gamma_0_to_fp16, mean = add_93_mean_0_to_fp16, variance = add_93_variance_0_to_fp16, x = reshape_185_cast);
            tensor<fp16, [2, 640, 32, 32]> input_405_cast = silu(x = add_93_cast);
            tensor<int32, [2]> var_9289 = const()[name = tensor<string, []>("op_9289"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9291 = const()[name = tensor<string, []>("op_9291"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_241_pad_type_0 = const()[name = tensor<string, []>("hidden_states_241_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_241_pad_0 = const()[name = tensor<string, []>("hidden_states_241_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [640, 640, 3, 3]> up_blocks_2_resnets_1_conv2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_1_conv2_weight_to_fp16"), val = tensor<fp16, [640, 640, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1617689536)))];
            tensor<fp16, [640]> up_blocks_2_resnets_1_conv2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_1_conv2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1625062400)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_241_cast = conv(bias = up_blocks_2_resnets_1_conv2_bias_to_fp16, dilations = var_9291, groups = var_8627, pad = hidden_states_241_pad_0, pad_type = hidden_states_241_pad_type_0, strides = var_9289, weight = up_blocks_2_resnets_1_conv2_weight_to_fp16, x = input_405_cast);
            tensor<int32, [2]> var_9296 = const()[name = tensor<string, []>("op_9296"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9298 = const()[name = tensor<string, []>("op_9298"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> x_19_pad_type_0 = const()[name = tensor<string, []>("x_19_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> x_19_pad_0 = const()[name = tensor<string, []>("x_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1280, 1, 1]> up_blocks_2_resnets_1_conv_shortcut_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_1_conv_shortcut_weight_to_fp16"), val = tensor<fp16, [640, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1625063744)))];
            tensor<fp16, [640]> up_blocks_2_resnets_1_conv_shortcut_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_1_conv_shortcut_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1626702208)))];
            tensor<fp16, [2, 640, 32, 32]> x_19_cast = conv(bias = up_blocks_2_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_9298, groups = var_8627, pad = x_19_pad_0, pad_type = x_19_pad_type_0, strides = var_9296, weight = up_blocks_2_resnets_1_conv_shortcut_weight_to_fp16, x = input_393_cast);
            tensor<fp16, [2, 640, 32, 32]> hidden_states_243_cast = add(x = x_19_cast, y = hidden_states_241_cast);
            tensor<int32, [5]> reshape_188_shape_0 = const()[name = tensor<string, []>("reshape_188_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 32, 32])];
            tensor<fp16, [2, 32, 20, 32, 32]> reshape_188_cast = reshape(shape = reshape_188_shape_0, x = hidden_states_243_cast);
            tensor<int32, [3]> reduce_mean_141_axes_0 = const()[name = tensor<string, []>("reduce_mean_141_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_141_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_141_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_141_cast = reduce_mean(axes = reduce_mean_141_axes_0, keep_dims = reduce_mean_141_keep_dims_0, x = reshape_188_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> sub_94_cast = sub(x = reshape_188_cast, y = reduce_mean_141_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> square_47_cast = square(x = sub_94_cast);
            tensor<int32, [3]> reduce_mean_143_axes_0 = const()[name = tensor<string, []>("reduce_mean_143_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_143_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_143_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_143_cast = reduce_mean(axes = reduce_mean_143_axes_0, keep_dims = reduce_mean_143_keep_dims_0, x = square_47_cast);
            tensor<fp16, []> add_94_y_0_to_fp16 = const()[name = tensor<string, []>("add_94_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_94_cast = add(x = reduce_mean_143_cast, y = add_94_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_47_cast = sqrt(x = add_94_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> real_div_47_cast = real_div(x = sub_94_cast, y = sqrt_47_cast);
            tensor<int32, [4]> reshape_189_shape_0 = const()[name = tensor<string, []>("reshape_189_shape_0"), val = tensor<int32, [4]>([2, 640, 32, 32])];
            tensor<fp16, [2, 640, 32, 32]> reshape_189_cast = reshape(shape = reshape_189_shape_0, x = real_div_47_cast);
            tensor<fp16, [640]> add_95_mean_0_to_fp16 = const()[name = tensor<string, []>("add_95_mean_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1626703552)))];
            tensor<fp16, [640]> add_95_variance_0_to_fp16 = const()[name = tensor<string, []>("add_95_variance_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1626704896)))];
            tensor<fp16, [640]> add_95_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_95_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1626706240)))];
            tensor<fp16, [640]> add_95_beta_0_to_fp16 = const()[name = tensor<string, []>("add_95_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1626707584)))];
            tensor<fp16, []> add_95_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_95_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 32, 32]> add_95_cast = batch_norm(beta = add_95_beta_0_to_fp16, epsilon = add_95_epsilon_0_to_fp16, gamma = add_95_gamma_0_to_fp16, mean = add_95_mean_0_to_fp16, variance = add_95_variance_0_to_fp16, x = reshape_189_cast);
            tensor<int32, [2]> var_9318 = const()[name = tensor<string, []>("op_9318"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9320 = const()[name = tensor<string, []>("op_9320"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_245_pad_type_0 = const()[name = tensor<string, []>("hidden_states_245_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_245_pad_0 = const()[name = tensor<string, []>("hidden_states_245_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_2_attentions_1_proj_in_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_1_proj_in_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1626708928)))];
            tensor<fp16, [640]> up_blocks_2_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_1_proj_in_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1627528192)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_245_cast = conv(bias = up_blocks_2_attentions_1_proj_in_bias_to_fp16, dilations = var_9320, groups = var_8627, pad = hidden_states_245_pad_0, pad_type = hidden_states_245_pad_type_0, strides = var_9318, weight = up_blocks_2_attentions_1_proj_in_weight_to_fp16, x = add_95_cast);
            tensor<int32, [4]> var_9325 = const()[name = tensor<string, []>("op_9325"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<fp16, [2, 640, 1, 1024]> inputs_67_cast = reshape(shape = var_9325, x = hidden_states_245_cast);
            tensor<int32, [1]> var_9335 = const()[name = tensor<string, []>("op_9335"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_67_cast = reduce_mean(axes = var_9335, keep_dims = var_8622, x = inputs_67_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_67_cast = sub(x = inputs_67_cast, y = channels_mean_67_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_sq_67_cast = mul(x = zero_mean_67_cast, y = zero_mean_67_cast);
            tensor<int32, [1]> var_9339 = const()[name = tensor<string, []>("op_9339"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_9340_cast = reduce_mean(axes = var_9339, keep_dims = var_8622, x = zero_mean_sq_67_cast);
            tensor<fp16, []> var_9341_to_fp16 = const()[name = tensor<string, []>("op_9341_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_9342_cast = add(x = var_9340_cast, y = var_9341_to_fp16);
            tensor<fp16, []> denom_67_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_67_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_67_cast = rsqrt(epsilon = denom_67_epsilon_0_to_fp16, x = var_9342_cast);
            tensor<fp16, [2, 640, 1, 1024]> out_67_cast = mul(x = zero_mean_67_cast, y = denom_67_cast);
            tensor<fp16, [1, 640, 1, 1]> var_9346_to_fp16 = const()[name = tensor<string, []>("op_9346_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1627529536)))];
            tensor<fp16, [2, 640, 1, 1024]> var_9347_cast = add(x = out_67_cast, y = var_9346_to_fp16);
            tensor<fp16, [1, 640, 1, 1]> var_9349_to_fp16 = const()[name = tensor<string, []>("op_9349_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1627530880)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_247_cast = mul(x = var_9347_cast, y = var_9349_to_fp16);
            tensor<int32, [2]> var_9356 = const()[name = tensor<string, []>("op_9356"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9358 = const()[name = tensor<string, []>("op_9358"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_45_pad_type_0 = const()[name = tensor<string, []>("q_45_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_45_pad_0 = const()[name = tensor<string, []>("q_45_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1627532224)))];
            tensor<fp16, [2, 640, 1, 1024]> q_45_cast = conv(dilations = var_9358, groups = var_8627, pad = q_45_pad_0, pad_type = q_45_pad_type_0, strides = var_9356, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_247_cast);
            tensor<int32, [2]> var_9362 = const()[name = tensor<string, []>("op_9362"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9364 = const()[name = tensor<string, []>("op_9364"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_89_pad_type_0 = const()[name = tensor<string, []>("k_89_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_89_pad_0 = const()[name = tensor<string, []>("k_89_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1628351488)))];
            tensor<fp16, [2, 640, 1, 1024]> k_89_cast = conv(dilations = var_9364, groups = var_8627, pad = k_89_pad_0, pad_type = k_89_pad_type_0, strides = var_9362, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_247_cast);
            tensor<int32, [2]> var_9368 = const()[name = tensor<string, []>("op_9368"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9370 = const()[name = tensor<string, []>("op_9370"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_45_pad_type_0 = const()[name = tensor<string, []>("v_45_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_45_pad_0 = const()[name = tensor<string, []>("v_45_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1629170752)))];
            tensor<fp16, [2, 640, 1, 1024]> v_45_cast = conv(dilations = var_9370, groups = var_8627, pad = v_45_pad_0, pad_type = v_45_pad_type_0, strides = var_9368, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_247_cast);
            tensor<int32, [4]> var_9374_begin_0 = const()[name = tensor<string, []>("op_9374_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9374_end_0 = const()[name = tensor<string, []>("op_9374_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_9374_end_mask_0 = const()[name = tensor<string, []>("op_9374_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9374_cast = slice_by_index(begin = var_9374_begin_0, end = var_9374_end_0, end_mask = var_9374_end_mask_0, x = q_45_cast);
            tensor<int32, [4]> var_9378_begin_0 = const()[name = tensor<string, []>("op_9378_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_9378_end_0 = const()[name = tensor<string, []>("op_9378_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_9378_end_mask_0 = const()[name = tensor<string, []>("op_9378_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9378_cast = slice_by_index(begin = var_9378_begin_0, end = var_9378_end_0, end_mask = var_9378_end_mask_0, x = q_45_cast);
            tensor<int32, [4]> var_9382_begin_0 = const()[name = tensor<string, []>("op_9382_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_9382_end_0 = const()[name = tensor<string, []>("op_9382_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_9382_end_mask_0 = const()[name = tensor<string, []>("op_9382_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9382_cast = slice_by_index(begin = var_9382_begin_0, end = var_9382_end_0, end_mask = var_9382_end_mask_0, x = q_45_cast);
            tensor<int32, [4]> var_9386_begin_0 = const()[name = tensor<string, []>("op_9386_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_9386_end_0 = const()[name = tensor<string, []>("op_9386_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_9386_end_mask_0 = const()[name = tensor<string, []>("op_9386_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9386_cast = slice_by_index(begin = var_9386_begin_0, end = var_9386_end_0, end_mask = var_9386_end_mask_0, x = q_45_cast);
            tensor<int32, [4]> var_9390_begin_0 = const()[name = tensor<string, []>("op_9390_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_9390_end_0 = const()[name = tensor<string, []>("op_9390_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_9390_end_mask_0 = const()[name = tensor<string, []>("op_9390_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9390_cast = slice_by_index(begin = var_9390_begin_0, end = var_9390_end_0, end_mask = var_9390_end_mask_0, x = q_45_cast);
            tensor<int32, [4]> var_9394_begin_0 = const()[name = tensor<string, []>("op_9394_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_9394_end_0 = const()[name = tensor<string, []>("op_9394_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_9394_end_mask_0 = const()[name = tensor<string, []>("op_9394_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9394_cast = slice_by_index(begin = var_9394_begin_0, end = var_9394_end_0, end_mask = var_9394_end_mask_0, x = q_45_cast);
            tensor<int32, [4]> var_9398_begin_0 = const()[name = tensor<string, []>("op_9398_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_9398_end_0 = const()[name = tensor<string, []>("op_9398_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_9398_end_mask_0 = const()[name = tensor<string, []>("op_9398_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9398_cast = slice_by_index(begin = var_9398_begin_0, end = var_9398_end_0, end_mask = var_9398_end_mask_0, x = q_45_cast);
            tensor<int32, [4]> var_9402_begin_0 = const()[name = tensor<string, []>("op_9402_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_9402_end_0 = const()[name = tensor<string, []>("op_9402_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_9402_end_mask_0 = const()[name = tensor<string, []>("op_9402_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9402_cast = slice_by_index(begin = var_9402_begin_0, end = var_9402_end_0, end_mask = var_9402_end_mask_0, x = q_45_cast);
            tensor<int32, [4]> var_9406_begin_0 = const()[name = tensor<string, []>("op_9406_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_9406_end_0 = const()[name = tensor<string, []>("op_9406_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_9406_end_mask_0 = const()[name = tensor<string, []>("op_9406_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9406_cast = slice_by_index(begin = var_9406_begin_0, end = var_9406_end_0, end_mask = var_9406_end_mask_0, x = q_45_cast);
            tensor<int32, [4]> var_9410_begin_0 = const()[name = tensor<string, []>("op_9410_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_9410_end_0 = const()[name = tensor<string, []>("op_9410_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_9410_end_mask_0 = const()[name = tensor<string, []>("op_9410_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9410_cast = slice_by_index(begin = var_9410_begin_0, end = var_9410_end_0, end_mask = var_9410_end_mask_0, x = q_45_cast);
            tensor<int32, [4]> k_91_perm_0 = const()[name = tensor<string, []>("k_91_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_9417_begin_0 = const()[name = tensor<string, []>("op_9417_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9417_end_0 = const()[name = tensor<string, []>("op_9417_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_9417_end_mask_0 = const()[name = tensor<string, []>("op_9417_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 640]> transpose_9 = transpose(perm = k_91_perm_0, x = k_89_cast);
            tensor<fp16, [2, 1024, 1, 64]> var_9417_cast = slice_by_index(begin = var_9417_begin_0, end = var_9417_end_0, end_mask = var_9417_end_mask_0, x = transpose_9);
            tensor<int32, [4]> var_9421_begin_0 = const()[name = tensor<string, []>("op_9421_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_9421_end_0 = const()[name = tensor<string, []>("op_9421_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_9421_end_mask_0 = const()[name = tensor<string, []>("op_9421_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_9421_cast = slice_by_index(begin = var_9421_begin_0, end = var_9421_end_0, end_mask = var_9421_end_mask_0, x = transpose_9);
            tensor<int32, [4]> var_9425_begin_0 = const()[name = tensor<string, []>("op_9425_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_9425_end_0 = const()[name = tensor<string, []>("op_9425_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_9425_end_mask_0 = const()[name = tensor<string, []>("op_9425_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_9425_cast = slice_by_index(begin = var_9425_begin_0, end = var_9425_end_0, end_mask = var_9425_end_mask_0, x = transpose_9);
            tensor<int32, [4]> var_9429_begin_0 = const()[name = tensor<string, []>("op_9429_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_9429_end_0 = const()[name = tensor<string, []>("op_9429_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_9429_end_mask_0 = const()[name = tensor<string, []>("op_9429_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_9429_cast = slice_by_index(begin = var_9429_begin_0, end = var_9429_end_0, end_mask = var_9429_end_mask_0, x = transpose_9);
            tensor<int32, [4]> var_9433_begin_0 = const()[name = tensor<string, []>("op_9433_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_9433_end_0 = const()[name = tensor<string, []>("op_9433_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_9433_end_mask_0 = const()[name = tensor<string, []>("op_9433_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_9433_cast = slice_by_index(begin = var_9433_begin_0, end = var_9433_end_0, end_mask = var_9433_end_mask_0, x = transpose_9);
            tensor<int32, [4]> var_9437_begin_0 = const()[name = tensor<string, []>("op_9437_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_9437_end_0 = const()[name = tensor<string, []>("op_9437_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_9437_end_mask_0 = const()[name = tensor<string, []>("op_9437_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_9437_cast = slice_by_index(begin = var_9437_begin_0, end = var_9437_end_0, end_mask = var_9437_end_mask_0, x = transpose_9);
            tensor<int32, [4]> var_9441_begin_0 = const()[name = tensor<string, []>("op_9441_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_9441_end_0 = const()[name = tensor<string, []>("op_9441_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_9441_end_mask_0 = const()[name = tensor<string, []>("op_9441_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_9441_cast = slice_by_index(begin = var_9441_begin_0, end = var_9441_end_0, end_mask = var_9441_end_mask_0, x = transpose_9);
            tensor<int32, [4]> var_9445_begin_0 = const()[name = tensor<string, []>("op_9445_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_9445_end_0 = const()[name = tensor<string, []>("op_9445_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_9445_end_mask_0 = const()[name = tensor<string, []>("op_9445_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_9445_cast = slice_by_index(begin = var_9445_begin_0, end = var_9445_end_0, end_mask = var_9445_end_mask_0, x = transpose_9);
            tensor<int32, [4]> var_9449_begin_0 = const()[name = tensor<string, []>("op_9449_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9449_end_0 = const()[name = tensor<string, []>("op_9449_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_9449_end_mask_0 = const()[name = tensor<string, []>("op_9449_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_9449_cast = slice_by_index(begin = var_9449_begin_0, end = var_9449_end_0, end_mask = var_9449_end_mask_0, x = transpose_9);
            tensor<int32, [4]> var_9453_begin_0 = const()[name = tensor<string, []>("op_9453_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_9453_end_0 = const()[name = tensor<string, []>("op_9453_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_9453_end_mask_0 = const()[name = tensor<string, []>("op_9453_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_9453_cast = slice_by_index(begin = var_9453_begin_0, end = var_9453_end_0, end_mask = var_9453_end_mask_0, x = transpose_9);
            tensor<int32, [4]> var_9455_begin_0 = const()[name = tensor<string, []>("op_9455_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9455_end_0 = const()[name = tensor<string, []>("op_9455_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_9455_end_mask_0 = const()[name = tensor<string, []>("op_9455_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9455_cast = slice_by_index(begin = var_9455_begin_0, end = var_9455_end_0, end_mask = var_9455_end_mask_0, x = v_45_cast);
            tensor<int32, [4]> var_9459_begin_0 = const()[name = tensor<string, []>("op_9459_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_9459_end_0 = const()[name = tensor<string, []>("op_9459_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_9459_end_mask_0 = const()[name = tensor<string, []>("op_9459_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9459_cast = slice_by_index(begin = var_9459_begin_0, end = var_9459_end_0, end_mask = var_9459_end_mask_0, x = v_45_cast);
            tensor<int32, [4]> var_9463_begin_0 = const()[name = tensor<string, []>("op_9463_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_9463_end_0 = const()[name = tensor<string, []>("op_9463_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_9463_end_mask_0 = const()[name = tensor<string, []>("op_9463_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9463_cast = slice_by_index(begin = var_9463_begin_0, end = var_9463_end_0, end_mask = var_9463_end_mask_0, x = v_45_cast);
            tensor<int32, [4]> var_9467_begin_0 = const()[name = tensor<string, []>("op_9467_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_9467_end_0 = const()[name = tensor<string, []>("op_9467_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_9467_end_mask_0 = const()[name = tensor<string, []>("op_9467_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9467_cast = slice_by_index(begin = var_9467_begin_0, end = var_9467_end_0, end_mask = var_9467_end_mask_0, x = v_45_cast);
            tensor<int32, [4]> var_9471_begin_0 = const()[name = tensor<string, []>("op_9471_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_9471_end_0 = const()[name = tensor<string, []>("op_9471_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_9471_end_mask_0 = const()[name = tensor<string, []>("op_9471_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9471_cast = slice_by_index(begin = var_9471_begin_0, end = var_9471_end_0, end_mask = var_9471_end_mask_0, x = v_45_cast);
            tensor<int32, [4]> var_9475_begin_0 = const()[name = tensor<string, []>("op_9475_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_9475_end_0 = const()[name = tensor<string, []>("op_9475_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_9475_end_mask_0 = const()[name = tensor<string, []>("op_9475_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9475_cast = slice_by_index(begin = var_9475_begin_0, end = var_9475_end_0, end_mask = var_9475_end_mask_0, x = v_45_cast);
            tensor<int32, [4]> var_9479_begin_0 = const()[name = tensor<string, []>("op_9479_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_9479_end_0 = const()[name = tensor<string, []>("op_9479_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_9479_end_mask_0 = const()[name = tensor<string, []>("op_9479_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9479_cast = slice_by_index(begin = var_9479_begin_0, end = var_9479_end_0, end_mask = var_9479_end_mask_0, x = v_45_cast);
            tensor<int32, [4]> var_9483_begin_0 = const()[name = tensor<string, []>("op_9483_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_9483_end_0 = const()[name = tensor<string, []>("op_9483_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_9483_end_mask_0 = const()[name = tensor<string, []>("op_9483_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9483_cast = slice_by_index(begin = var_9483_begin_0, end = var_9483_end_0, end_mask = var_9483_end_mask_0, x = v_45_cast);
            tensor<int32, [4]> var_9487_begin_0 = const()[name = tensor<string, []>("op_9487_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_9487_end_0 = const()[name = tensor<string, []>("op_9487_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_9487_end_mask_0 = const()[name = tensor<string, []>("op_9487_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9487_cast = slice_by_index(begin = var_9487_begin_0, end = var_9487_end_0, end_mask = var_9487_end_mask_0, x = v_45_cast);
            tensor<int32, [4]> var_9491_begin_0 = const()[name = tensor<string, []>("op_9491_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_9491_end_0 = const()[name = tensor<string, []>("op_9491_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_9491_end_mask_0 = const()[name = tensor<string, []>("op_9491_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9491_cast = slice_by_index(begin = var_9491_begin_0, end = var_9491_end_0, end_mask = var_9491_end_mask_0, x = v_45_cast);
            tensor<string, []> var_9495_equation_0 = const()[name = tensor<string, []>("op_9495_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9495_cast = einsum(equation = var_9495_equation_0, values = (var_9417_cast, var_9374_cast));
            tensor<fp16, []> var_9496_to_fp16 = const()[name = tensor<string, []>("op_9496_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_641_cast = mul(x = var_9495_cast, y = var_9496_to_fp16);
            tensor<string, []> var_9499_equation_0 = const()[name = tensor<string, []>("op_9499_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9499_cast = einsum(equation = var_9499_equation_0, values = (var_9421_cast, var_9378_cast));
            tensor<fp16, []> var_9500_to_fp16 = const()[name = tensor<string, []>("op_9500_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_643_cast = mul(x = var_9499_cast, y = var_9500_to_fp16);
            tensor<string, []> var_9503_equation_0 = const()[name = tensor<string, []>("op_9503_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9503_cast = einsum(equation = var_9503_equation_0, values = (var_9425_cast, var_9382_cast));
            tensor<fp16, []> var_9504_to_fp16 = const()[name = tensor<string, []>("op_9504_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_645_cast = mul(x = var_9503_cast, y = var_9504_to_fp16);
            tensor<string, []> var_9507_equation_0 = const()[name = tensor<string, []>("op_9507_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9507_cast = einsum(equation = var_9507_equation_0, values = (var_9429_cast, var_9386_cast));
            tensor<fp16, []> var_9508_to_fp16 = const()[name = tensor<string, []>("op_9508_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_647_cast = mul(x = var_9507_cast, y = var_9508_to_fp16);
            tensor<string, []> var_9511_equation_0 = const()[name = tensor<string, []>("op_9511_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9511_cast = einsum(equation = var_9511_equation_0, values = (var_9433_cast, var_9390_cast));
            tensor<fp16, []> var_9512_to_fp16 = const()[name = tensor<string, []>("op_9512_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_649_cast = mul(x = var_9511_cast, y = var_9512_to_fp16);
            tensor<string, []> var_9515_equation_0 = const()[name = tensor<string, []>("op_9515_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9515_cast = einsum(equation = var_9515_equation_0, values = (var_9437_cast, var_9394_cast));
            tensor<fp16, []> var_9516_to_fp16 = const()[name = tensor<string, []>("op_9516_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_651_cast = mul(x = var_9515_cast, y = var_9516_to_fp16);
            tensor<string, []> var_9519_equation_0 = const()[name = tensor<string, []>("op_9519_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9519_cast = einsum(equation = var_9519_equation_0, values = (var_9441_cast, var_9398_cast));
            tensor<fp16, []> var_9520_to_fp16 = const()[name = tensor<string, []>("op_9520_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_653_cast = mul(x = var_9519_cast, y = var_9520_to_fp16);
            tensor<string, []> var_9523_equation_0 = const()[name = tensor<string, []>("op_9523_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9523_cast = einsum(equation = var_9523_equation_0, values = (var_9445_cast, var_9402_cast));
            tensor<fp16, []> var_9524_to_fp16 = const()[name = tensor<string, []>("op_9524_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_655_cast = mul(x = var_9523_cast, y = var_9524_to_fp16);
            tensor<string, []> var_9527_equation_0 = const()[name = tensor<string, []>("op_9527_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9527_cast = einsum(equation = var_9527_equation_0, values = (var_9449_cast, var_9406_cast));
            tensor<fp16, []> var_9528_to_fp16 = const()[name = tensor<string, []>("op_9528_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_657_cast = mul(x = var_9527_cast, y = var_9528_to_fp16);
            tensor<string, []> var_9531_equation_0 = const()[name = tensor<string, []>("op_9531_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_9531_cast = einsum(equation = var_9531_equation_0, values = (var_9453_cast, var_9410_cast));
            tensor<fp16, []> var_9532_to_fp16 = const()[name = tensor<string, []>("op_9532_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_659_cast = mul(x = var_9531_cast, y = var_9532_to_fp16);
            tensor<fp16, [2, 1024, 1, 1024]> var_9534_cast = softmax(axis = var_8627, x = aw_641_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_9535_cast = softmax(axis = var_8627, x = aw_643_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_9536_cast = softmax(axis = var_8627, x = aw_645_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_9537_cast = softmax(axis = var_8627, x = aw_647_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_9538_cast = softmax(axis = var_8627, x = aw_649_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_9539_cast = softmax(axis = var_8627, x = aw_651_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_9540_cast = softmax(axis = var_8627, x = aw_653_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_9541_cast = softmax(axis = var_8627, x = aw_655_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_9542_cast = softmax(axis = var_8627, x = aw_657_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_9543_cast = softmax(axis = var_8627, x = aw_659_cast);
            tensor<string, []> var_9545_equation_0 = const()[name = tensor<string, []>("op_9545_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9545_cast = einsum(equation = var_9545_equation_0, values = (var_9455_cast, var_9534_cast));
            tensor<string, []> var_9547_equation_0 = const()[name = tensor<string, []>("op_9547_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9547_cast = einsum(equation = var_9547_equation_0, values = (var_9459_cast, var_9535_cast));
            tensor<string, []> var_9549_equation_0 = const()[name = tensor<string, []>("op_9549_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9549_cast = einsum(equation = var_9549_equation_0, values = (var_9463_cast, var_9536_cast));
            tensor<string, []> var_9551_equation_0 = const()[name = tensor<string, []>("op_9551_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9551_cast = einsum(equation = var_9551_equation_0, values = (var_9467_cast, var_9537_cast));
            tensor<string, []> var_9553_equation_0 = const()[name = tensor<string, []>("op_9553_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9553_cast = einsum(equation = var_9553_equation_0, values = (var_9471_cast, var_9538_cast));
            tensor<string, []> var_9555_equation_0 = const()[name = tensor<string, []>("op_9555_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9555_cast = einsum(equation = var_9555_equation_0, values = (var_9475_cast, var_9539_cast));
            tensor<string, []> var_9557_equation_0 = const()[name = tensor<string, []>("op_9557_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9557_cast = einsum(equation = var_9557_equation_0, values = (var_9479_cast, var_9540_cast));
            tensor<string, []> var_9559_equation_0 = const()[name = tensor<string, []>("op_9559_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9559_cast = einsum(equation = var_9559_equation_0, values = (var_9483_cast, var_9541_cast));
            tensor<string, []> var_9561_equation_0 = const()[name = tensor<string, []>("op_9561_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9561_cast = einsum(equation = var_9561_equation_0, values = (var_9487_cast, var_9542_cast));
            tensor<string, []> var_9563_equation_0 = const()[name = tensor<string, []>("op_9563_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9563_cast = einsum(equation = var_9563_equation_0, values = (var_9491_cast, var_9543_cast));
            tensor<bool, []> input_409_interleave_0 = const()[name = tensor<string, []>("input_409_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 1024]> input_409_cast = concat(axis = var_8627, interleave = input_409_interleave_0, values = (var_9545_cast, var_9547_cast, var_9549_cast, var_9551_cast, var_9553_cast, var_9555_cast, var_9557_cast, var_9559_cast, var_9561_cast, var_9563_cast));
            tensor<int32, [2]> var_9569 = const()[name = tensor<string, []>("op_9569"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9571 = const()[name = tensor<string, []>("op_9571"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_9573_pad_type_0 = const()[name = tensor<string, []>("op_9573_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_9573_pad_0 = const()[name = tensor<string, []>("op_9573_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1629990016)))];
            tensor<fp16, [640]> up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1630809280)))];
            tensor<fp16, [2, 640, 1, 1024]> var_9573_cast = conv(bias = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_9571, groups = var_8627, pad = var_9573_pad_0, pad_type = var_9573_pad_type_0, strides = var_9569, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_409_cast);
            tensor<fp16, [2, 640, 1, 1024]> inputs_69_cast = add(x = var_9573_cast, y = inputs_67_cast);
            tensor<int32, [1]> var_9577 = const()[name = tensor<string, []>("op_9577"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_69_cast = reduce_mean(axes = var_9577, keep_dims = var_8622, x = inputs_69_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_69_cast = sub(x = inputs_69_cast, y = channels_mean_69_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_sq_69_cast = mul(x = zero_mean_69_cast, y = zero_mean_69_cast);
            tensor<int32, [1]> var_9581 = const()[name = tensor<string, []>("op_9581"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_9582_cast = reduce_mean(axes = var_9581, keep_dims = var_8622, x = zero_mean_sq_69_cast);
            tensor<fp16, []> var_9583_to_fp16 = const()[name = tensor<string, []>("op_9583_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_9584_cast = add(x = var_9582_cast, y = var_9583_to_fp16);
            tensor<fp16, []> denom_69_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_69_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_69_cast = rsqrt(epsilon = denom_69_epsilon_0_to_fp16, x = var_9584_cast);
            tensor<fp16, [2, 640, 1, 1024]> out_69_cast = mul(x = zero_mean_69_cast, y = denom_69_cast);
            tensor<fp16, [1, 640, 1, 1]> var_9588_to_fp16 = const()[name = tensor<string, []>("op_9588_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1630810624)))];
            tensor<fp16, [2, 640, 1, 1024]> var_9589_cast = add(x = out_69_cast, y = var_9588_to_fp16);
            tensor<fp16, [1, 640, 1, 1]> var_9591_to_fp16 = const()[name = tensor<string, []>("op_9591_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1630811968)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_249_cast = mul(x = var_9589_cast, y = var_9591_to_fp16);
            tensor<int32, [2]> var_9598 = const()[name = tensor<string, []>("op_9598"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9600 = const()[name = tensor<string, []>("op_9600"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_47_pad_type_0 = const()[name = tensor<string, []>("q_47_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_47_pad_0 = const()[name = tensor<string, []>("q_47_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1630813312)))];
            tensor<fp16, [2, 640, 1, 1024]> q_47_cast = conv(dilations = var_9600, groups = var_8627, pad = q_47_pad_0, pad_type = q_47_pad_type_0, strides = var_9598, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_249_cast);
            tensor<int32, [2]> var_9604 = const()[name = tensor<string, []>("op_9604"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9606 = const()[name = tensor<string, []>("op_9606"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_93_pad_type_0 = const()[name = tensor<string, []>("k_93_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_93_pad_0 = const()[name = tensor<string, []>("k_93_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1024, 1, 1]> up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [640, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1631632576)))];
            tensor<fp16, [2, 640, 1, 77]> k_93_cast = conv(dilations = var_9606, groups = var_8627, pad = k_93_pad_0, pad_type = k_93_pad_type_0, strides = var_9604, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [2]> var_9610 = const()[name = tensor<string, []>("op_9610"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9612 = const()[name = tensor<string, []>("op_9612"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_47_pad_type_0 = const()[name = tensor<string, []>("v_47_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_47_pad_0 = const()[name = tensor<string, []>("v_47_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1024, 1, 1]> up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [640, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1632943360)))];
            tensor<fp16, [2, 640, 1, 77]> v_47_cast = conv(dilations = var_9612, groups = var_8627, pad = v_47_pad_0, pad_type = v_47_pad_type_0, strides = var_9610, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [4]> var_9616_begin_0 = const()[name = tensor<string, []>("op_9616_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9616_end_0 = const()[name = tensor<string, []>("op_9616_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_9616_end_mask_0 = const()[name = tensor<string, []>("op_9616_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9616_cast = slice_by_index(begin = var_9616_begin_0, end = var_9616_end_0, end_mask = var_9616_end_mask_0, x = q_47_cast);
            tensor<int32, [4]> var_9620_begin_0 = const()[name = tensor<string, []>("op_9620_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_9620_end_0 = const()[name = tensor<string, []>("op_9620_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_9620_end_mask_0 = const()[name = tensor<string, []>("op_9620_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9620_cast = slice_by_index(begin = var_9620_begin_0, end = var_9620_end_0, end_mask = var_9620_end_mask_0, x = q_47_cast);
            tensor<int32, [4]> var_9624_begin_0 = const()[name = tensor<string, []>("op_9624_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_9624_end_0 = const()[name = tensor<string, []>("op_9624_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_9624_end_mask_0 = const()[name = tensor<string, []>("op_9624_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9624_cast = slice_by_index(begin = var_9624_begin_0, end = var_9624_end_0, end_mask = var_9624_end_mask_0, x = q_47_cast);
            tensor<int32, [4]> var_9628_begin_0 = const()[name = tensor<string, []>("op_9628_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_9628_end_0 = const()[name = tensor<string, []>("op_9628_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_9628_end_mask_0 = const()[name = tensor<string, []>("op_9628_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9628_cast = slice_by_index(begin = var_9628_begin_0, end = var_9628_end_0, end_mask = var_9628_end_mask_0, x = q_47_cast);
            tensor<int32, [4]> var_9632_begin_0 = const()[name = tensor<string, []>("op_9632_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_9632_end_0 = const()[name = tensor<string, []>("op_9632_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_9632_end_mask_0 = const()[name = tensor<string, []>("op_9632_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9632_cast = slice_by_index(begin = var_9632_begin_0, end = var_9632_end_0, end_mask = var_9632_end_mask_0, x = q_47_cast);
            tensor<int32, [4]> var_9636_begin_0 = const()[name = tensor<string, []>("op_9636_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_9636_end_0 = const()[name = tensor<string, []>("op_9636_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_9636_end_mask_0 = const()[name = tensor<string, []>("op_9636_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9636_cast = slice_by_index(begin = var_9636_begin_0, end = var_9636_end_0, end_mask = var_9636_end_mask_0, x = q_47_cast);
            tensor<int32, [4]> var_9640_begin_0 = const()[name = tensor<string, []>("op_9640_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_9640_end_0 = const()[name = tensor<string, []>("op_9640_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_9640_end_mask_0 = const()[name = tensor<string, []>("op_9640_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9640_cast = slice_by_index(begin = var_9640_begin_0, end = var_9640_end_0, end_mask = var_9640_end_mask_0, x = q_47_cast);
            tensor<int32, [4]> var_9644_begin_0 = const()[name = tensor<string, []>("op_9644_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_9644_end_0 = const()[name = tensor<string, []>("op_9644_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_9644_end_mask_0 = const()[name = tensor<string, []>("op_9644_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9644_cast = slice_by_index(begin = var_9644_begin_0, end = var_9644_end_0, end_mask = var_9644_end_mask_0, x = q_47_cast);
            tensor<int32, [4]> var_9648_begin_0 = const()[name = tensor<string, []>("op_9648_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_9648_end_0 = const()[name = tensor<string, []>("op_9648_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_9648_end_mask_0 = const()[name = tensor<string, []>("op_9648_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9648_cast = slice_by_index(begin = var_9648_begin_0, end = var_9648_end_0, end_mask = var_9648_end_mask_0, x = q_47_cast);
            tensor<int32, [4]> var_9652_begin_0 = const()[name = tensor<string, []>("op_9652_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_9652_end_0 = const()[name = tensor<string, []>("op_9652_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_9652_end_mask_0 = const()[name = tensor<string, []>("op_9652_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9652_cast = slice_by_index(begin = var_9652_begin_0, end = var_9652_end_0, end_mask = var_9652_end_mask_0, x = q_47_cast);
            tensor<int32, [4]> k_95_perm_0 = const()[name = tensor<string, []>("k_95_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_9659_begin_0 = const()[name = tensor<string, []>("op_9659_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9659_end_0 = const()[name = tensor<string, []>("op_9659_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_9659_end_mask_0 = const()[name = tensor<string, []>("op_9659_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 640]> transpose_8 = transpose(perm = k_95_perm_0, x = k_93_cast);
            tensor<fp16, [2, 77, 1, 64]> var_9659_cast = slice_by_index(begin = var_9659_begin_0, end = var_9659_end_0, end_mask = var_9659_end_mask_0, x = transpose_8);
            tensor<int32, [4]> var_9663_begin_0 = const()[name = tensor<string, []>("op_9663_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_9663_end_0 = const()[name = tensor<string, []>("op_9663_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_9663_end_mask_0 = const()[name = tensor<string, []>("op_9663_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9663_cast = slice_by_index(begin = var_9663_begin_0, end = var_9663_end_0, end_mask = var_9663_end_mask_0, x = transpose_8);
            tensor<int32, [4]> var_9667_begin_0 = const()[name = tensor<string, []>("op_9667_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_9667_end_0 = const()[name = tensor<string, []>("op_9667_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_9667_end_mask_0 = const()[name = tensor<string, []>("op_9667_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9667_cast = slice_by_index(begin = var_9667_begin_0, end = var_9667_end_0, end_mask = var_9667_end_mask_0, x = transpose_8);
            tensor<int32, [4]> var_9671_begin_0 = const()[name = tensor<string, []>("op_9671_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_9671_end_0 = const()[name = tensor<string, []>("op_9671_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_9671_end_mask_0 = const()[name = tensor<string, []>("op_9671_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9671_cast = slice_by_index(begin = var_9671_begin_0, end = var_9671_end_0, end_mask = var_9671_end_mask_0, x = transpose_8);
            tensor<int32, [4]> var_9675_begin_0 = const()[name = tensor<string, []>("op_9675_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_9675_end_0 = const()[name = tensor<string, []>("op_9675_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_9675_end_mask_0 = const()[name = tensor<string, []>("op_9675_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9675_cast = slice_by_index(begin = var_9675_begin_0, end = var_9675_end_0, end_mask = var_9675_end_mask_0, x = transpose_8);
            tensor<int32, [4]> var_9679_begin_0 = const()[name = tensor<string, []>("op_9679_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_9679_end_0 = const()[name = tensor<string, []>("op_9679_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_9679_end_mask_0 = const()[name = tensor<string, []>("op_9679_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9679_cast = slice_by_index(begin = var_9679_begin_0, end = var_9679_end_0, end_mask = var_9679_end_mask_0, x = transpose_8);
            tensor<int32, [4]> var_9683_begin_0 = const()[name = tensor<string, []>("op_9683_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_9683_end_0 = const()[name = tensor<string, []>("op_9683_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_9683_end_mask_0 = const()[name = tensor<string, []>("op_9683_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9683_cast = slice_by_index(begin = var_9683_begin_0, end = var_9683_end_0, end_mask = var_9683_end_mask_0, x = transpose_8);
            tensor<int32, [4]> var_9687_begin_0 = const()[name = tensor<string, []>("op_9687_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_9687_end_0 = const()[name = tensor<string, []>("op_9687_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_9687_end_mask_0 = const()[name = tensor<string, []>("op_9687_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9687_cast = slice_by_index(begin = var_9687_begin_0, end = var_9687_end_0, end_mask = var_9687_end_mask_0, x = transpose_8);
            tensor<int32, [4]> var_9691_begin_0 = const()[name = tensor<string, []>("op_9691_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9691_end_0 = const()[name = tensor<string, []>("op_9691_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_9691_end_mask_0 = const()[name = tensor<string, []>("op_9691_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9691_cast = slice_by_index(begin = var_9691_begin_0, end = var_9691_end_0, end_mask = var_9691_end_mask_0, x = transpose_8);
            tensor<int32, [4]> var_9695_begin_0 = const()[name = tensor<string, []>("op_9695_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_9695_end_0 = const()[name = tensor<string, []>("op_9695_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_9695_end_mask_0 = const()[name = tensor<string, []>("op_9695_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9695_cast = slice_by_index(begin = var_9695_begin_0, end = var_9695_end_0, end_mask = var_9695_end_mask_0, x = transpose_8);
            tensor<int32, [4]> var_9697_begin_0 = const()[name = tensor<string, []>("op_9697_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9697_end_0 = const()[name = tensor<string, []>("op_9697_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_9697_end_mask_0 = const()[name = tensor<string, []>("op_9697_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9697_cast = slice_by_index(begin = var_9697_begin_0, end = var_9697_end_0, end_mask = var_9697_end_mask_0, x = v_47_cast);
            tensor<int32, [4]> var_9701_begin_0 = const()[name = tensor<string, []>("op_9701_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_9701_end_0 = const()[name = tensor<string, []>("op_9701_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_9701_end_mask_0 = const()[name = tensor<string, []>("op_9701_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9701_cast = slice_by_index(begin = var_9701_begin_0, end = var_9701_end_0, end_mask = var_9701_end_mask_0, x = v_47_cast);
            tensor<int32, [4]> var_9705_begin_0 = const()[name = tensor<string, []>("op_9705_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_9705_end_0 = const()[name = tensor<string, []>("op_9705_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_9705_end_mask_0 = const()[name = tensor<string, []>("op_9705_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9705_cast = slice_by_index(begin = var_9705_begin_0, end = var_9705_end_0, end_mask = var_9705_end_mask_0, x = v_47_cast);
            tensor<int32, [4]> var_9709_begin_0 = const()[name = tensor<string, []>("op_9709_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_9709_end_0 = const()[name = tensor<string, []>("op_9709_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_9709_end_mask_0 = const()[name = tensor<string, []>("op_9709_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9709_cast = slice_by_index(begin = var_9709_begin_0, end = var_9709_end_0, end_mask = var_9709_end_mask_0, x = v_47_cast);
            tensor<int32, [4]> var_9713_begin_0 = const()[name = tensor<string, []>("op_9713_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_9713_end_0 = const()[name = tensor<string, []>("op_9713_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_9713_end_mask_0 = const()[name = tensor<string, []>("op_9713_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9713_cast = slice_by_index(begin = var_9713_begin_0, end = var_9713_end_0, end_mask = var_9713_end_mask_0, x = v_47_cast);
            tensor<int32, [4]> var_9717_begin_0 = const()[name = tensor<string, []>("op_9717_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_9717_end_0 = const()[name = tensor<string, []>("op_9717_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_9717_end_mask_0 = const()[name = tensor<string, []>("op_9717_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9717_cast = slice_by_index(begin = var_9717_begin_0, end = var_9717_end_0, end_mask = var_9717_end_mask_0, x = v_47_cast);
            tensor<int32, [4]> var_9721_begin_0 = const()[name = tensor<string, []>("op_9721_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_9721_end_0 = const()[name = tensor<string, []>("op_9721_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_9721_end_mask_0 = const()[name = tensor<string, []>("op_9721_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9721_cast = slice_by_index(begin = var_9721_begin_0, end = var_9721_end_0, end_mask = var_9721_end_mask_0, x = v_47_cast);
            tensor<int32, [4]> var_9725_begin_0 = const()[name = tensor<string, []>("op_9725_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_9725_end_0 = const()[name = tensor<string, []>("op_9725_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_9725_end_mask_0 = const()[name = tensor<string, []>("op_9725_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9725_cast = slice_by_index(begin = var_9725_begin_0, end = var_9725_end_0, end_mask = var_9725_end_mask_0, x = v_47_cast);
            tensor<int32, [4]> var_9729_begin_0 = const()[name = tensor<string, []>("op_9729_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_9729_end_0 = const()[name = tensor<string, []>("op_9729_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_9729_end_mask_0 = const()[name = tensor<string, []>("op_9729_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9729_cast = slice_by_index(begin = var_9729_begin_0, end = var_9729_end_0, end_mask = var_9729_end_mask_0, x = v_47_cast);
            tensor<int32, [4]> var_9733_begin_0 = const()[name = tensor<string, []>("op_9733_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_9733_end_0 = const()[name = tensor<string, []>("op_9733_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_9733_end_mask_0 = const()[name = tensor<string, []>("op_9733_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9733_cast = slice_by_index(begin = var_9733_begin_0, end = var_9733_end_0, end_mask = var_9733_end_mask_0, x = v_47_cast);
            tensor<string, []> var_9737_equation_0 = const()[name = tensor<string, []>("op_9737_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9737_cast = einsum(equation = var_9737_equation_0, values = (var_9659_cast, var_9616_cast));
            tensor<fp16, []> var_9738_to_fp16 = const()[name = tensor<string, []>("op_9738_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_661_cast = mul(x = var_9737_cast, y = var_9738_to_fp16);
            tensor<string, []> var_9741_equation_0 = const()[name = tensor<string, []>("op_9741_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9741_cast = einsum(equation = var_9741_equation_0, values = (var_9663_cast, var_9620_cast));
            tensor<fp16, []> var_9742_to_fp16 = const()[name = tensor<string, []>("op_9742_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_663_cast = mul(x = var_9741_cast, y = var_9742_to_fp16);
            tensor<string, []> var_9745_equation_0 = const()[name = tensor<string, []>("op_9745_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9745_cast = einsum(equation = var_9745_equation_0, values = (var_9667_cast, var_9624_cast));
            tensor<fp16, []> var_9746_to_fp16 = const()[name = tensor<string, []>("op_9746_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_665_cast = mul(x = var_9745_cast, y = var_9746_to_fp16);
            tensor<string, []> var_9749_equation_0 = const()[name = tensor<string, []>("op_9749_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9749_cast = einsum(equation = var_9749_equation_0, values = (var_9671_cast, var_9628_cast));
            tensor<fp16, []> var_9750_to_fp16 = const()[name = tensor<string, []>("op_9750_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_667_cast = mul(x = var_9749_cast, y = var_9750_to_fp16);
            tensor<string, []> var_9753_equation_0 = const()[name = tensor<string, []>("op_9753_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9753_cast = einsum(equation = var_9753_equation_0, values = (var_9675_cast, var_9632_cast));
            tensor<fp16, []> var_9754_to_fp16 = const()[name = tensor<string, []>("op_9754_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_669_cast = mul(x = var_9753_cast, y = var_9754_to_fp16);
            tensor<string, []> var_9757_equation_0 = const()[name = tensor<string, []>("op_9757_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9757_cast = einsum(equation = var_9757_equation_0, values = (var_9679_cast, var_9636_cast));
            tensor<fp16, []> var_9758_to_fp16 = const()[name = tensor<string, []>("op_9758_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_671_cast = mul(x = var_9757_cast, y = var_9758_to_fp16);
            tensor<string, []> var_9761_equation_0 = const()[name = tensor<string, []>("op_9761_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9761_cast = einsum(equation = var_9761_equation_0, values = (var_9683_cast, var_9640_cast));
            tensor<fp16, []> var_9762_to_fp16 = const()[name = tensor<string, []>("op_9762_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_673_cast = mul(x = var_9761_cast, y = var_9762_to_fp16);
            tensor<string, []> var_9765_equation_0 = const()[name = tensor<string, []>("op_9765_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9765_cast = einsum(equation = var_9765_equation_0, values = (var_9687_cast, var_9644_cast));
            tensor<fp16, []> var_9766_to_fp16 = const()[name = tensor<string, []>("op_9766_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_675_cast = mul(x = var_9765_cast, y = var_9766_to_fp16);
            tensor<string, []> var_9769_equation_0 = const()[name = tensor<string, []>("op_9769_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9769_cast = einsum(equation = var_9769_equation_0, values = (var_9691_cast, var_9648_cast));
            tensor<fp16, []> var_9770_to_fp16 = const()[name = tensor<string, []>("op_9770_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_677_cast = mul(x = var_9769_cast, y = var_9770_to_fp16);
            tensor<string, []> var_9773_equation_0 = const()[name = tensor<string, []>("op_9773_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_9773_cast = einsum(equation = var_9773_equation_0, values = (var_9695_cast, var_9652_cast));
            tensor<fp16, []> var_9774_to_fp16 = const()[name = tensor<string, []>("op_9774_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_679_cast = mul(x = var_9773_cast, y = var_9774_to_fp16);
            tensor<fp16, [2, 77, 1, 1024]> var_9776_cast = softmax(axis = var_8627, x = aw_661_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_9777_cast = softmax(axis = var_8627, x = aw_663_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_9778_cast = softmax(axis = var_8627, x = aw_665_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_9779_cast = softmax(axis = var_8627, x = aw_667_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_9780_cast = softmax(axis = var_8627, x = aw_669_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_9781_cast = softmax(axis = var_8627, x = aw_671_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_9782_cast = softmax(axis = var_8627, x = aw_673_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_9783_cast = softmax(axis = var_8627, x = aw_675_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_9784_cast = softmax(axis = var_8627, x = aw_677_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_9785_cast = softmax(axis = var_8627, x = aw_679_cast);
            tensor<string, []> var_9787_equation_0 = const()[name = tensor<string, []>("op_9787_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9787_cast = einsum(equation = var_9787_equation_0, values = (var_9697_cast, var_9776_cast));
            tensor<string, []> var_9789_equation_0 = const()[name = tensor<string, []>("op_9789_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9789_cast = einsum(equation = var_9789_equation_0, values = (var_9701_cast, var_9777_cast));
            tensor<string, []> var_9791_equation_0 = const()[name = tensor<string, []>("op_9791_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9791_cast = einsum(equation = var_9791_equation_0, values = (var_9705_cast, var_9778_cast));
            tensor<string, []> var_9793_equation_0 = const()[name = tensor<string, []>("op_9793_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9793_cast = einsum(equation = var_9793_equation_0, values = (var_9709_cast, var_9779_cast));
            tensor<string, []> var_9795_equation_0 = const()[name = tensor<string, []>("op_9795_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9795_cast = einsum(equation = var_9795_equation_0, values = (var_9713_cast, var_9780_cast));
            tensor<string, []> var_9797_equation_0 = const()[name = tensor<string, []>("op_9797_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9797_cast = einsum(equation = var_9797_equation_0, values = (var_9717_cast, var_9781_cast));
            tensor<string, []> var_9799_equation_0 = const()[name = tensor<string, []>("op_9799_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9799_cast = einsum(equation = var_9799_equation_0, values = (var_9721_cast, var_9782_cast));
            tensor<string, []> var_9801_equation_0 = const()[name = tensor<string, []>("op_9801_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9801_cast = einsum(equation = var_9801_equation_0, values = (var_9725_cast, var_9783_cast));
            tensor<string, []> var_9803_equation_0 = const()[name = tensor<string, []>("op_9803_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9803_cast = einsum(equation = var_9803_equation_0, values = (var_9729_cast, var_9784_cast));
            tensor<string, []> var_9805_equation_0 = const()[name = tensor<string, []>("op_9805_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_9805_cast = einsum(equation = var_9805_equation_0, values = (var_9733_cast, var_9785_cast));
            tensor<bool, []> input_411_interleave_0 = const()[name = tensor<string, []>("input_411_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 1024]> input_411_cast = concat(axis = var_8627, interleave = input_411_interleave_0, values = (var_9787_cast, var_9789_cast, var_9791_cast, var_9793_cast, var_9795_cast, var_9797_cast, var_9799_cast, var_9801_cast, var_9803_cast, var_9805_cast));
            tensor<int32, [2]> var_9811 = const()[name = tensor<string, []>("op_9811"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9813 = const()[name = tensor<string, []>("op_9813"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_9815_pad_type_0 = const()[name = tensor<string, []>("op_9815_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_9815_pad_0 = const()[name = tensor<string, []>("op_9815_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1634254144)))];
            tensor<fp16, [640]> up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1635073408)))];
            tensor<fp16, [2, 640, 1, 1024]> var_9815_cast = conv(bias = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_9813, groups = var_8627, pad = var_9815_pad_0, pad_type = var_9815_pad_type_0, strides = var_9811, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_411_cast);
            tensor<fp16, [2, 640, 1, 1024]> inputs_71_cast = add(x = var_9815_cast, y = inputs_69_cast);
            tensor<int32, [1]> var_9819 = const()[name = tensor<string, []>("op_9819"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_71_cast = reduce_mean(axes = var_9819, keep_dims = var_8622, x = inputs_71_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_71_cast = sub(x = inputs_71_cast, y = channels_mean_71_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_sq_71_cast = mul(x = zero_mean_71_cast, y = zero_mean_71_cast);
            tensor<int32, [1]> var_9823 = const()[name = tensor<string, []>("op_9823"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_9824_cast = reduce_mean(axes = var_9823, keep_dims = var_8622, x = zero_mean_sq_71_cast);
            tensor<fp16, []> var_9825_to_fp16 = const()[name = tensor<string, []>("op_9825_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_9826_cast = add(x = var_9824_cast, y = var_9825_to_fp16);
            tensor<fp16, []> denom_71_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_71_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_71_cast = rsqrt(epsilon = denom_71_epsilon_0_to_fp16, x = var_9826_cast);
            tensor<fp16, [2, 640, 1, 1024]> out_71_cast = mul(x = zero_mean_71_cast, y = denom_71_cast);
            tensor<fp16, [1, 640, 1, 1]> var_9830_to_fp16 = const()[name = tensor<string, []>("op_9830_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1635074752)))];
            tensor<fp16, [2, 640, 1, 1024]> var_9831_cast = add(x = out_71_cast, y = var_9830_to_fp16);
            tensor<fp16, [1, 640, 1, 1]> var_9833_to_fp16 = const()[name = tensor<string, []>("op_9833_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1635076096)))];
            tensor<fp16, [2, 640, 1, 1024]> input_413_cast = mul(x = var_9831_cast, y = var_9833_to_fp16);
            tensor<int32, [2]> var_9841 = const()[name = tensor<string, []>("op_9841"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9843 = const()[name = tensor<string, []>("op_9843"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_9845_pad_type_0 = const()[name = tensor<string, []>("op_9845_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_9845_pad_0 = const()[name = tensor<string, []>("op_9845_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [5120, 640, 1, 1]> up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [5120, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1635077440)))];
            tensor<fp16, [5120]> up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1641631104)))];
            tensor<fp16, [2, 5120, 1, 1024]> var_9845_cast = conv(bias = up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_9843, groups = var_8627, pad = var_9845_pad_0, pad_type = var_9845_pad_type_0, strides = var_9841, weight = up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_413_cast);
            tensor<int32, [2]> var_9846_split_sizes_0 = const()[name = tensor<string, []>("op_9846_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_9846_axis_0 = const()[name = tensor<string, []>("op_9846_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 1024]> var_9846_cast_0, tensor<fp16, [2, 2560, 1, 1024]> var_9846_cast_1 = split(axis = var_9846_axis_0, split_sizes = var_9846_split_sizes_0, x = var_9845_cast);
            tensor<string, []> var_9848_mode_0 = const()[name = tensor<string, []>("op_9848_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 1024]> var_9848_cast = gelu(mode = var_9848_mode_0, x = var_9846_cast_1);
            tensor<fp16, [2, 2560, 1, 1024]> input_415_cast = mul(x = var_9846_cast_0, y = var_9848_cast);
            tensor<int32, [2]> var_9852 = const()[name = tensor<string, []>("op_9852"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9854 = const()[name = tensor<string, []>("op_9854"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_9856_pad_type_0 = const()[name = tensor<string, []>("op_9856_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_9856_pad_0 = const()[name = tensor<string, []>("op_9856_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2560, 1, 1]> up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [640, 2560, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1641641408)))];
            tensor<fp16, [640]> up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1644918272)))];
            tensor<fp16, [2, 640, 1, 1024]> var_9856_cast = conv(bias = up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_9854, groups = var_8627, pad = var_9856_pad_0, pad_type = var_9856_pad_type_0, strides = var_9852, weight = up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_415_cast);
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_253_cast = add(x = var_9856_cast, y = inputs_71_cast);
            tensor<int32, [4]> var_9858 = const()[name = tensor<string, []>("op_9858"), val = tensor<int32, [4]>([2, 640, 32, 32])];
            tensor<fp16, [2, 640, 32, 32]> input_417_cast = reshape(shape = var_9858, x = hidden_states_253_cast);
            tensor<int32, [2]> var_9862 = const()[name = tensor<string, []>("op_9862"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9864 = const()[name = tensor<string, []>("op_9864"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_255_pad_type_0 = const()[name = tensor<string, []>("hidden_states_255_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_255_pad_0 = const()[name = tensor<string, []>("hidden_states_255_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_2_attentions_1_proj_out_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_1_proj_out_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1644919616)))];
            tensor<fp16, [640]> up_blocks_2_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_1_proj_out_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1645738880)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_255_cast = conv(bias = up_blocks_2_attentions_1_proj_out_bias_to_fp16, dilations = var_9864, groups = var_8627, pad = hidden_states_255_pad_0, pad_type = hidden_states_255_pad_type_0, strides = var_9862, weight = up_blocks_2_attentions_1_proj_out_weight_to_fp16, x = input_417_cast);
            tensor<fp16, [2, 640, 32, 32]> hidden_states_257_cast = add(x = hidden_states_255_cast, y = hidden_states_243_cast);
            tensor<bool, []> input_419_interleave_0 = const()[name = tensor<string, []>("input_419_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 960, 32, 32]> input_419_cast = concat(axis = var_8627, interleave = input_419_interleave_0, values = (hidden_states_257_cast, input_63_cast));
            tensor<int32, [5]> reshape_192_shape_0 = const()[name = tensor<string, []>("reshape_192_shape_0"), val = tensor<int32, [5]>([2, 32, 30, 32, 32])];
            tensor<fp16, [2, 32, 30, 32, 32]> reshape_192_cast = reshape(shape = reshape_192_shape_0, x = input_419_cast);
            tensor<int32, [3]> reduce_mean_144_axes_0 = const()[name = tensor<string, []>("reduce_mean_144_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_144_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_144_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_144_cast = reduce_mean(axes = reduce_mean_144_axes_0, keep_dims = reduce_mean_144_keep_dims_0, x = reshape_192_cast);
            tensor<fp16, [2, 32, 30, 32, 32]> sub_96_cast = sub(x = reshape_192_cast, y = reduce_mean_144_cast);
            tensor<fp16, [2, 32, 30, 32, 32]> square_48_cast = square(x = sub_96_cast);
            tensor<int32, [3]> reduce_mean_146_axes_0 = const()[name = tensor<string, []>("reduce_mean_146_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_146_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_146_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_146_cast = reduce_mean(axes = reduce_mean_146_axes_0, keep_dims = reduce_mean_146_keep_dims_0, x = square_48_cast);
            tensor<fp16, []> add_96_y_0_to_fp16 = const()[name = tensor<string, []>("add_96_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_96_cast = add(x = reduce_mean_146_cast, y = add_96_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_48_cast = sqrt(x = add_96_cast);
            tensor<fp16, [2, 32, 30, 32, 32]> real_div_48_cast = real_div(x = sub_96_cast, y = sqrt_48_cast);
            tensor<int32, [4]> reshape_193_shape_0 = const()[name = tensor<string, []>("reshape_193_shape_0"), val = tensor<int32, [4]>([2, 960, 32, 32])];
            tensor<fp16, [2, 960, 32, 32]> reshape_193_cast = reshape(shape = reshape_193_shape_0, x = real_div_48_cast);
            tensor<fp16, [960]> add_97_mean_0_to_fp16 = const()[name = tensor<string, []>("add_97_mean_0_to_fp16"), val = tensor<fp16, [960]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1645740224)))];
            tensor<fp16, [960]> add_97_variance_0_to_fp16 = const()[name = tensor<string, []>("add_97_variance_0_to_fp16"), val = tensor<fp16, [960]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1645742208)))];
            tensor<fp16, [960]> add_97_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_97_gamma_0_to_fp16"), val = tensor<fp16, [960]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1645744192)))];
            tensor<fp16, [960]> add_97_beta_0_to_fp16 = const()[name = tensor<string, []>("add_97_beta_0_to_fp16"), val = tensor<fp16, [960]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1645746176)))];
            tensor<fp16, []> add_97_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_97_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 960, 32, 32]> add_97_cast = batch_norm(beta = add_97_beta_0_to_fp16, epsilon = add_97_epsilon_0_to_fp16, gamma = add_97_gamma_0_to_fp16, mean = add_97_mean_0_to_fp16, variance = add_97_variance_0_to_fp16, x = reshape_193_cast);
            tensor<fp16, [2, 960, 32, 32]> input_423_cast = silu(x = add_97_cast);
            tensor<int32, [2]> var_9882 = const()[name = tensor<string, []>("op_9882"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9884 = const()[name = tensor<string, []>("op_9884"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_259_pad_type_0 = const()[name = tensor<string, []>("hidden_states_259_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_259_pad_0 = const()[name = tensor<string, []>("hidden_states_259_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [640, 960, 3, 3]> up_blocks_2_resnets_2_conv1_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_2_conv1_weight_to_fp16"), val = tensor<fp16, [640, 960, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1645748160)))];
            tensor<fp16, [640]> up_blocks_2_resnets_2_conv1_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_2_conv1_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1656807424)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_259_cast = conv(bias = up_blocks_2_resnets_2_conv1_bias_to_fp16, dilations = var_9884, groups = var_8627, pad = hidden_states_259_pad_0, pad_type = hidden_states_259_pad_type_0, strides = var_9882, weight = up_blocks_2_resnets_2_conv1_weight_to_fp16, x = input_423_cast);
            tensor<int32, [2]> var_9890 = const()[name = tensor<string, []>("op_9890"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9892 = const()[name = tensor<string, []>("op_9892"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_37_pad_type_0 = const()[name = tensor<string, []>("temb_37_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_37_pad_0 = const()[name = tensor<string, []>("temb_37_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1280, 1, 1]> up_blocks_2_resnets_2_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_2_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [640, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1656808768)))];
            tensor<fp16, [640]> up_blocks_2_resnets_2_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_2_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1658447232)))];
            tensor<fp16, [2, 640, 1, 1]> temb_37_cast = conv(bias = up_blocks_2_resnets_2_time_emb_proj_bias_to_fp16, dilations = var_9892, groups = var_8627, pad = temb_37_pad_0, pad_type = temb_37_pad_type_0, strides = var_9890, weight = up_blocks_2_resnets_2_time_emb_proj_weight_to_fp16, x = input_15_cast);
            tensor<fp16, [2, 640, 32, 32]> input_427_cast = add(x = hidden_states_259_cast, y = temb_37_cast);
            tensor<int32, [5]> reshape_196_shape_0 = const()[name = tensor<string, []>("reshape_196_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 32, 32])];
            tensor<fp16, [2, 32, 20, 32, 32]> reshape_196_cast = reshape(shape = reshape_196_shape_0, x = input_427_cast);
            tensor<int32, [3]> reduce_mean_147_axes_0 = const()[name = tensor<string, []>("reduce_mean_147_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_147_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_147_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_147_cast = reduce_mean(axes = reduce_mean_147_axes_0, keep_dims = reduce_mean_147_keep_dims_0, x = reshape_196_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> sub_98_cast = sub(x = reshape_196_cast, y = reduce_mean_147_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> square_49_cast = square(x = sub_98_cast);
            tensor<int32, [3]> reduce_mean_149_axes_0 = const()[name = tensor<string, []>("reduce_mean_149_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_149_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_149_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_149_cast = reduce_mean(axes = reduce_mean_149_axes_0, keep_dims = reduce_mean_149_keep_dims_0, x = square_49_cast);
            tensor<fp16, []> add_98_y_0_to_fp16 = const()[name = tensor<string, []>("add_98_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_98_cast = add(x = reduce_mean_149_cast, y = add_98_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_49_cast = sqrt(x = add_98_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> real_div_49_cast = real_div(x = sub_98_cast, y = sqrt_49_cast);
            tensor<int32, [4]> reshape_197_shape_0 = const()[name = tensor<string, []>("reshape_197_shape_0"), val = tensor<int32, [4]>([2, 640, 32, 32])];
            tensor<fp16, [2, 640, 32, 32]> reshape_197_cast = reshape(shape = reshape_197_shape_0, x = real_div_49_cast);
            tensor<fp16, [640]> add_99_mean_0_to_fp16 = const()[name = tensor<string, []>("add_99_mean_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1658448576)))];
            tensor<fp16, [640]> add_99_variance_0_to_fp16 = const()[name = tensor<string, []>("add_99_variance_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1658449920)))];
            tensor<fp16, [640]> add_99_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_99_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1658451264)))];
            tensor<fp16, [640]> add_99_beta_0_to_fp16 = const()[name = tensor<string, []>("add_99_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1658452608)))];
            tensor<fp16, []> add_99_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_99_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 32, 32]> add_99_cast = batch_norm(beta = add_99_beta_0_to_fp16, epsilon = add_99_epsilon_0_to_fp16, gamma = add_99_gamma_0_to_fp16, mean = add_99_mean_0_to_fp16, variance = add_99_variance_0_to_fp16, x = reshape_197_cast);
            tensor<fp16, [2, 640, 32, 32]> input_431_cast = silu(x = add_99_cast);
            tensor<int32, [2]> var_9902 = const()[name = tensor<string, []>("op_9902"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9904 = const()[name = tensor<string, []>("op_9904"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_261_pad_type_0 = const()[name = tensor<string, []>("hidden_states_261_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_261_pad_0 = const()[name = tensor<string, []>("hidden_states_261_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [640, 640, 3, 3]> up_blocks_2_resnets_2_conv2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_2_conv2_weight_to_fp16"), val = tensor<fp16, [640, 640, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1658453952)))];
            tensor<fp16, [640]> up_blocks_2_resnets_2_conv2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_2_conv2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1665826816)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_261_cast = conv(bias = up_blocks_2_resnets_2_conv2_bias_to_fp16, dilations = var_9904, groups = var_8627, pad = hidden_states_261_pad_0, pad_type = hidden_states_261_pad_type_0, strides = var_9902, weight = up_blocks_2_resnets_2_conv2_weight_to_fp16, x = input_431_cast);
            tensor<int32, [2]> var_9909 = const()[name = tensor<string, []>("op_9909"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9911 = const()[name = tensor<string, []>("op_9911"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> x_21_pad_type_0 = const()[name = tensor<string, []>("x_21_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> x_21_pad_0 = const()[name = tensor<string, []>("x_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 960, 1, 1]> up_blocks_2_resnets_2_conv_shortcut_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_2_conv_shortcut_weight_to_fp16"), val = tensor<fp16, [640, 960, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1665828160)))];
            tensor<fp16, [640]> up_blocks_2_resnets_2_conv_shortcut_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_resnets_2_conv_shortcut_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1667057024)))];
            tensor<fp16, [2, 640, 32, 32]> x_21_cast = conv(bias = up_blocks_2_resnets_2_conv_shortcut_bias_to_fp16, dilations = var_9911, groups = var_8627, pad = x_21_pad_0, pad_type = x_21_pad_type_0, strides = var_9909, weight = up_blocks_2_resnets_2_conv_shortcut_weight_to_fp16, x = input_419_cast);
            tensor<fp16, [2, 640, 32, 32]> hidden_states_263_cast = add(x = x_21_cast, y = hidden_states_261_cast);
            tensor<int32, [5]> reshape_200_shape_0 = const()[name = tensor<string, []>("reshape_200_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 32, 32])];
            tensor<fp16, [2, 32, 20, 32, 32]> reshape_200_cast = reshape(shape = reshape_200_shape_0, x = hidden_states_263_cast);
            tensor<int32, [3]> reduce_mean_150_axes_0 = const()[name = tensor<string, []>("reduce_mean_150_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_150_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_150_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_150_cast = reduce_mean(axes = reduce_mean_150_axes_0, keep_dims = reduce_mean_150_keep_dims_0, x = reshape_200_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> sub_100_cast = sub(x = reshape_200_cast, y = reduce_mean_150_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> square_50_cast = square(x = sub_100_cast);
            tensor<int32, [3]> reduce_mean_152_axes_0 = const()[name = tensor<string, []>("reduce_mean_152_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_152_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_152_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_152_cast = reduce_mean(axes = reduce_mean_152_axes_0, keep_dims = reduce_mean_152_keep_dims_0, x = square_50_cast);
            tensor<fp16, []> add_100_y_0_to_fp16 = const()[name = tensor<string, []>("add_100_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_100_cast = add(x = reduce_mean_152_cast, y = add_100_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_50_cast = sqrt(x = add_100_cast);
            tensor<fp16, [2, 32, 20, 32, 32]> real_div_50_cast = real_div(x = sub_100_cast, y = sqrt_50_cast);
            tensor<int32, [4]> reshape_201_shape_0 = const()[name = tensor<string, []>("reshape_201_shape_0"), val = tensor<int32, [4]>([2, 640, 32, 32])];
            tensor<fp16, [2, 640, 32, 32]> reshape_201_cast = reshape(shape = reshape_201_shape_0, x = real_div_50_cast);
            tensor<fp16, [640]> add_101_mean_0_to_fp16 = const()[name = tensor<string, []>("add_101_mean_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1667058368)))];
            tensor<fp16, [640]> add_101_variance_0_to_fp16 = const()[name = tensor<string, []>("add_101_variance_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1667059712)))];
            tensor<fp16, [640]> add_101_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_101_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1667061056)))];
            tensor<fp16, [640]> add_101_beta_0_to_fp16 = const()[name = tensor<string, []>("add_101_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1667062400)))];
            tensor<fp16, []> add_101_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_101_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 32, 32]> add_101_cast = batch_norm(beta = add_101_beta_0_to_fp16, epsilon = add_101_epsilon_0_to_fp16, gamma = add_101_gamma_0_to_fp16, mean = add_101_mean_0_to_fp16, variance = add_101_variance_0_to_fp16, x = reshape_201_cast);
            tensor<int32, [2]> var_9931 = const()[name = tensor<string, []>("op_9931"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9933 = const()[name = tensor<string, []>("op_9933"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_265_pad_type_0 = const()[name = tensor<string, []>("hidden_states_265_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_265_pad_0 = const()[name = tensor<string, []>("hidden_states_265_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_2_attentions_2_proj_in_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_2_proj_in_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1667063744)))];
            tensor<fp16, [640]> up_blocks_2_attentions_2_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_2_proj_in_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1667883008)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_265_cast = conv(bias = up_blocks_2_attentions_2_proj_in_bias_to_fp16, dilations = var_9933, groups = var_8627, pad = hidden_states_265_pad_0, pad_type = hidden_states_265_pad_type_0, strides = var_9931, weight = up_blocks_2_attentions_2_proj_in_weight_to_fp16, x = add_101_cast);
            tensor<int32, [4]> var_9938 = const()[name = tensor<string, []>("op_9938"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<fp16, [2, 640, 1, 1024]> inputs_73_cast = reshape(shape = var_9938, x = hidden_states_265_cast);
            tensor<int32, [1]> var_9948 = const()[name = tensor<string, []>("op_9948"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_73_cast = reduce_mean(axes = var_9948, keep_dims = var_8622, x = inputs_73_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_73_cast = sub(x = inputs_73_cast, y = channels_mean_73_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_sq_73_cast = mul(x = zero_mean_73_cast, y = zero_mean_73_cast);
            tensor<int32, [1]> var_9952 = const()[name = tensor<string, []>("op_9952"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_9953_cast = reduce_mean(axes = var_9952, keep_dims = var_8622, x = zero_mean_sq_73_cast);
            tensor<fp16, []> var_9954_to_fp16 = const()[name = tensor<string, []>("op_9954_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_9955_cast = add(x = var_9953_cast, y = var_9954_to_fp16);
            tensor<fp16, []> denom_73_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_73_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_73_cast = rsqrt(epsilon = denom_73_epsilon_0_to_fp16, x = var_9955_cast);
            tensor<fp16, [2, 640, 1, 1024]> out_73_cast = mul(x = zero_mean_73_cast, y = denom_73_cast);
            tensor<fp16, [1, 640, 1, 1]> var_9959_to_fp16 = const()[name = tensor<string, []>("op_9959_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1667884352)))];
            tensor<fp16, [2, 640, 1, 1024]> var_9960_cast = add(x = out_73_cast, y = var_9959_to_fp16);
            tensor<fp16, [1, 640, 1, 1]> var_9962_to_fp16 = const()[name = tensor<string, []>("op_9962_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1667885696)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_267_cast = mul(x = var_9960_cast, y = var_9962_to_fp16);
            tensor<int32, [2]> var_9969 = const()[name = tensor<string, []>("op_9969"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9971 = const()[name = tensor<string, []>("op_9971"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_49_pad_type_0 = const()[name = tensor<string, []>("q_49_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_49_pad_0 = const()[name = tensor<string, []>("q_49_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1667887040)))];
            tensor<fp16, [2, 640, 1, 1024]> q_49_cast = conv(dilations = var_9971, groups = var_8627, pad = q_49_pad_0, pad_type = q_49_pad_type_0, strides = var_9969, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_267_cast);
            tensor<int32, [2]> var_9975 = const()[name = tensor<string, []>("op_9975"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9977 = const()[name = tensor<string, []>("op_9977"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_97_pad_type_0 = const()[name = tensor<string, []>("k_97_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_97_pad_0 = const()[name = tensor<string, []>("k_97_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1668706304)))];
            tensor<fp16, [2, 640, 1, 1024]> k_97_cast = conv(dilations = var_9977, groups = var_8627, pad = k_97_pad_0, pad_type = k_97_pad_type_0, strides = var_9975, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_267_cast);
            tensor<int32, [2]> var_9981 = const()[name = tensor<string, []>("op_9981"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9983 = const()[name = tensor<string, []>("op_9983"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_49_pad_type_0 = const()[name = tensor<string, []>("v_49_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_49_pad_0 = const()[name = tensor<string, []>("v_49_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1669525568)))];
            tensor<fp16, [2, 640, 1, 1024]> v_49_cast = conv(dilations = var_9983, groups = var_8627, pad = v_49_pad_0, pad_type = v_49_pad_type_0, strides = var_9981, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_267_cast);
            tensor<int32, [4]> var_9987_begin_0 = const()[name = tensor<string, []>("op_9987_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9987_end_0 = const()[name = tensor<string, []>("op_9987_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_9987_end_mask_0 = const()[name = tensor<string, []>("op_9987_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9987_cast = slice_by_index(begin = var_9987_begin_0, end = var_9987_end_0, end_mask = var_9987_end_mask_0, x = q_49_cast);
            tensor<int32, [4]> var_9991_begin_0 = const()[name = tensor<string, []>("op_9991_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_9991_end_0 = const()[name = tensor<string, []>("op_9991_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_9991_end_mask_0 = const()[name = tensor<string, []>("op_9991_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9991_cast = slice_by_index(begin = var_9991_begin_0, end = var_9991_end_0, end_mask = var_9991_end_mask_0, x = q_49_cast);
            tensor<int32, [4]> var_9995_begin_0 = const()[name = tensor<string, []>("op_9995_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_9995_end_0 = const()[name = tensor<string, []>("op_9995_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_9995_end_mask_0 = const()[name = tensor<string, []>("op_9995_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9995_cast = slice_by_index(begin = var_9995_begin_0, end = var_9995_end_0, end_mask = var_9995_end_mask_0, x = q_49_cast);
            tensor<int32, [4]> var_9999_begin_0 = const()[name = tensor<string, []>("op_9999_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_9999_end_0 = const()[name = tensor<string, []>("op_9999_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_9999_end_mask_0 = const()[name = tensor<string, []>("op_9999_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_9999_cast = slice_by_index(begin = var_9999_begin_0, end = var_9999_end_0, end_mask = var_9999_end_mask_0, x = q_49_cast);
            tensor<int32, [4]> var_10003_begin_0 = const()[name = tensor<string, []>("op_10003_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_10003_end_0 = const()[name = tensor<string, []>("op_10003_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_10003_end_mask_0 = const()[name = tensor<string, []>("op_10003_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10003_cast = slice_by_index(begin = var_10003_begin_0, end = var_10003_end_0, end_mask = var_10003_end_mask_0, x = q_49_cast);
            tensor<int32, [4]> var_10007_begin_0 = const()[name = tensor<string, []>("op_10007_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_10007_end_0 = const()[name = tensor<string, []>("op_10007_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_10007_end_mask_0 = const()[name = tensor<string, []>("op_10007_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10007_cast = slice_by_index(begin = var_10007_begin_0, end = var_10007_end_0, end_mask = var_10007_end_mask_0, x = q_49_cast);
            tensor<int32, [4]> var_10011_begin_0 = const()[name = tensor<string, []>("op_10011_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_10011_end_0 = const()[name = tensor<string, []>("op_10011_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_10011_end_mask_0 = const()[name = tensor<string, []>("op_10011_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10011_cast = slice_by_index(begin = var_10011_begin_0, end = var_10011_end_0, end_mask = var_10011_end_mask_0, x = q_49_cast);
            tensor<int32, [4]> var_10015_begin_0 = const()[name = tensor<string, []>("op_10015_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_10015_end_0 = const()[name = tensor<string, []>("op_10015_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_10015_end_mask_0 = const()[name = tensor<string, []>("op_10015_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10015_cast = slice_by_index(begin = var_10015_begin_0, end = var_10015_end_0, end_mask = var_10015_end_mask_0, x = q_49_cast);
            tensor<int32, [4]> var_10019_begin_0 = const()[name = tensor<string, []>("op_10019_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_10019_end_0 = const()[name = tensor<string, []>("op_10019_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_10019_end_mask_0 = const()[name = tensor<string, []>("op_10019_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10019_cast = slice_by_index(begin = var_10019_begin_0, end = var_10019_end_0, end_mask = var_10019_end_mask_0, x = q_49_cast);
            tensor<int32, [4]> var_10023_begin_0 = const()[name = tensor<string, []>("op_10023_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_10023_end_0 = const()[name = tensor<string, []>("op_10023_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_10023_end_mask_0 = const()[name = tensor<string, []>("op_10023_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10023_cast = slice_by_index(begin = var_10023_begin_0, end = var_10023_end_0, end_mask = var_10023_end_mask_0, x = q_49_cast);
            tensor<int32, [4]> k_99_perm_0 = const()[name = tensor<string, []>("k_99_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_10030_begin_0 = const()[name = tensor<string, []>("op_10030_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10030_end_0 = const()[name = tensor<string, []>("op_10030_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_10030_end_mask_0 = const()[name = tensor<string, []>("op_10030_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 640]> transpose_7 = transpose(perm = k_99_perm_0, x = k_97_cast);
            tensor<fp16, [2, 1024, 1, 64]> var_10030_cast = slice_by_index(begin = var_10030_begin_0, end = var_10030_end_0, end_mask = var_10030_end_mask_0, x = transpose_7);
            tensor<int32, [4]> var_10034_begin_0 = const()[name = tensor<string, []>("op_10034_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_10034_end_0 = const()[name = tensor<string, []>("op_10034_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_10034_end_mask_0 = const()[name = tensor<string, []>("op_10034_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10034_cast = slice_by_index(begin = var_10034_begin_0, end = var_10034_end_0, end_mask = var_10034_end_mask_0, x = transpose_7);
            tensor<int32, [4]> var_10038_begin_0 = const()[name = tensor<string, []>("op_10038_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_10038_end_0 = const()[name = tensor<string, []>("op_10038_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_10038_end_mask_0 = const()[name = tensor<string, []>("op_10038_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10038_cast = slice_by_index(begin = var_10038_begin_0, end = var_10038_end_0, end_mask = var_10038_end_mask_0, x = transpose_7);
            tensor<int32, [4]> var_10042_begin_0 = const()[name = tensor<string, []>("op_10042_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_10042_end_0 = const()[name = tensor<string, []>("op_10042_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_10042_end_mask_0 = const()[name = tensor<string, []>("op_10042_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10042_cast = slice_by_index(begin = var_10042_begin_0, end = var_10042_end_0, end_mask = var_10042_end_mask_0, x = transpose_7);
            tensor<int32, [4]> var_10046_begin_0 = const()[name = tensor<string, []>("op_10046_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_10046_end_0 = const()[name = tensor<string, []>("op_10046_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_10046_end_mask_0 = const()[name = tensor<string, []>("op_10046_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10046_cast = slice_by_index(begin = var_10046_begin_0, end = var_10046_end_0, end_mask = var_10046_end_mask_0, x = transpose_7);
            tensor<int32, [4]> var_10050_begin_0 = const()[name = tensor<string, []>("op_10050_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_10050_end_0 = const()[name = tensor<string, []>("op_10050_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_10050_end_mask_0 = const()[name = tensor<string, []>("op_10050_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10050_cast = slice_by_index(begin = var_10050_begin_0, end = var_10050_end_0, end_mask = var_10050_end_mask_0, x = transpose_7);
            tensor<int32, [4]> var_10054_begin_0 = const()[name = tensor<string, []>("op_10054_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_10054_end_0 = const()[name = tensor<string, []>("op_10054_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_10054_end_mask_0 = const()[name = tensor<string, []>("op_10054_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10054_cast = slice_by_index(begin = var_10054_begin_0, end = var_10054_end_0, end_mask = var_10054_end_mask_0, x = transpose_7);
            tensor<int32, [4]> var_10058_begin_0 = const()[name = tensor<string, []>("op_10058_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_10058_end_0 = const()[name = tensor<string, []>("op_10058_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_10058_end_mask_0 = const()[name = tensor<string, []>("op_10058_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10058_cast = slice_by_index(begin = var_10058_begin_0, end = var_10058_end_0, end_mask = var_10058_end_mask_0, x = transpose_7);
            tensor<int32, [4]> var_10062_begin_0 = const()[name = tensor<string, []>("op_10062_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10062_end_0 = const()[name = tensor<string, []>("op_10062_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_10062_end_mask_0 = const()[name = tensor<string, []>("op_10062_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10062_cast = slice_by_index(begin = var_10062_begin_0, end = var_10062_end_0, end_mask = var_10062_end_mask_0, x = transpose_7);
            tensor<int32, [4]> var_10066_begin_0 = const()[name = tensor<string, []>("op_10066_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_10066_end_0 = const()[name = tensor<string, []>("op_10066_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_10066_end_mask_0 = const()[name = tensor<string, []>("op_10066_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10066_cast = slice_by_index(begin = var_10066_begin_0, end = var_10066_end_0, end_mask = var_10066_end_mask_0, x = transpose_7);
            tensor<int32, [4]> var_10068_begin_0 = const()[name = tensor<string, []>("op_10068_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10068_end_0 = const()[name = tensor<string, []>("op_10068_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10068_end_mask_0 = const()[name = tensor<string, []>("op_10068_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10068_cast = slice_by_index(begin = var_10068_begin_0, end = var_10068_end_0, end_mask = var_10068_end_mask_0, x = v_49_cast);
            tensor<int32, [4]> var_10072_begin_0 = const()[name = tensor<string, []>("op_10072_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_10072_end_0 = const()[name = tensor<string, []>("op_10072_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_10072_end_mask_0 = const()[name = tensor<string, []>("op_10072_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10072_cast = slice_by_index(begin = var_10072_begin_0, end = var_10072_end_0, end_mask = var_10072_end_mask_0, x = v_49_cast);
            tensor<int32, [4]> var_10076_begin_0 = const()[name = tensor<string, []>("op_10076_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_10076_end_0 = const()[name = tensor<string, []>("op_10076_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_10076_end_mask_0 = const()[name = tensor<string, []>("op_10076_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10076_cast = slice_by_index(begin = var_10076_begin_0, end = var_10076_end_0, end_mask = var_10076_end_mask_0, x = v_49_cast);
            tensor<int32, [4]> var_10080_begin_0 = const()[name = tensor<string, []>("op_10080_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_10080_end_0 = const()[name = tensor<string, []>("op_10080_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_10080_end_mask_0 = const()[name = tensor<string, []>("op_10080_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10080_cast = slice_by_index(begin = var_10080_begin_0, end = var_10080_end_0, end_mask = var_10080_end_mask_0, x = v_49_cast);
            tensor<int32, [4]> var_10084_begin_0 = const()[name = tensor<string, []>("op_10084_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_10084_end_0 = const()[name = tensor<string, []>("op_10084_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_10084_end_mask_0 = const()[name = tensor<string, []>("op_10084_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10084_cast = slice_by_index(begin = var_10084_begin_0, end = var_10084_end_0, end_mask = var_10084_end_mask_0, x = v_49_cast);
            tensor<int32, [4]> var_10088_begin_0 = const()[name = tensor<string, []>("op_10088_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_10088_end_0 = const()[name = tensor<string, []>("op_10088_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_10088_end_mask_0 = const()[name = tensor<string, []>("op_10088_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10088_cast = slice_by_index(begin = var_10088_begin_0, end = var_10088_end_0, end_mask = var_10088_end_mask_0, x = v_49_cast);
            tensor<int32, [4]> var_10092_begin_0 = const()[name = tensor<string, []>("op_10092_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_10092_end_0 = const()[name = tensor<string, []>("op_10092_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_10092_end_mask_0 = const()[name = tensor<string, []>("op_10092_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10092_cast = slice_by_index(begin = var_10092_begin_0, end = var_10092_end_0, end_mask = var_10092_end_mask_0, x = v_49_cast);
            tensor<int32, [4]> var_10096_begin_0 = const()[name = tensor<string, []>("op_10096_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_10096_end_0 = const()[name = tensor<string, []>("op_10096_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_10096_end_mask_0 = const()[name = tensor<string, []>("op_10096_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10096_cast = slice_by_index(begin = var_10096_begin_0, end = var_10096_end_0, end_mask = var_10096_end_mask_0, x = v_49_cast);
            tensor<int32, [4]> var_10100_begin_0 = const()[name = tensor<string, []>("op_10100_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_10100_end_0 = const()[name = tensor<string, []>("op_10100_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_10100_end_mask_0 = const()[name = tensor<string, []>("op_10100_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10100_cast = slice_by_index(begin = var_10100_begin_0, end = var_10100_end_0, end_mask = var_10100_end_mask_0, x = v_49_cast);
            tensor<int32, [4]> var_10104_begin_0 = const()[name = tensor<string, []>("op_10104_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_10104_end_0 = const()[name = tensor<string, []>("op_10104_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_10104_end_mask_0 = const()[name = tensor<string, []>("op_10104_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10104_cast = slice_by_index(begin = var_10104_begin_0, end = var_10104_end_0, end_mask = var_10104_end_mask_0, x = v_49_cast);
            tensor<string, []> var_10108_equation_0 = const()[name = tensor<string, []>("op_10108_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10108_cast = einsum(equation = var_10108_equation_0, values = (var_10030_cast, var_9987_cast));
            tensor<fp16, []> var_10109_to_fp16 = const()[name = tensor<string, []>("op_10109_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_681_cast = mul(x = var_10108_cast, y = var_10109_to_fp16);
            tensor<string, []> var_10112_equation_0 = const()[name = tensor<string, []>("op_10112_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10112_cast = einsum(equation = var_10112_equation_0, values = (var_10034_cast, var_9991_cast));
            tensor<fp16, []> var_10113_to_fp16 = const()[name = tensor<string, []>("op_10113_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_683_cast = mul(x = var_10112_cast, y = var_10113_to_fp16);
            tensor<string, []> var_10116_equation_0 = const()[name = tensor<string, []>("op_10116_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10116_cast = einsum(equation = var_10116_equation_0, values = (var_10038_cast, var_9995_cast));
            tensor<fp16, []> var_10117_to_fp16 = const()[name = tensor<string, []>("op_10117_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_685_cast = mul(x = var_10116_cast, y = var_10117_to_fp16);
            tensor<string, []> var_10120_equation_0 = const()[name = tensor<string, []>("op_10120_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10120_cast = einsum(equation = var_10120_equation_0, values = (var_10042_cast, var_9999_cast));
            tensor<fp16, []> var_10121_to_fp16 = const()[name = tensor<string, []>("op_10121_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_687_cast = mul(x = var_10120_cast, y = var_10121_to_fp16);
            tensor<string, []> var_10124_equation_0 = const()[name = tensor<string, []>("op_10124_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10124_cast = einsum(equation = var_10124_equation_0, values = (var_10046_cast, var_10003_cast));
            tensor<fp16, []> var_10125_to_fp16 = const()[name = tensor<string, []>("op_10125_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_689_cast = mul(x = var_10124_cast, y = var_10125_to_fp16);
            tensor<string, []> var_10128_equation_0 = const()[name = tensor<string, []>("op_10128_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10128_cast = einsum(equation = var_10128_equation_0, values = (var_10050_cast, var_10007_cast));
            tensor<fp16, []> var_10129_to_fp16 = const()[name = tensor<string, []>("op_10129_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_691_cast = mul(x = var_10128_cast, y = var_10129_to_fp16);
            tensor<string, []> var_10132_equation_0 = const()[name = tensor<string, []>("op_10132_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10132_cast = einsum(equation = var_10132_equation_0, values = (var_10054_cast, var_10011_cast));
            tensor<fp16, []> var_10133_to_fp16 = const()[name = tensor<string, []>("op_10133_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_693_cast = mul(x = var_10132_cast, y = var_10133_to_fp16);
            tensor<string, []> var_10136_equation_0 = const()[name = tensor<string, []>("op_10136_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10136_cast = einsum(equation = var_10136_equation_0, values = (var_10058_cast, var_10015_cast));
            tensor<fp16, []> var_10137_to_fp16 = const()[name = tensor<string, []>("op_10137_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_695_cast = mul(x = var_10136_cast, y = var_10137_to_fp16);
            tensor<string, []> var_10140_equation_0 = const()[name = tensor<string, []>("op_10140_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10140_cast = einsum(equation = var_10140_equation_0, values = (var_10062_cast, var_10019_cast));
            tensor<fp16, []> var_10141_to_fp16 = const()[name = tensor<string, []>("op_10141_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_697_cast = mul(x = var_10140_cast, y = var_10141_to_fp16);
            tensor<string, []> var_10144_equation_0 = const()[name = tensor<string, []>("op_10144_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 1024]> var_10144_cast = einsum(equation = var_10144_equation_0, values = (var_10066_cast, var_10023_cast));
            tensor<fp16, []> var_10145_to_fp16 = const()[name = tensor<string, []>("op_10145_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 1024]> aw_699_cast = mul(x = var_10144_cast, y = var_10145_to_fp16);
            tensor<fp16, [2, 1024, 1, 1024]> var_10147_cast = softmax(axis = var_8627, x = aw_681_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_10148_cast = softmax(axis = var_8627, x = aw_683_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_10149_cast = softmax(axis = var_8627, x = aw_685_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_10150_cast = softmax(axis = var_8627, x = aw_687_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_10151_cast = softmax(axis = var_8627, x = aw_689_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_10152_cast = softmax(axis = var_8627, x = aw_691_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_10153_cast = softmax(axis = var_8627, x = aw_693_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_10154_cast = softmax(axis = var_8627, x = aw_695_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_10155_cast = softmax(axis = var_8627, x = aw_697_cast);
            tensor<fp16, [2, 1024, 1, 1024]> var_10156_cast = softmax(axis = var_8627, x = aw_699_cast);
            tensor<string, []> var_10158_equation_0 = const()[name = tensor<string, []>("op_10158_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10158_cast = einsum(equation = var_10158_equation_0, values = (var_10068_cast, var_10147_cast));
            tensor<string, []> var_10160_equation_0 = const()[name = tensor<string, []>("op_10160_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10160_cast = einsum(equation = var_10160_equation_0, values = (var_10072_cast, var_10148_cast));
            tensor<string, []> var_10162_equation_0 = const()[name = tensor<string, []>("op_10162_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10162_cast = einsum(equation = var_10162_equation_0, values = (var_10076_cast, var_10149_cast));
            tensor<string, []> var_10164_equation_0 = const()[name = tensor<string, []>("op_10164_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10164_cast = einsum(equation = var_10164_equation_0, values = (var_10080_cast, var_10150_cast));
            tensor<string, []> var_10166_equation_0 = const()[name = tensor<string, []>("op_10166_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10166_cast = einsum(equation = var_10166_equation_0, values = (var_10084_cast, var_10151_cast));
            tensor<string, []> var_10168_equation_0 = const()[name = tensor<string, []>("op_10168_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10168_cast = einsum(equation = var_10168_equation_0, values = (var_10088_cast, var_10152_cast));
            tensor<string, []> var_10170_equation_0 = const()[name = tensor<string, []>("op_10170_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10170_cast = einsum(equation = var_10170_equation_0, values = (var_10092_cast, var_10153_cast));
            tensor<string, []> var_10172_equation_0 = const()[name = tensor<string, []>("op_10172_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10172_cast = einsum(equation = var_10172_equation_0, values = (var_10096_cast, var_10154_cast));
            tensor<string, []> var_10174_equation_0 = const()[name = tensor<string, []>("op_10174_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10174_cast = einsum(equation = var_10174_equation_0, values = (var_10100_cast, var_10155_cast));
            tensor<string, []> var_10176_equation_0 = const()[name = tensor<string, []>("op_10176_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10176_cast = einsum(equation = var_10176_equation_0, values = (var_10104_cast, var_10156_cast));
            tensor<bool, []> input_435_interleave_0 = const()[name = tensor<string, []>("input_435_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 1024]> input_435_cast = concat(axis = var_8627, interleave = input_435_interleave_0, values = (var_10158_cast, var_10160_cast, var_10162_cast, var_10164_cast, var_10166_cast, var_10168_cast, var_10170_cast, var_10172_cast, var_10174_cast, var_10176_cast));
            tensor<int32, [2]> var_10182 = const()[name = tensor<string, []>("op_10182"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10184 = const()[name = tensor<string, []>("op_10184"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_10186_pad_type_0 = const()[name = tensor<string, []>("op_10186_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_10186_pad_0 = const()[name = tensor<string, []>("op_10186_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1670344832)))];
            tensor<fp16, [640]> up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1671164096)))];
            tensor<fp16, [2, 640, 1, 1024]> var_10186_cast = conv(bias = up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_10184, groups = var_8627, pad = var_10186_pad_0, pad_type = var_10186_pad_type_0, strides = var_10182, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_435_cast);
            tensor<fp16, [2, 640, 1, 1024]> inputs_75_cast = add(x = var_10186_cast, y = inputs_73_cast);
            tensor<int32, [1]> var_10190 = const()[name = tensor<string, []>("op_10190"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_75_cast = reduce_mean(axes = var_10190, keep_dims = var_8622, x = inputs_75_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_75_cast = sub(x = inputs_75_cast, y = channels_mean_75_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_sq_75_cast = mul(x = zero_mean_75_cast, y = zero_mean_75_cast);
            tensor<int32, [1]> var_10194 = const()[name = tensor<string, []>("op_10194"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_10195_cast = reduce_mean(axes = var_10194, keep_dims = var_8622, x = zero_mean_sq_75_cast);
            tensor<fp16, []> var_10196_to_fp16 = const()[name = tensor<string, []>("op_10196_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_10197_cast = add(x = var_10195_cast, y = var_10196_to_fp16);
            tensor<fp16, []> denom_75_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_75_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_75_cast = rsqrt(epsilon = denom_75_epsilon_0_to_fp16, x = var_10197_cast);
            tensor<fp16, [2, 640, 1, 1024]> out_75_cast = mul(x = zero_mean_75_cast, y = denom_75_cast);
            tensor<fp16, [1, 640, 1, 1]> var_10201_to_fp16 = const()[name = tensor<string, []>("op_10201_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1671165440)))];
            tensor<fp16, [2, 640, 1, 1024]> var_10202_cast = add(x = out_75_cast, y = var_10201_to_fp16);
            tensor<fp16, [1, 640, 1, 1]> var_10204_to_fp16 = const()[name = tensor<string, []>("op_10204_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1671166784)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_269_cast = mul(x = var_10202_cast, y = var_10204_to_fp16);
            tensor<int32, [2]> var_10211 = const()[name = tensor<string, []>("op_10211"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10213 = const()[name = tensor<string, []>("op_10213"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_51_pad_type_0 = const()[name = tensor<string, []>("q_51_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_51_pad_0 = const()[name = tensor<string, []>("q_51_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1671168128)))];
            tensor<fp16, [2, 640, 1, 1024]> q_51_cast = conv(dilations = var_10213, groups = var_8627, pad = q_51_pad_0, pad_type = q_51_pad_type_0, strides = var_10211, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_269_cast);
            tensor<int32, [2]> var_10217 = const()[name = tensor<string, []>("op_10217"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10219 = const()[name = tensor<string, []>("op_10219"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_101_pad_type_0 = const()[name = tensor<string, []>("k_101_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_101_pad_0 = const()[name = tensor<string, []>("k_101_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1024, 1, 1]> up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [640, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1671987392)))];
            tensor<fp16, [2, 640, 1, 77]> k_101_cast = conv(dilations = var_10219, groups = var_8627, pad = k_101_pad_0, pad_type = k_101_pad_type_0, strides = var_10217, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [2]> var_10223 = const()[name = tensor<string, []>("op_10223"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10225 = const()[name = tensor<string, []>("op_10225"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_51_pad_type_0 = const()[name = tensor<string, []>("v_51_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_51_pad_0 = const()[name = tensor<string, []>("v_51_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1024, 1, 1]> up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [640, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1673298176)))];
            tensor<fp16, [2, 640, 1, 77]> v_51_cast = conv(dilations = var_10225, groups = var_8627, pad = v_51_pad_0, pad_type = v_51_pad_type_0, strides = var_10223, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [4]> var_10229_begin_0 = const()[name = tensor<string, []>("op_10229_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10229_end_0 = const()[name = tensor<string, []>("op_10229_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10229_end_mask_0 = const()[name = tensor<string, []>("op_10229_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10229_cast = slice_by_index(begin = var_10229_begin_0, end = var_10229_end_0, end_mask = var_10229_end_mask_0, x = q_51_cast);
            tensor<int32, [4]> var_10233_begin_0 = const()[name = tensor<string, []>("op_10233_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_10233_end_0 = const()[name = tensor<string, []>("op_10233_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_10233_end_mask_0 = const()[name = tensor<string, []>("op_10233_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10233_cast = slice_by_index(begin = var_10233_begin_0, end = var_10233_end_0, end_mask = var_10233_end_mask_0, x = q_51_cast);
            tensor<int32, [4]> var_10237_begin_0 = const()[name = tensor<string, []>("op_10237_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_10237_end_0 = const()[name = tensor<string, []>("op_10237_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_10237_end_mask_0 = const()[name = tensor<string, []>("op_10237_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10237_cast = slice_by_index(begin = var_10237_begin_0, end = var_10237_end_0, end_mask = var_10237_end_mask_0, x = q_51_cast);
            tensor<int32, [4]> var_10241_begin_0 = const()[name = tensor<string, []>("op_10241_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_10241_end_0 = const()[name = tensor<string, []>("op_10241_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_10241_end_mask_0 = const()[name = tensor<string, []>("op_10241_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10241_cast = slice_by_index(begin = var_10241_begin_0, end = var_10241_end_0, end_mask = var_10241_end_mask_0, x = q_51_cast);
            tensor<int32, [4]> var_10245_begin_0 = const()[name = tensor<string, []>("op_10245_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_10245_end_0 = const()[name = tensor<string, []>("op_10245_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_10245_end_mask_0 = const()[name = tensor<string, []>("op_10245_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10245_cast = slice_by_index(begin = var_10245_begin_0, end = var_10245_end_0, end_mask = var_10245_end_mask_0, x = q_51_cast);
            tensor<int32, [4]> var_10249_begin_0 = const()[name = tensor<string, []>("op_10249_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_10249_end_0 = const()[name = tensor<string, []>("op_10249_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_10249_end_mask_0 = const()[name = tensor<string, []>("op_10249_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10249_cast = slice_by_index(begin = var_10249_begin_0, end = var_10249_end_0, end_mask = var_10249_end_mask_0, x = q_51_cast);
            tensor<int32, [4]> var_10253_begin_0 = const()[name = tensor<string, []>("op_10253_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_10253_end_0 = const()[name = tensor<string, []>("op_10253_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_10253_end_mask_0 = const()[name = tensor<string, []>("op_10253_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10253_cast = slice_by_index(begin = var_10253_begin_0, end = var_10253_end_0, end_mask = var_10253_end_mask_0, x = q_51_cast);
            tensor<int32, [4]> var_10257_begin_0 = const()[name = tensor<string, []>("op_10257_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_10257_end_0 = const()[name = tensor<string, []>("op_10257_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_10257_end_mask_0 = const()[name = tensor<string, []>("op_10257_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10257_cast = slice_by_index(begin = var_10257_begin_0, end = var_10257_end_0, end_mask = var_10257_end_mask_0, x = q_51_cast);
            tensor<int32, [4]> var_10261_begin_0 = const()[name = tensor<string, []>("op_10261_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_10261_end_0 = const()[name = tensor<string, []>("op_10261_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_10261_end_mask_0 = const()[name = tensor<string, []>("op_10261_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10261_cast = slice_by_index(begin = var_10261_begin_0, end = var_10261_end_0, end_mask = var_10261_end_mask_0, x = q_51_cast);
            tensor<int32, [4]> var_10265_begin_0 = const()[name = tensor<string, []>("op_10265_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_10265_end_0 = const()[name = tensor<string, []>("op_10265_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_10265_end_mask_0 = const()[name = tensor<string, []>("op_10265_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10265_cast = slice_by_index(begin = var_10265_begin_0, end = var_10265_end_0, end_mask = var_10265_end_mask_0, x = q_51_cast);
            tensor<int32, [4]> k_103_perm_0 = const()[name = tensor<string, []>("k_103_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_10272_begin_0 = const()[name = tensor<string, []>("op_10272_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10272_end_0 = const()[name = tensor<string, []>("op_10272_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_10272_end_mask_0 = const()[name = tensor<string, []>("op_10272_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 640]> transpose_6 = transpose(perm = k_103_perm_0, x = k_101_cast);
            tensor<fp16, [2, 77, 1, 64]> var_10272_cast = slice_by_index(begin = var_10272_begin_0, end = var_10272_end_0, end_mask = var_10272_end_mask_0, x = transpose_6);
            tensor<int32, [4]> var_10276_begin_0 = const()[name = tensor<string, []>("op_10276_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_10276_end_0 = const()[name = tensor<string, []>("op_10276_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_10276_end_mask_0 = const()[name = tensor<string, []>("op_10276_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10276_cast = slice_by_index(begin = var_10276_begin_0, end = var_10276_end_0, end_mask = var_10276_end_mask_0, x = transpose_6);
            tensor<int32, [4]> var_10280_begin_0 = const()[name = tensor<string, []>("op_10280_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_10280_end_0 = const()[name = tensor<string, []>("op_10280_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_10280_end_mask_0 = const()[name = tensor<string, []>("op_10280_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10280_cast = slice_by_index(begin = var_10280_begin_0, end = var_10280_end_0, end_mask = var_10280_end_mask_0, x = transpose_6);
            tensor<int32, [4]> var_10284_begin_0 = const()[name = tensor<string, []>("op_10284_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_10284_end_0 = const()[name = tensor<string, []>("op_10284_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_10284_end_mask_0 = const()[name = tensor<string, []>("op_10284_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10284_cast = slice_by_index(begin = var_10284_begin_0, end = var_10284_end_0, end_mask = var_10284_end_mask_0, x = transpose_6);
            tensor<int32, [4]> var_10288_begin_0 = const()[name = tensor<string, []>("op_10288_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_10288_end_0 = const()[name = tensor<string, []>("op_10288_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_10288_end_mask_0 = const()[name = tensor<string, []>("op_10288_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10288_cast = slice_by_index(begin = var_10288_begin_0, end = var_10288_end_0, end_mask = var_10288_end_mask_0, x = transpose_6);
            tensor<int32, [4]> var_10292_begin_0 = const()[name = tensor<string, []>("op_10292_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_10292_end_0 = const()[name = tensor<string, []>("op_10292_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_10292_end_mask_0 = const()[name = tensor<string, []>("op_10292_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10292_cast = slice_by_index(begin = var_10292_begin_0, end = var_10292_end_0, end_mask = var_10292_end_mask_0, x = transpose_6);
            tensor<int32, [4]> var_10296_begin_0 = const()[name = tensor<string, []>("op_10296_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_10296_end_0 = const()[name = tensor<string, []>("op_10296_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_10296_end_mask_0 = const()[name = tensor<string, []>("op_10296_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10296_cast = slice_by_index(begin = var_10296_begin_0, end = var_10296_end_0, end_mask = var_10296_end_mask_0, x = transpose_6);
            tensor<int32, [4]> var_10300_begin_0 = const()[name = tensor<string, []>("op_10300_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_10300_end_0 = const()[name = tensor<string, []>("op_10300_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_10300_end_mask_0 = const()[name = tensor<string, []>("op_10300_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10300_cast = slice_by_index(begin = var_10300_begin_0, end = var_10300_end_0, end_mask = var_10300_end_mask_0, x = transpose_6);
            tensor<int32, [4]> var_10304_begin_0 = const()[name = tensor<string, []>("op_10304_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10304_end_0 = const()[name = tensor<string, []>("op_10304_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_10304_end_mask_0 = const()[name = tensor<string, []>("op_10304_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10304_cast = slice_by_index(begin = var_10304_begin_0, end = var_10304_end_0, end_mask = var_10304_end_mask_0, x = transpose_6);
            tensor<int32, [4]> var_10308_begin_0 = const()[name = tensor<string, []>("op_10308_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_10308_end_0 = const()[name = tensor<string, []>("op_10308_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_10308_end_mask_0 = const()[name = tensor<string, []>("op_10308_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10308_cast = slice_by_index(begin = var_10308_begin_0, end = var_10308_end_0, end_mask = var_10308_end_mask_0, x = transpose_6);
            tensor<int32, [4]> var_10310_begin_0 = const()[name = tensor<string, []>("op_10310_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10310_end_0 = const()[name = tensor<string, []>("op_10310_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_10310_end_mask_0 = const()[name = tensor<string, []>("op_10310_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10310_cast = slice_by_index(begin = var_10310_begin_0, end = var_10310_end_0, end_mask = var_10310_end_mask_0, x = v_51_cast);
            tensor<int32, [4]> var_10314_begin_0 = const()[name = tensor<string, []>("op_10314_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_10314_end_0 = const()[name = tensor<string, []>("op_10314_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_10314_end_mask_0 = const()[name = tensor<string, []>("op_10314_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10314_cast = slice_by_index(begin = var_10314_begin_0, end = var_10314_end_0, end_mask = var_10314_end_mask_0, x = v_51_cast);
            tensor<int32, [4]> var_10318_begin_0 = const()[name = tensor<string, []>("op_10318_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_10318_end_0 = const()[name = tensor<string, []>("op_10318_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_10318_end_mask_0 = const()[name = tensor<string, []>("op_10318_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10318_cast = slice_by_index(begin = var_10318_begin_0, end = var_10318_end_0, end_mask = var_10318_end_mask_0, x = v_51_cast);
            tensor<int32, [4]> var_10322_begin_0 = const()[name = tensor<string, []>("op_10322_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_10322_end_0 = const()[name = tensor<string, []>("op_10322_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_10322_end_mask_0 = const()[name = tensor<string, []>("op_10322_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10322_cast = slice_by_index(begin = var_10322_begin_0, end = var_10322_end_0, end_mask = var_10322_end_mask_0, x = v_51_cast);
            tensor<int32, [4]> var_10326_begin_0 = const()[name = tensor<string, []>("op_10326_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_10326_end_0 = const()[name = tensor<string, []>("op_10326_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_10326_end_mask_0 = const()[name = tensor<string, []>("op_10326_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10326_cast = slice_by_index(begin = var_10326_begin_0, end = var_10326_end_0, end_mask = var_10326_end_mask_0, x = v_51_cast);
            tensor<int32, [4]> var_10330_begin_0 = const()[name = tensor<string, []>("op_10330_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_10330_end_0 = const()[name = tensor<string, []>("op_10330_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_10330_end_mask_0 = const()[name = tensor<string, []>("op_10330_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10330_cast = slice_by_index(begin = var_10330_begin_0, end = var_10330_end_0, end_mask = var_10330_end_mask_0, x = v_51_cast);
            tensor<int32, [4]> var_10334_begin_0 = const()[name = tensor<string, []>("op_10334_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_10334_end_0 = const()[name = tensor<string, []>("op_10334_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_10334_end_mask_0 = const()[name = tensor<string, []>("op_10334_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10334_cast = slice_by_index(begin = var_10334_begin_0, end = var_10334_end_0, end_mask = var_10334_end_mask_0, x = v_51_cast);
            tensor<int32, [4]> var_10338_begin_0 = const()[name = tensor<string, []>("op_10338_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_10338_end_0 = const()[name = tensor<string, []>("op_10338_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_10338_end_mask_0 = const()[name = tensor<string, []>("op_10338_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10338_cast = slice_by_index(begin = var_10338_begin_0, end = var_10338_end_0, end_mask = var_10338_end_mask_0, x = v_51_cast);
            tensor<int32, [4]> var_10342_begin_0 = const()[name = tensor<string, []>("op_10342_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_10342_end_0 = const()[name = tensor<string, []>("op_10342_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_10342_end_mask_0 = const()[name = tensor<string, []>("op_10342_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10342_cast = slice_by_index(begin = var_10342_begin_0, end = var_10342_end_0, end_mask = var_10342_end_mask_0, x = v_51_cast);
            tensor<int32, [4]> var_10346_begin_0 = const()[name = tensor<string, []>("op_10346_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_10346_end_0 = const()[name = tensor<string, []>("op_10346_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_10346_end_mask_0 = const()[name = tensor<string, []>("op_10346_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10346_cast = slice_by_index(begin = var_10346_begin_0, end = var_10346_end_0, end_mask = var_10346_end_mask_0, x = v_51_cast);
            tensor<string, []> var_10350_equation_0 = const()[name = tensor<string, []>("op_10350_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10350_cast = einsum(equation = var_10350_equation_0, values = (var_10272_cast, var_10229_cast));
            tensor<fp16, []> var_10351_to_fp16 = const()[name = tensor<string, []>("op_10351_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_701_cast = mul(x = var_10350_cast, y = var_10351_to_fp16);
            tensor<string, []> var_10354_equation_0 = const()[name = tensor<string, []>("op_10354_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10354_cast = einsum(equation = var_10354_equation_0, values = (var_10276_cast, var_10233_cast));
            tensor<fp16, []> var_10355_to_fp16 = const()[name = tensor<string, []>("op_10355_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_703_cast = mul(x = var_10354_cast, y = var_10355_to_fp16);
            tensor<string, []> var_10358_equation_0 = const()[name = tensor<string, []>("op_10358_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10358_cast = einsum(equation = var_10358_equation_0, values = (var_10280_cast, var_10237_cast));
            tensor<fp16, []> var_10359_to_fp16 = const()[name = tensor<string, []>("op_10359_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_705_cast = mul(x = var_10358_cast, y = var_10359_to_fp16);
            tensor<string, []> var_10362_equation_0 = const()[name = tensor<string, []>("op_10362_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10362_cast = einsum(equation = var_10362_equation_0, values = (var_10284_cast, var_10241_cast));
            tensor<fp16, []> var_10363_to_fp16 = const()[name = tensor<string, []>("op_10363_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_707_cast = mul(x = var_10362_cast, y = var_10363_to_fp16);
            tensor<string, []> var_10366_equation_0 = const()[name = tensor<string, []>("op_10366_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10366_cast = einsum(equation = var_10366_equation_0, values = (var_10288_cast, var_10245_cast));
            tensor<fp16, []> var_10367_to_fp16 = const()[name = tensor<string, []>("op_10367_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_709_cast = mul(x = var_10366_cast, y = var_10367_to_fp16);
            tensor<string, []> var_10370_equation_0 = const()[name = tensor<string, []>("op_10370_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10370_cast = einsum(equation = var_10370_equation_0, values = (var_10292_cast, var_10249_cast));
            tensor<fp16, []> var_10371_to_fp16 = const()[name = tensor<string, []>("op_10371_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_711_cast = mul(x = var_10370_cast, y = var_10371_to_fp16);
            tensor<string, []> var_10374_equation_0 = const()[name = tensor<string, []>("op_10374_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10374_cast = einsum(equation = var_10374_equation_0, values = (var_10296_cast, var_10253_cast));
            tensor<fp16, []> var_10375_to_fp16 = const()[name = tensor<string, []>("op_10375_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_713_cast = mul(x = var_10374_cast, y = var_10375_to_fp16);
            tensor<string, []> var_10378_equation_0 = const()[name = tensor<string, []>("op_10378_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10378_cast = einsum(equation = var_10378_equation_0, values = (var_10300_cast, var_10257_cast));
            tensor<fp16, []> var_10379_to_fp16 = const()[name = tensor<string, []>("op_10379_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_715_cast = mul(x = var_10378_cast, y = var_10379_to_fp16);
            tensor<string, []> var_10382_equation_0 = const()[name = tensor<string, []>("op_10382_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10382_cast = einsum(equation = var_10382_equation_0, values = (var_10304_cast, var_10261_cast));
            tensor<fp16, []> var_10383_to_fp16 = const()[name = tensor<string, []>("op_10383_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_717_cast = mul(x = var_10382_cast, y = var_10383_to_fp16);
            tensor<string, []> var_10386_equation_0 = const()[name = tensor<string, []>("op_10386_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 1024]> var_10386_cast = einsum(equation = var_10386_equation_0, values = (var_10308_cast, var_10265_cast));
            tensor<fp16, []> var_10387_to_fp16 = const()[name = tensor<string, []>("op_10387_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 1024]> aw_719_cast = mul(x = var_10386_cast, y = var_10387_to_fp16);
            tensor<fp16, [2, 77, 1, 1024]> var_10389_cast = softmax(axis = var_8627, x = aw_701_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_10390_cast = softmax(axis = var_8627, x = aw_703_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_10391_cast = softmax(axis = var_8627, x = aw_705_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_10392_cast = softmax(axis = var_8627, x = aw_707_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_10393_cast = softmax(axis = var_8627, x = aw_709_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_10394_cast = softmax(axis = var_8627, x = aw_711_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_10395_cast = softmax(axis = var_8627, x = aw_713_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_10396_cast = softmax(axis = var_8627, x = aw_715_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_10397_cast = softmax(axis = var_8627, x = aw_717_cast);
            tensor<fp16, [2, 77, 1, 1024]> var_10398_cast = softmax(axis = var_8627, x = aw_719_cast);
            tensor<string, []> var_10400_equation_0 = const()[name = tensor<string, []>("op_10400_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10400_cast = einsum(equation = var_10400_equation_0, values = (var_10310_cast, var_10389_cast));
            tensor<string, []> var_10402_equation_0 = const()[name = tensor<string, []>("op_10402_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10402_cast = einsum(equation = var_10402_equation_0, values = (var_10314_cast, var_10390_cast));
            tensor<string, []> var_10404_equation_0 = const()[name = tensor<string, []>("op_10404_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10404_cast = einsum(equation = var_10404_equation_0, values = (var_10318_cast, var_10391_cast));
            tensor<string, []> var_10406_equation_0 = const()[name = tensor<string, []>("op_10406_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10406_cast = einsum(equation = var_10406_equation_0, values = (var_10322_cast, var_10392_cast));
            tensor<string, []> var_10408_equation_0 = const()[name = tensor<string, []>("op_10408_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10408_cast = einsum(equation = var_10408_equation_0, values = (var_10326_cast, var_10393_cast));
            tensor<string, []> var_10410_equation_0 = const()[name = tensor<string, []>("op_10410_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10410_cast = einsum(equation = var_10410_equation_0, values = (var_10330_cast, var_10394_cast));
            tensor<string, []> var_10412_equation_0 = const()[name = tensor<string, []>("op_10412_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10412_cast = einsum(equation = var_10412_equation_0, values = (var_10334_cast, var_10395_cast));
            tensor<string, []> var_10414_equation_0 = const()[name = tensor<string, []>("op_10414_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10414_cast = einsum(equation = var_10414_equation_0, values = (var_10338_cast, var_10396_cast));
            tensor<string, []> var_10416_equation_0 = const()[name = tensor<string, []>("op_10416_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10416_cast = einsum(equation = var_10416_equation_0, values = (var_10342_cast, var_10397_cast));
            tensor<string, []> var_10418_equation_0 = const()[name = tensor<string, []>("op_10418_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 1024]> var_10418_cast = einsum(equation = var_10418_equation_0, values = (var_10346_cast, var_10398_cast));
            tensor<bool, []> input_437_interleave_0 = const()[name = tensor<string, []>("input_437_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 1, 1024]> input_437_cast = concat(axis = var_8627, interleave = input_437_interleave_0, values = (var_10400_cast, var_10402_cast, var_10404_cast, var_10406_cast, var_10408_cast, var_10410_cast, var_10412_cast, var_10414_cast, var_10416_cast, var_10418_cast));
            tensor<int32, [2]> var_10424 = const()[name = tensor<string, []>("op_10424"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10426 = const()[name = tensor<string, []>("op_10426"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_10428_pad_type_0 = const()[name = tensor<string, []>("op_10428_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_10428_pad_0 = const()[name = tensor<string, []>("op_10428_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1674608960)))];
            tensor<fp16, [640]> up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1675428224)))];
            tensor<fp16, [2, 640, 1, 1024]> var_10428_cast = conv(bias = up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_10426, groups = var_8627, pad = var_10428_pad_0, pad_type = var_10428_pad_type_0, strides = var_10424, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_437_cast);
            tensor<fp16, [2, 640, 1, 1024]> inputs_77_cast = add(x = var_10428_cast, y = inputs_75_cast);
            tensor<int32, [1]> var_10432 = const()[name = tensor<string, []>("op_10432"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_77_cast = reduce_mean(axes = var_10432, keep_dims = var_8622, x = inputs_77_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_77_cast = sub(x = inputs_77_cast, y = channels_mean_77_cast);
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_sq_77_cast = mul(x = zero_mean_77_cast, y = zero_mean_77_cast);
            tensor<int32, [1]> var_10436 = const()[name = tensor<string, []>("op_10436"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_10437_cast = reduce_mean(axes = var_10436, keep_dims = var_8622, x = zero_mean_sq_77_cast);
            tensor<fp16, []> var_10438_to_fp16 = const()[name = tensor<string, []>("op_10438_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_10439_cast = add(x = var_10437_cast, y = var_10438_to_fp16);
            tensor<fp16, []> denom_77_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_77_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 1024]> denom_77_cast = rsqrt(epsilon = denom_77_epsilon_0_to_fp16, x = var_10439_cast);
            tensor<fp16, [2, 640, 1, 1024]> out_77_cast = mul(x = zero_mean_77_cast, y = denom_77_cast);
            tensor<fp16, [1, 640, 1, 1]> var_10443_to_fp16 = const()[name = tensor<string, []>("op_10443_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1675429568)))];
            tensor<fp16, [2, 640, 1, 1024]> var_10444_cast = add(x = out_77_cast, y = var_10443_to_fp16);
            tensor<fp16, [1, 640, 1, 1]> var_10446_to_fp16 = const()[name = tensor<string, []>("op_10446_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1675430912)))];
            tensor<fp16, [2, 640, 1, 1024]> input_439_cast = mul(x = var_10444_cast, y = var_10446_to_fp16);
            tensor<int32, [2]> var_10454 = const()[name = tensor<string, []>("op_10454"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10456 = const()[name = tensor<string, []>("op_10456"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_10458_pad_type_0 = const()[name = tensor<string, []>("op_10458_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_10458_pad_0 = const()[name = tensor<string, []>("op_10458_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [5120, 640, 1, 1]> up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [5120, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1675432256)))];
            tensor<fp16, [5120]> up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1681985920)))];
            tensor<fp16, [2, 5120, 1, 1024]> var_10458_cast = conv(bias = up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_10456, groups = var_8627, pad = var_10458_pad_0, pad_type = var_10458_pad_type_0, strides = var_10454, weight = up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_439_cast);
            tensor<int32, [2]> var_10459_split_sizes_0 = const()[name = tensor<string, []>("op_10459_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_10459_axis_0 = const()[name = tensor<string, []>("op_10459_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 1024]> var_10459_cast_0, tensor<fp16, [2, 2560, 1, 1024]> var_10459_cast_1 = split(axis = var_10459_axis_0, split_sizes = var_10459_split_sizes_0, x = var_10458_cast);
            tensor<string, []> var_10461_mode_0 = const()[name = tensor<string, []>("op_10461_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 1024]> var_10461_cast = gelu(mode = var_10461_mode_0, x = var_10459_cast_1);
            tensor<fp16, [2, 2560, 1, 1024]> input_441_cast = mul(x = var_10459_cast_0, y = var_10461_cast);
            tensor<int32, [2]> var_10465 = const()[name = tensor<string, []>("op_10465"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10467 = const()[name = tensor<string, []>("op_10467"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_10469_pad_type_0 = const()[name = tensor<string, []>("op_10469_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_10469_pad_0 = const()[name = tensor<string, []>("op_10469_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 2560, 1, 1]> up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [640, 2560, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1681996224)))];
            tensor<fp16, [640]> up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1685273088)))];
            tensor<fp16, [2, 640, 1, 1024]> var_10469_cast = conv(bias = up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_10467, groups = var_8627, pad = var_10469_pad_0, pad_type = var_10469_pad_type_0, strides = var_10465, weight = up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_441_cast);
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_273_cast = add(x = var_10469_cast, y = inputs_77_cast);
            tensor<int32, [4]> var_10471 = const()[name = tensor<string, []>("op_10471"), val = tensor<int32, [4]>([2, 640, 32, 32])];
            tensor<fp16, [2, 640, 32, 32]> input_443_cast = reshape(shape = var_10471, x = hidden_states_273_cast);
            tensor<int32, [2]> var_10475 = const()[name = tensor<string, []>("op_10475"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10477 = const()[name = tensor<string, []>("op_10477"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_275_pad_type_0 = const()[name = tensor<string, []>("hidden_states_275_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_275_pad_0 = const()[name = tensor<string, []>("hidden_states_275_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 640, 1, 1]> up_blocks_2_attentions_2_proj_out_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_2_proj_out_weight_to_fp16"), val = tensor<fp16, [640, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1685274432)))];
            tensor<fp16, [640]> up_blocks_2_attentions_2_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_attentions_2_proj_out_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1686093696)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_275_cast = conv(bias = up_blocks_2_attentions_2_proj_out_bias_to_fp16, dilations = var_10477, groups = var_8627, pad = hidden_states_275_pad_0, pad_type = hidden_states_275_pad_type_0, strides = var_10475, weight = up_blocks_2_attentions_2_proj_out_weight_to_fp16, x = input_443_cast);
            tensor<fp16, [2, 640, 32, 32]> input_445_cast = add(x = hidden_states_275_cast, y = hidden_states_263_cast);
            tensor<fp32, []> input_447_scale_factor_height_0 = const()[name = tensor<string, []>("input_447_scale_factor_height_0"), val = tensor<fp32, []>(0x1p+1)];
            tensor<fp32, []> input_447_scale_factor_width_0 = const()[name = tensor<string, []>("input_447_scale_factor_width_0"), val = tensor<fp32, []>(0x1p+1)];
            tensor<fp16, [2, 640, 64, 64]> input_447_cast = upsample_nearest_neighbor(scale_factor_height = input_447_scale_factor_height_0, scale_factor_width = input_447_scale_factor_width_0, x = input_445_cast);
            tensor<int32, [2]> var_10486 = const()[name = tensor<string, []>("op_10486"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10488 = const()[name = tensor<string, []>("op_10488"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_277_pad_type_0 = const()[name = tensor<string, []>("hidden_states_277_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_277_pad_0 = const()[name = tensor<string, []>("hidden_states_277_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [640, 640, 3, 3]> up_blocks_2_upsamplers_0_conv_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_upsamplers_0_conv_weight_to_fp16"), val = tensor<fp16, [640, 640, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1686095040)))];
            tensor<fp16, [640]> up_blocks_2_upsamplers_0_conv_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_2_upsamplers_0_conv_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1693467904)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_277_cast = conv(bias = up_blocks_2_upsamplers_0_conv_bias_to_fp16, dilations = var_10488, groups = var_8627, pad = hidden_states_277_pad_0, pad_type = hidden_states_277_pad_type_0, strides = var_10486, weight = up_blocks_2_upsamplers_0_conv_weight_to_fp16, x = input_447_cast);
            tensor<bool, []> var_10504 = const()[name = tensor<string, []>("op_10504"), val = tensor<bool, []>(true)];
            tensor<int32, []> var_10509 = const()[name = tensor<string, []>("op_10509"), val = tensor<int32, []>(1)];
            tensor<bool, []> input_449_interleave_0 = const()[name = tensor<string, []>("input_449_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 960, 64, 64]> input_449_cast = concat(axis = var_10509, interleave = input_449_interleave_0, values = (hidden_states_277_cast, input_61_cast));
            tensor<int32, [5]> reshape_204_shape_0 = const()[name = tensor<string, []>("reshape_204_shape_0"), val = tensor<int32, [5]>([2, 32, 30, 64, 64])];
            tensor<fp16, [2, 32, 30, 64, 64]> reshape_204_cast = reshape(shape = reshape_204_shape_0, x = input_449_cast);
            tensor<int32, [3]> reduce_mean_153_axes_0 = const()[name = tensor<string, []>("reduce_mean_153_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_153_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_153_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_153_cast = reduce_mean(axes = reduce_mean_153_axes_0, keep_dims = reduce_mean_153_keep_dims_0, x = reshape_204_cast);
            tensor<fp16, [2, 32, 30, 64, 64]> sub_102_cast = sub(x = reshape_204_cast, y = reduce_mean_153_cast);
            tensor<fp16, [2, 32, 30, 64, 64]> square_51_cast = square(x = sub_102_cast);
            tensor<int32, [3]> reduce_mean_155_axes_0 = const()[name = tensor<string, []>("reduce_mean_155_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_155_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_155_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_155_cast = reduce_mean(axes = reduce_mean_155_axes_0, keep_dims = reduce_mean_155_keep_dims_0, x = square_51_cast);
            tensor<fp16, []> add_102_y_0_to_fp16 = const()[name = tensor<string, []>("add_102_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_102_cast = add(x = reduce_mean_155_cast, y = add_102_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_51_cast = sqrt(x = add_102_cast);
            tensor<fp16, [2, 32, 30, 64, 64]> real_div_51_cast = real_div(x = sub_102_cast, y = sqrt_51_cast);
            tensor<int32, [4]> reshape_205_shape_0 = const()[name = tensor<string, []>("reshape_205_shape_0"), val = tensor<int32, [4]>([2, 960, 64, 64])];
            tensor<fp16, [2, 960, 64, 64]> reshape_205_cast = reshape(shape = reshape_205_shape_0, x = real_div_51_cast);
            tensor<fp16, [960]> add_103_mean_0_to_fp16 = const()[name = tensor<string, []>("add_103_mean_0_to_fp16"), val = tensor<fp16, [960]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1693469248)))];
            tensor<fp16, [960]> add_103_variance_0_to_fp16 = const()[name = tensor<string, []>("add_103_variance_0_to_fp16"), val = tensor<fp16, [960]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1693471232)))];
            tensor<fp16, [960]> add_103_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_103_gamma_0_to_fp16"), val = tensor<fp16, [960]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1693473216)))];
            tensor<fp16, [960]> add_103_beta_0_to_fp16 = const()[name = tensor<string, []>("add_103_beta_0_to_fp16"), val = tensor<fp16, [960]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1693475200)))];
            tensor<fp16, []> add_103_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_103_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 960, 64, 64]> add_103_cast = batch_norm(beta = add_103_beta_0_to_fp16, epsilon = add_103_epsilon_0_to_fp16, gamma = add_103_gamma_0_to_fp16, mean = add_103_mean_0_to_fp16, variance = add_103_variance_0_to_fp16, x = reshape_205_cast);
            tensor<fp16, [2, 960, 64, 64]> input_453_cast = silu(x = add_103_cast);
            tensor<int32, [2]> var_10536 = const()[name = tensor<string, []>("op_10536"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10538 = const()[name = tensor<string, []>("op_10538"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_279_pad_type_0 = const()[name = tensor<string, []>("hidden_states_279_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_279_pad_0 = const()[name = tensor<string, []>("hidden_states_279_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [320, 960, 3, 3]> up_blocks_3_resnets_0_conv1_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_resnets_0_conv1_weight_to_fp16"), val = tensor<fp16, [320, 960, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1693477184)))];
            tensor<fp16, [320]> up_blocks_3_resnets_0_conv1_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_resnets_0_conv1_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1699006848)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_279_cast = conv(bias = up_blocks_3_resnets_0_conv1_bias_to_fp16, dilations = var_10538, groups = var_10509, pad = hidden_states_279_pad_0, pad_type = hidden_states_279_pad_type_0, strides = var_10536, weight = up_blocks_3_resnets_0_conv1_weight_to_fp16, x = input_453_cast);
            tensor<int32, [2]> var_10544 = const()[name = tensor<string, []>("op_10544"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10546 = const()[name = tensor<string, []>("op_10546"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_39_pad_type_0 = const()[name = tensor<string, []>("temb_39_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_39_pad_0 = const()[name = tensor<string, []>("temb_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1280, 1, 1]> up_blocks_3_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [320, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1699007552)))];
            tensor<fp16, [320]> up_blocks_3_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1699826816)))];
            tensor<fp16, [2, 320, 1, 1]> temb_39_cast = conv(bias = up_blocks_3_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_10546, groups = var_10509, pad = temb_39_pad_0, pad_type = temb_39_pad_type_0, strides = var_10544, weight = up_blocks_3_resnets_0_time_emb_proj_weight_to_fp16, x = input_15_cast);
            tensor<fp16, [2, 320, 64, 64]> input_457_cast = add(x = hidden_states_279_cast, y = temb_39_cast);
            tensor<int32, [5]> reshape_208_shape_0 = const()[name = tensor<string, []>("reshape_208_shape_0"), val = tensor<int32, [5]>([2, 32, 10, 64, 64])];
            tensor<fp16, [2, 32, 10, 64, 64]> reshape_208_cast = reshape(shape = reshape_208_shape_0, x = input_457_cast);
            tensor<int32, [3]> reduce_mean_156_axes_0 = const()[name = tensor<string, []>("reduce_mean_156_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_156_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_156_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_156_cast = reduce_mean(axes = reduce_mean_156_axes_0, keep_dims = reduce_mean_156_keep_dims_0, x = reshape_208_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> sub_104_cast = sub(x = reshape_208_cast, y = reduce_mean_156_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> square_52_cast = square(x = sub_104_cast);
            tensor<int32, [3]> reduce_mean_158_axes_0 = const()[name = tensor<string, []>("reduce_mean_158_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_158_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_158_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_158_cast = reduce_mean(axes = reduce_mean_158_axes_0, keep_dims = reduce_mean_158_keep_dims_0, x = square_52_cast);
            tensor<fp16, []> add_104_y_0_to_fp16 = const()[name = tensor<string, []>("add_104_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_104_cast = add(x = reduce_mean_158_cast, y = add_104_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_52_cast = sqrt(x = add_104_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> real_div_52_cast = real_div(x = sub_104_cast, y = sqrt_52_cast);
            tensor<int32, [4]> reshape_209_shape_0 = const()[name = tensor<string, []>("reshape_209_shape_0"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> reshape_209_cast = reshape(shape = reshape_209_shape_0, x = real_div_52_cast);
            tensor<fp16, [320]> add_105_mean_0_to_fp16 = const()[name = tensor<string, []>("add_105_mean_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1699827520)))];
            tensor<fp16, [320]> add_105_variance_0_to_fp16 = const()[name = tensor<string, []>("add_105_variance_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1699828224)))];
            tensor<fp16, [320]> add_105_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_105_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1699828928)))];
            tensor<fp16, [320]> add_105_beta_0_to_fp16 = const()[name = tensor<string, []>("add_105_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1699829632)))];
            tensor<fp16, []> add_105_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_105_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 320, 64, 64]> add_105_cast = batch_norm(beta = add_105_beta_0_to_fp16, epsilon = add_105_epsilon_0_to_fp16, gamma = add_105_gamma_0_to_fp16, mean = add_105_mean_0_to_fp16, variance = add_105_variance_0_to_fp16, x = reshape_209_cast);
            tensor<fp16, [2, 320, 64, 64]> input_461_cast = silu(x = add_105_cast);
            tensor<int32, [2]> var_10556 = const()[name = tensor<string, []>("op_10556"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10558 = const()[name = tensor<string, []>("op_10558"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_281_pad_type_0 = const()[name = tensor<string, []>("hidden_states_281_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_281_pad_0 = const()[name = tensor<string, []>("hidden_states_281_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [320, 320, 3, 3]> up_blocks_3_resnets_0_conv2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_resnets_0_conv2_weight_to_fp16"), val = tensor<fp16, [320, 320, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1699830336)))];
            tensor<fp16, [320]> up_blocks_3_resnets_0_conv2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_resnets_0_conv2_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1701673600)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_281_cast = conv(bias = up_blocks_3_resnets_0_conv2_bias_to_fp16, dilations = var_10558, groups = var_10509, pad = hidden_states_281_pad_0, pad_type = hidden_states_281_pad_type_0, strides = var_10556, weight = up_blocks_3_resnets_0_conv2_weight_to_fp16, x = input_461_cast);
            tensor<int32, [2]> var_10563 = const()[name = tensor<string, []>("op_10563"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10565 = const()[name = tensor<string, []>("op_10565"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> x_23_pad_type_0 = const()[name = tensor<string, []>("x_23_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> x_23_pad_0 = const()[name = tensor<string, []>("x_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 960, 1, 1]> up_blocks_3_resnets_0_conv_shortcut_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_resnets_0_conv_shortcut_weight_to_fp16"), val = tensor<fp16, [320, 960, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1701674304)))];
            tensor<fp16, [320]> up_blocks_3_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1702288768)))];
            tensor<fp16, [2, 320, 64, 64]> x_23_cast = conv(bias = up_blocks_3_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_10565, groups = var_10509, pad = x_23_pad_0, pad_type = x_23_pad_type_0, strides = var_10563, weight = up_blocks_3_resnets_0_conv_shortcut_weight_to_fp16, x = input_449_cast);
            tensor<fp16, [2, 320, 64, 64]> hidden_states_283_cast = add(x = x_23_cast, y = hidden_states_281_cast);
            tensor<int32, [5]> reshape_212_shape_0 = const()[name = tensor<string, []>("reshape_212_shape_0"), val = tensor<int32, [5]>([2, 32, 10, 64, 64])];
            tensor<fp16, [2, 32, 10, 64, 64]> reshape_212_cast = reshape(shape = reshape_212_shape_0, x = hidden_states_283_cast);
            tensor<int32, [3]> reduce_mean_159_axes_0 = const()[name = tensor<string, []>("reduce_mean_159_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_159_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_159_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_159_cast = reduce_mean(axes = reduce_mean_159_axes_0, keep_dims = reduce_mean_159_keep_dims_0, x = reshape_212_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> sub_106_cast = sub(x = reshape_212_cast, y = reduce_mean_159_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> square_53_cast = square(x = sub_106_cast);
            tensor<int32, [3]> reduce_mean_161_axes_0 = const()[name = tensor<string, []>("reduce_mean_161_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_161_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_161_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_161_cast = reduce_mean(axes = reduce_mean_161_axes_0, keep_dims = reduce_mean_161_keep_dims_0, x = square_53_cast);
            tensor<fp16, []> add_106_y_0_to_fp16 = const()[name = tensor<string, []>("add_106_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_106_cast = add(x = reduce_mean_161_cast, y = add_106_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_53_cast = sqrt(x = add_106_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> real_div_53_cast = real_div(x = sub_106_cast, y = sqrt_53_cast);
            tensor<int32, [4]> reshape_213_shape_0 = const()[name = tensor<string, []>("reshape_213_shape_0"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> reshape_213_cast = reshape(shape = reshape_213_shape_0, x = real_div_53_cast);
            tensor<fp16, [320]> add_107_mean_0_to_fp16 = const()[name = tensor<string, []>("add_107_mean_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1702289472)))];
            tensor<fp16, [320]> add_107_variance_0_to_fp16 = const()[name = tensor<string, []>("add_107_variance_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1702290176)))];
            tensor<fp16, [320]> add_107_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_107_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1702290880)))];
            tensor<fp16, [320]> add_107_beta_0_to_fp16 = const()[name = tensor<string, []>("add_107_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1702291584)))];
            tensor<fp16, []> add_107_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_107_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 320, 64, 64]> add_107_cast = batch_norm(beta = add_107_beta_0_to_fp16, epsilon = add_107_epsilon_0_to_fp16, gamma = add_107_gamma_0_to_fp16, mean = add_107_mean_0_to_fp16, variance = add_107_variance_0_to_fp16, x = reshape_213_cast);
            tensor<int32, [2]> var_10585 = const()[name = tensor<string, []>("op_10585"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10587 = const()[name = tensor<string, []>("op_10587"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_285_pad_type_0 = const()[name = tensor<string, []>("hidden_states_285_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_285_pad_0 = const()[name = tensor<string, []>("hidden_states_285_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> up_blocks_3_attentions_0_proj_in_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_0_proj_in_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1702292288)))];
            tensor<fp16, [320]> up_blocks_3_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_0_proj_in_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1702497152)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_285_cast = conv(bias = up_blocks_3_attentions_0_proj_in_bias_to_fp16, dilations = var_10587, groups = var_10509, pad = hidden_states_285_pad_0, pad_type = hidden_states_285_pad_type_0, strides = var_10585, weight = up_blocks_3_attentions_0_proj_in_weight_to_fp16, x = add_107_cast);
            tensor<int32, [4]> var_10592 = const()[name = tensor<string, []>("op_10592"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<fp16, [2, 320, 1, 4096]> inputs_79_cast = reshape(shape = var_10592, x = hidden_states_285_cast);
            tensor<int32, [1]> var_10602 = const()[name = tensor<string, []>("op_10602"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_79_cast = reduce_mean(axes = var_10602, keep_dims = var_10504, x = inputs_79_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_79_cast = sub(x = inputs_79_cast, y = channels_mean_79_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_sq_79_cast = mul(x = zero_mean_79_cast, y = zero_mean_79_cast);
            tensor<int32, [1]> var_10606 = const()[name = tensor<string, []>("op_10606"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_10607_cast = reduce_mean(axes = var_10606, keep_dims = var_10504, x = zero_mean_sq_79_cast);
            tensor<fp16, []> var_10608_to_fp16 = const()[name = tensor<string, []>("op_10608_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_10609_cast = add(x = var_10607_cast, y = var_10608_to_fp16);
            tensor<fp16, []> denom_79_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_79_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_79_cast = rsqrt(epsilon = denom_79_epsilon_0_to_fp16, x = var_10609_cast);
            tensor<fp16, [2, 320, 1, 4096]> out_79_cast = mul(x = zero_mean_79_cast, y = denom_79_cast);
            tensor<fp16, [1, 320, 1, 1]> var_10613_to_fp16 = const()[name = tensor<string, []>("op_10613_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1702497856)))];
            tensor<fp16, [2, 320, 1, 4096]> var_10614_cast = add(x = out_79_cast, y = var_10613_to_fp16);
            tensor<fp16, [1, 320, 1, 1]> var_10616_to_fp16 = const()[name = tensor<string, []>("op_10616_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1702498560)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_287_cast = mul(x = var_10614_cast, y = var_10616_to_fp16);
            tensor<int32, [2]> var_10623 = const()[name = tensor<string, []>("op_10623"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10625 = const()[name = tensor<string, []>("op_10625"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_53_pad_type_0 = const()[name = tensor<string, []>("q_53_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_53_pad_0 = const()[name = tensor<string, []>("q_53_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1702499264)))];
            tensor<fp16, [2, 320, 1, 4096]> q_53_cast = conv(dilations = var_10625, groups = var_10509, pad = q_53_pad_0, pad_type = q_53_pad_type_0, strides = var_10623, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_287_cast);
            tensor<int32, [2]> var_10629 = const()[name = tensor<string, []>("op_10629"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10631 = const()[name = tensor<string, []>("op_10631"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_105_pad_type_0 = const()[name = tensor<string, []>("k_105_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_105_pad_0 = const()[name = tensor<string, []>("k_105_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1702704128)))];
            tensor<fp16, [2, 320, 1, 4096]> k_105_cast = conv(dilations = var_10631, groups = var_10509, pad = k_105_pad_0, pad_type = k_105_pad_type_0, strides = var_10629, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_287_cast);
            tensor<int32, [2]> var_10635 = const()[name = tensor<string, []>("op_10635"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10637 = const()[name = tensor<string, []>("op_10637"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_53_pad_type_0 = const()[name = tensor<string, []>("v_53_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_53_pad_0 = const()[name = tensor<string, []>("v_53_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1702908992)))];
            tensor<fp16, [2, 320, 1, 4096]> v_53_cast = conv(dilations = var_10637, groups = var_10509, pad = v_53_pad_0, pad_type = v_53_pad_type_0, strides = var_10635, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_287_cast);
            tensor<int32, [4]> var_10641_begin_0 = const()[name = tensor<string, []>("op_10641_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10641_end_0 = const()[name = tensor<string, []>("op_10641_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_10641_end_mask_0 = const()[name = tensor<string, []>("op_10641_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_10641_cast = slice_by_index(begin = var_10641_begin_0, end = var_10641_end_0, end_mask = var_10641_end_mask_0, x = q_53_cast);
            tensor<int32, [4]> var_10645_begin_0 = const()[name = tensor<string, []>("op_10645_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_10645_end_0 = const()[name = tensor<string, []>("op_10645_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_10645_end_mask_0 = const()[name = tensor<string, []>("op_10645_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_10645_cast = slice_by_index(begin = var_10645_begin_0, end = var_10645_end_0, end_mask = var_10645_end_mask_0, x = q_53_cast);
            tensor<int32, [4]> var_10649_begin_0 = const()[name = tensor<string, []>("op_10649_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_10649_end_0 = const()[name = tensor<string, []>("op_10649_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_10649_end_mask_0 = const()[name = tensor<string, []>("op_10649_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_10649_cast = slice_by_index(begin = var_10649_begin_0, end = var_10649_end_0, end_mask = var_10649_end_mask_0, x = q_53_cast);
            tensor<int32, [4]> var_10653_begin_0 = const()[name = tensor<string, []>("op_10653_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_10653_end_0 = const()[name = tensor<string, []>("op_10653_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_10653_end_mask_0 = const()[name = tensor<string, []>("op_10653_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_10653_cast = slice_by_index(begin = var_10653_begin_0, end = var_10653_end_0, end_mask = var_10653_end_mask_0, x = q_53_cast);
            tensor<int32, [4]> var_10657_begin_0 = const()[name = tensor<string, []>("op_10657_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_10657_end_0 = const()[name = tensor<string, []>("op_10657_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_10657_end_mask_0 = const()[name = tensor<string, []>("op_10657_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_10657_cast = slice_by_index(begin = var_10657_begin_0, end = var_10657_end_0, end_mask = var_10657_end_mask_0, x = q_53_cast);
            tensor<int32, [4]> k_107_perm_0 = const()[name = tensor<string, []>("k_107_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_10664_begin_0 = const()[name = tensor<string, []>("op_10664_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10664_end_0 = const()[name = tensor<string, []>("op_10664_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 64])];
            tensor<bool, [4]> var_10664_end_mask_0 = const()[name = tensor<string, []>("op_10664_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 320]> transpose_5 = transpose(perm = k_107_perm_0, x = k_105_cast);
            tensor<fp16, [2, 4096, 1, 64]> var_10664_cast = slice_by_index(begin = var_10664_begin_0, end = var_10664_end_0, end_mask = var_10664_end_mask_0, x = transpose_5);
            tensor<int32, [4]> var_10668_begin_0 = const()[name = tensor<string, []>("op_10668_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_10668_end_0 = const()[name = tensor<string, []>("op_10668_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 128])];
            tensor<bool, [4]> var_10668_end_mask_0 = const()[name = tensor<string, []>("op_10668_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_10668_cast = slice_by_index(begin = var_10668_begin_0, end = var_10668_end_0, end_mask = var_10668_end_mask_0, x = transpose_5);
            tensor<int32, [4]> var_10672_begin_0 = const()[name = tensor<string, []>("op_10672_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_10672_end_0 = const()[name = tensor<string, []>("op_10672_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 192])];
            tensor<bool, [4]> var_10672_end_mask_0 = const()[name = tensor<string, []>("op_10672_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_10672_cast = slice_by_index(begin = var_10672_begin_0, end = var_10672_end_0, end_mask = var_10672_end_mask_0, x = transpose_5);
            tensor<int32, [4]> var_10676_begin_0 = const()[name = tensor<string, []>("op_10676_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_10676_end_0 = const()[name = tensor<string, []>("op_10676_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 256])];
            tensor<bool, [4]> var_10676_end_mask_0 = const()[name = tensor<string, []>("op_10676_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_10676_cast = slice_by_index(begin = var_10676_begin_0, end = var_10676_end_0, end_mask = var_10676_end_mask_0, x = transpose_5);
            tensor<int32, [4]> var_10680_begin_0 = const()[name = tensor<string, []>("op_10680_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_10680_end_0 = const()[name = tensor<string, []>("op_10680_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_10680_end_mask_0 = const()[name = tensor<string, []>("op_10680_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_10680_cast = slice_by_index(begin = var_10680_begin_0, end = var_10680_end_0, end_mask = var_10680_end_mask_0, x = transpose_5);
            tensor<int32, [4]> var_10682_begin_0 = const()[name = tensor<string, []>("op_10682_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10682_end_0 = const()[name = tensor<string, []>("op_10682_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_10682_end_mask_0 = const()[name = tensor<string, []>("op_10682_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_10682_cast = slice_by_index(begin = var_10682_begin_0, end = var_10682_end_0, end_mask = var_10682_end_mask_0, x = v_53_cast);
            tensor<int32, [4]> var_10686_begin_0 = const()[name = tensor<string, []>("op_10686_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_10686_end_0 = const()[name = tensor<string, []>("op_10686_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_10686_end_mask_0 = const()[name = tensor<string, []>("op_10686_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_10686_cast = slice_by_index(begin = var_10686_begin_0, end = var_10686_end_0, end_mask = var_10686_end_mask_0, x = v_53_cast);
            tensor<int32, [4]> var_10690_begin_0 = const()[name = tensor<string, []>("op_10690_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_10690_end_0 = const()[name = tensor<string, []>("op_10690_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_10690_end_mask_0 = const()[name = tensor<string, []>("op_10690_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_10690_cast = slice_by_index(begin = var_10690_begin_0, end = var_10690_end_0, end_mask = var_10690_end_mask_0, x = v_53_cast);
            tensor<int32, [4]> var_10694_begin_0 = const()[name = tensor<string, []>("op_10694_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_10694_end_0 = const()[name = tensor<string, []>("op_10694_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_10694_end_mask_0 = const()[name = tensor<string, []>("op_10694_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_10694_cast = slice_by_index(begin = var_10694_begin_0, end = var_10694_end_0, end_mask = var_10694_end_mask_0, x = v_53_cast);
            tensor<int32, [4]> var_10698_begin_0 = const()[name = tensor<string, []>("op_10698_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_10698_end_0 = const()[name = tensor<string, []>("op_10698_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_10698_end_mask_0 = const()[name = tensor<string, []>("op_10698_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_10698_cast = slice_by_index(begin = var_10698_begin_0, end = var_10698_end_0, end_mask = var_10698_end_mask_0, x = v_53_cast);
            tensor<string, []> var_10702_equation_0 = const()[name = tensor<string, []>("op_10702_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_10702_cast = einsum(equation = var_10702_equation_0, values = (var_10664_cast, var_10641_cast));
            tensor<fp16, []> var_10703_to_fp16 = const()[name = tensor<string, []>("op_10703_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_721_cast = mul(x = var_10702_cast, y = var_10703_to_fp16);
            tensor<string, []> var_10706_equation_0 = const()[name = tensor<string, []>("op_10706_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_10706_cast = einsum(equation = var_10706_equation_0, values = (var_10668_cast, var_10645_cast));
            tensor<fp16, []> var_10707_to_fp16 = const()[name = tensor<string, []>("op_10707_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_723_cast = mul(x = var_10706_cast, y = var_10707_to_fp16);
            tensor<string, []> var_10710_equation_0 = const()[name = tensor<string, []>("op_10710_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_10710_cast = einsum(equation = var_10710_equation_0, values = (var_10672_cast, var_10649_cast));
            tensor<fp16, []> var_10711_to_fp16 = const()[name = tensor<string, []>("op_10711_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_725_cast = mul(x = var_10710_cast, y = var_10711_to_fp16);
            tensor<string, []> var_10714_equation_0 = const()[name = tensor<string, []>("op_10714_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_10714_cast = einsum(equation = var_10714_equation_0, values = (var_10676_cast, var_10653_cast));
            tensor<fp16, []> var_10715_to_fp16 = const()[name = tensor<string, []>("op_10715_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_727_cast = mul(x = var_10714_cast, y = var_10715_to_fp16);
            tensor<string, []> var_10718_equation_0 = const()[name = tensor<string, []>("op_10718_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_10718_cast = einsum(equation = var_10718_equation_0, values = (var_10680_cast, var_10657_cast));
            tensor<fp16, []> var_10719_to_fp16 = const()[name = tensor<string, []>("op_10719_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_729_cast = mul(x = var_10718_cast, y = var_10719_to_fp16);
            tensor<fp16, [2, 4096, 1, 4096]> var_10721_cast = softmax(axis = var_10509, x = aw_721_cast);
            tensor<fp16, [2, 4096, 1, 4096]> var_10722_cast = softmax(axis = var_10509, x = aw_723_cast);
            tensor<fp16, [2, 4096, 1, 4096]> var_10723_cast = softmax(axis = var_10509, x = aw_725_cast);
            tensor<fp16, [2, 4096, 1, 4096]> var_10724_cast = softmax(axis = var_10509, x = aw_727_cast);
            tensor<fp16, [2, 4096, 1, 4096]> var_10725_cast = softmax(axis = var_10509, x = aw_729_cast);
            tensor<string, []> var_10727_equation_0 = const()[name = tensor<string, []>("op_10727_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_10727_cast = einsum(equation = var_10727_equation_0, values = (var_10682_cast, var_10721_cast));
            tensor<string, []> var_10729_equation_0 = const()[name = tensor<string, []>("op_10729_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_10729_cast = einsum(equation = var_10729_equation_0, values = (var_10686_cast, var_10722_cast));
            tensor<string, []> var_10731_equation_0 = const()[name = tensor<string, []>("op_10731_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_10731_cast = einsum(equation = var_10731_equation_0, values = (var_10690_cast, var_10723_cast));
            tensor<string, []> var_10733_equation_0 = const()[name = tensor<string, []>("op_10733_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_10733_cast = einsum(equation = var_10733_equation_0, values = (var_10694_cast, var_10724_cast));
            tensor<string, []> var_10735_equation_0 = const()[name = tensor<string, []>("op_10735_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_10735_cast = einsum(equation = var_10735_equation_0, values = (var_10698_cast, var_10725_cast));
            tensor<bool, []> input_465_interleave_0 = const()[name = tensor<string, []>("input_465_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 320, 1, 4096]> input_465_cast = concat(axis = var_10509, interleave = input_465_interleave_0, values = (var_10727_cast, var_10729_cast, var_10731_cast, var_10733_cast, var_10735_cast));
            tensor<int32, [2]> var_10741 = const()[name = tensor<string, []>("op_10741"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10743 = const()[name = tensor<string, []>("op_10743"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_10745_pad_type_0 = const()[name = tensor<string, []>("op_10745_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_10745_pad_0 = const()[name = tensor<string, []>("op_10745_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1703113856)))];
            tensor<fp16, [320]> up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1703318720)))];
            tensor<fp16, [2, 320, 1, 4096]> var_10745_cast = conv(bias = up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_10743, groups = var_10509, pad = var_10745_pad_0, pad_type = var_10745_pad_type_0, strides = var_10741, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_465_cast);
            tensor<fp16, [2, 320, 1, 4096]> inputs_81_cast = add(x = var_10745_cast, y = inputs_79_cast);
            tensor<int32, [1]> var_10749 = const()[name = tensor<string, []>("op_10749"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_81_cast = reduce_mean(axes = var_10749, keep_dims = var_10504, x = inputs_81_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_81_cast = sub(x = inputs_81_cast, y = channels_mean_81_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_sq_81_cast = mul(x = zero_mean_81_cast, y = zero_mean_81_cast);
            tensor<int32, [1]> var_10753 = const()[name = tensor<string, []>("op_10753"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_10754_cast = reduce_mean(axes = var_10753, keep_dims = var_10504, x = zero_mean_sq_81_cast);
            tensor<fp16, []> var_10755_to_fp16 = const()[name = tensor<string, []>("op_10755_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_10756_cast = add(x = var_10754_cast, y = var_10755_to_fp16);
            tensor<fp16, []> denom_81_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_81_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_81_cast = rsqrt(epsilon = denom_81_epsilon_0_to_fp16, x = var_10756_cast);
            tensor<fp16, [2, 320, 1, 4096]> out_81_cast = mul(x = zero_mean_81_cast, y = denom_81_cast);
            tensor<fp16, [1, 320, 1, 1]> var_10760_to_fp16 = const()[name = tensor<string, []>("op_10760_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1703319424)))];
            tensor<fp16, [2, 320, 1, 4096]> var_10761_cast = add(x = out_81_cast, y = var_10760_to_fp16);
            tensor<fp16, [1, 320, 1, 1]> var_10763_to_fp16 = const()[name = tensor<string, []>("op_10763_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1703320128)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_289_cast = mul(x = var_10761_cast, y = var_10763_to_fp16);
            tensor<int32, [2]> var_10770 = const()[name = tensor<string, []>("op_10770"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10772 = const()[name = tensor<string, []>("op_10772"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_55_pad_type_0 = const()[name = tensor<string, []>("q_55_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_55_pad_0 = const()[name = tensor<string, []>("q_55_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1703320832)))];
            tensor<fp16, [2, 320, 1, 4096]> q_55_cast = conv(dilations = var_10772, groups = var_10509, pad = q_55_pad_0, pad_type = q_55_pad_type_0, strides = var_10770, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_289_cast);
            tensor<int32, [2]> var_10776 = const()[name = tensor<string, []>("op_10776"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10778 = const()[name = tensor<string, []>("op_10778"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_109_pad_type_0 = const()[name = tensor<string, []>("k_109_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_109_pad_0 = const()[name = tensor<string, []>("k_109_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1024, 1, 1]> up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [320, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1703525696)))];
            tensor<fp16, [2, 320, 1, 77]> k_109_cast = conv(dilations = var_10778, groups = var_10509, pad = k_109_pad_0, pad_type = k_109_pad_type_0, strides = var_10776, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [2]> var_10782 = const()[name = tensor<string, []>("op_10782"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10784 = const()[name = tensor<string, []>("op_10784"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_55_pad_type_0 = const()[name = tensor<string, []>("v_55_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_55_pad_0 = const()[name = tensor<string, []>("v_55_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1024, 1, 1]> up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [320, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1704181120)))];
            tensor<fp16, [2, 320, 1, 77]> v_55_cast = conv(dilations = var_10784, groups = var_10509, pad = v_55_pad_0, pad_type = v_55_pad_type_0, strides = var_10782, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [4]> var_10788_begin_0 = const()[name = tensor<string, []>("op_10788_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10788_end_0 = const()[name = tensor<string, []>("op_10788_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_10788_end_mask_0 = const()[name = tensor<string, []>("op_10788_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_10788_cast = slice_by_index(begin = var_10788_begin_0, end = var_10788_end_0, end_mask = var_10788_end_mask_0, x = q_55_cast);
            tensor<int32, [4]> var_10792_begin_0 = const()[name = tensor<string, []>("op_10792_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_10792_end_0 = const()[name = tensor<string, []>("op_10792_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_10792_end_mask_0 = const()[name = tensor<string, []>("op_10792_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_10792_cast = slice_by_index(begin = var_10792_begin_0, end = var_10792_end_0, end_mask = var_10792_end_mask_0, x = q_55_cast);
            tensor<int32, [4]> var_10796_begin_0 = const()[name = tensor<string, []>("op_10796_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_10796_end_0 = const()[name = tensor<string, []>("op_10796_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_10796_end_mask_0 = const()[name = tensor<string, []>("op_10796_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_10796_cast = slice_by_index(begin = var_10796_begin_0, end = var_10796_end_0, end_mask = var_10796_end_mask_0, x = q_55_cast);
            tensor<int32, [4]> var_10800_begin_0 = const()[name = tensor<string, []>("op_10800_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_10800_end_0 = const()[name = tensor<string, []>("op_10800_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_10800_end_mask_0 = const()[name = tensor<string, []>("op_10800_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_10800_cast = slice_by_index(begin = var_10800_begin_0, end = var_10800_end_0, end_mask = var_10800_end_mask_0, x = q_55_cast);
            tensor<int32, [4]> var_10804_begin_0 = const()[name = tensor<string, []>("op_10804_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_10804_end_0 = const()[name = tensor<string, []>("op_10804_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_10804_end_mask_0 = const()[name = tensor<string, []>("op_10804_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_10804_cast = slice_by_index(begin = var_10804_begin_0, end = var_10804_end_0, end_mask = var_10804_end_mask_0, x = q_55_cast);
            tensor<int32, [4]> k_111_perm_0 = const()[name = tensor<string, []>("k_111_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_10811_begin_0 = const()[name = tensor<string, []>("op_10811_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10811_end_0 = const()[name = tensor<string, []>("op_10811_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_10811_end_mask_0 = const()[name = tensor<string, []>("op_10811_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 320]> transpose_4 = transpose(perm = k_111_perm_0, x = k_109_cast);
            tensor<fp16, [2, 77, 1, 64]> var_10811_cast = slice_by_index(begin = var_10811_begin_0, end = var_10811_end_0, end_mask = var_10811_end_mask_0, x = transpose_4);
            tensor<int32, [4]> var_10815_begin_0 = const()[name = tensor<string, []>("op_10815_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_10815_end_0 = const()[name = tensor<string, []>("op_10815_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_10815_end_mask_0 = const()[name = tensor<string, []>("op_10815_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10815_cast = slice_by_index(begin = var_10815_begin_0, end = var_10815_end_0, end_mask = var_10815_end_mask_0, x = transpose_4);
            tensor<int32, [4]> var_10819_begin_0 = const()[name = tensor<string, []>("op_10819_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_10819_end_0 = const()[name = tensor<string, []>("op_10819_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_10819_end_mask_0 = const()[name = tensor<string, []>("op_10819_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10819_cast = slice_by_index(begin = var_10819_begin_0, end = var_10819_end_0, end_mask = var_10819_end_mask_0, x = transpose_4);
            tensor<int32, [4]> var_10823_begin_0 = const()[name = tensor<string, []>("op_10823_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_10823_end_0 = const()[name = tensor<string, []>("op_10823_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_10823_end_mask_0 = const()[name = tensor<string, []>("op_10823_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10823_cast = slice_by_index(begin = var_10823_begin_0, end = var_10823_end_0, end_mask = var_10823_end_mask_0, x = transpose_4);
            tensor<int32, [4]> var_10827_begin_0 = const()[name = tensor<string, []>("op_10827_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_10827_end_0 = const()[name = tensor<string, []>("op_10827_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_10827_end_mask_0 = const()[name = tensor<string, []>("op_10827_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10827_cast = slice_by_index(begin = var_10827_begin_0, end = var_10827_end_0, end_mask = var_10827_end_mask_0, x = transpose_4);
            tensor<int32, [4]> var_10829_begin_0 = const()[name = tensor<string, []>("op_10829_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10829_end_0 = const()[name = tensor<string, []>("op_10829_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_10829_end_mask_0 = const()[name = tensor<string, []>("op_10829_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10829_cast = slice_by_index(begin = var_10829_begin_0, end = var_10829_end_0, end_mask = var_10829_end_mask_0, x = v_55_cast);
            tensor<int32, [4]> var_10833_begin_0 = const()[name = tensor<string, []>("op_10833_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_10833_end_0 = const()[name = tensor<string, []>("op_10833_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_10833_end_mask_0 = const()[name = tensor<string, []>("op_10833_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10833_cast = slice_by_index(begin = var_10833_begin_0, end = var_10833_end_0, end_mask = var_10833_end_mask_0, x = v_55_cast);
            tensor<int32, [4]> var_10837_begin_0 = const()[name = tensor<string, []>("op_10837_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_10837_end_0 = const()[name = tensor<string, []>("op_10837_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_10837_end_mask_0 = const()[name = tensor<string, []>("op_10837_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10837_cast = slice_by_index(begin = var_10837_begin_0, end = var_10837_end_0, end_mask = var_10837_end_mask_0, x = v_55_cast);
            tensor<int32, [4]> var_10841_begin_0 = const()[name = tensor<string, []>("op_10841_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_10841_end_0 = const()[name = tensor<string, []>("op_10841_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_10841_end_mask_0 = const()[name = tensor<string, []>("op_10841_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10841_cast = slice_by_index(begin = var_10841_begin_0, end = var_10841_end_0, end_mask = var_10841_end_mask_0, x = v_55_cast);
            tensor<int32, [4]> var_10845_begin_0 = const()[name = tensor<string, []>("op_10845_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_10845_end_0 = const()[name = tensor<string, []>("op_10845_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_10845_end_mask_0 = const()[name = tensor<string, []>("op_10845_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10845_cast = slice_by_index(begin = var_10845_begin_0, end = var_10845_end_0, end_mask = var_10845_end_mask_0, x = v_55_cast);
            tensor<string, []> var_10849_equation_0 = const()[name = tensor<string, []>("op_10849_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_10849_cast = einsum(equation = var_10849_equation_0, values = (var_10811_cast, var_10788_cast));
            tensor<fp16, []> var_10850_to_fp16 = const()[name = tensor<string, []>("op_10850_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_731_cast = mul(x = var_10849_cast, y = var_10850_to_fp16);
            tensor<string, []> var_10853_equation_0 = const()[name = tensor<string, []>("op_10853_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_10853_cast = einsum(equation = var_10853_equation_0, values = (var_10815_cast, var_10792_cast));
            tensor<fp16, []> var_10854_to_fp16 = const()[name = tensor<string, []>("op_10854_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_733_cast = mul(x = var_10853_cast, y = var_10854_to_fp16);
            tensor<string, []> var_10857_equation_0 = const()[name = tensor<string, []>("op_10857_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_10857_cast = einsum(equation = var_10857_equation_0, values = (var_10819_cast, var_10796_cast));
            tensor<fp16, []> var_10858_to_fp16 = const()[name = tensor<string, []>("op_10858_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_735_cast = mul(x = var_10857_cast, y = var_10858_to_fp16);
            tensor<string, []> var_10861_equation_0 = const()[name = tensor<string, []>("op_10861_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_10861_cast = einsum(equation = var_10861_equation_0, values = (var_10823_cast, var_10800_cast));
            tensor<fp16, []> var_10862_to_fp16 = const()[name = tensor<string, []>("op_10862_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_737_cast = mul(x = var_10861_cast, y = var_10862_to_fp16);
            tensor<string, []> var_10865_equation_0 = const()[name = tensor<string, []>("op_10865_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_10865_cast = einsum(equation = var_10865_equation_0, values = (var_10827_cast, var_10804_cast));
            tensor<fp16, []> var_10866_to_fp16 = const()[name = tensor<string, []>("op_10866_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_739_cast = mul(x = var_10865_cast, y = var_10866_to_fp16);
            tensor<fp16, [2, 77, 1, 4096]> var_10868_cast = softmax(axis = var_10509, x = aw_731_cast);
            tensor<fp16, [2, 77, 1, 4096]> var_10869_cast = softmax(axis = var_10509, x = aw_733_cast);
            tensor<fp16, [2, 77, 1, 4096]> var_10870_cast = softmax(axis = var_10509, x = aw_735_cast);
            tensor<fp16, [2, 77, 1, 4096]> var_10871_cast = softmax(axis = var_10509, x = aw_737_cast);
            tensor<fp16, [2, 77, 1, 4096]> var_10872_cast = softmax(axis = var_10509, x = aw_739_cast);
            tensor<string, []> var_10874_equation_0 = const()[name = tensor<string, []>("op_10874_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_10874_cast = einsum(equation = var_10874_equation_0, values = (var_10829_cast, var_10868_cast));
            tensor<string, []> var_10876_equation_0 = const()[name = tensor<string, []>("op_10876_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_10876_cast = einsum(equation = var_10876_equation_0, values = (var_10833_cast, var_10869_cast));
            tensor<string, []> var_10878_equation_0 = const()[name = tensor<string, []>("op_10878_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_10878_cast = einsum(equation = var_10878_equation_0, values = (var_10837_cast, var_10870_cast));
            tensor<string, []> var_10880_equation_0 = const()[name = tensor<string, []>("op_10880_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_10880_cast = einsum(equation = var_10880_equation_0, values = (var_10841_cast, var_10871_cast));
            tensor<string, []> var_10882_equation_0 = const()[name = tensor<string, []>("op_10882_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_10882_cast = einsum(equation = var_10882_equation_0, values = (var_10845_cast, var_10872_cast));
            tensor<bool, []> input_467_interleave_0 = const()[name = tensor<string, []>("input_467_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 320, 1, 4096]> input_467_cast = concat(axis = var_10509, interleave = input_467_interleave_0, values = (var_10874_cast, var_10876_cast, var_10878_cast, var_10880_cast, var_10882_cast));
            tensor<int32, [2]> var_10888 = const()[name = tensor<string, []>("op_10888"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10890 = const()[name = tensor<string, []>("op_10890"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_10892_pad_type_0 = const()[name = tensor<string, []>("op_10892_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_10892_pad_0 = const()[name = tensor<string, []>("op_10892_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1704836544)))];
            tensor<fp16, [320]> up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1705041408)))];
            tensor<fp16, [2, 320, 1, 4096]> var_10892_cast = conv(bias = up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_10890, groups = var_10509, pad = var_10892_pad_0, pad_type = var_10892_pad_type_0, strides = var_10888, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_467_cast);
            tensor<fp16, [2, 320, 1, 4096]> inputs_83_cast = add(x = var_10892_cast, y = inputs_81_cast);
            tensor<int32, [1]> var_10896 = const()[name = tensor<string, []>("op_10896"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_83_cast = reduce_mean(axes = var_10896, keep_dims = var_10504, x = inputs_83_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_83_cast = sub(x = inputs_83_cast, y = channels_mean_83_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_sq_83_cast = mul(x = zero_mean_83_cast, y = zero_mean_83_cast);
            tensor<int32, [1]> var_10900 = const()[name = tensor<string, []>("op_10900"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_10901_cast = reduce_mean(axes = var_10900, keep_dims = var_10504, x = zero_mean_sq_83_cast);
            tensor<fp16, []> var_10902_to_fp16 = const()[name = tensor<string, []>("op_10902_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_10903_cast = add(x = var_10901_cast, y = var_10902_to_fp16);
            tensor<fp16, []> denom_83_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_83_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_83_cast = rsqrt(epsilon = denom_83_epsilon_0_to_fp16, x = var_10903_cast);
            tensor<fp16, [2, 320, 1, 4096]> out_83_cast = mul(x = zero_mean_83_cast, y = denom_83_cast);
            tensor<fp16, [1, 320, 1, 1]> var_10907_to_fp16 = const()[name = tensor<string, []>("op_10907_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1705042112)))];
            tensor<fp16, [2, 320, 1, 4096]> var_10908_cast = add(x = out_83_cast, y = var_10907_to_fp16);
            tensor<fp16, [1, 320, 1, 1]> var_10910_to_fp16 = const()[name = tensor<string, []>("op_10910_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1705042816)))];
            tensor<fp16, [2, 320, 1, 4096]> input_469_cast = mul(x = var_10908_cast, y = var_10910_to_fp16);
            tensor<int32, [2]> var_10918 = const()[name = tensor<string, []>("op_10918"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10920 = const()[name = tensor<string, []>("op_10920"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_10922_pad_type_0 = const()[name = tensor<string, []>("op_10922_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_10922_pad_0 = const()[name = tensor<string, []>("op_10922_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [2560, 320, 1, 1]> up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [2560, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1705043520)))];
            tensor<fp16, [2560]> up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1706681984)))];
            tensor<fp16, [2, 2560, 1, 4096]> var_10922_cast = conv(bias = up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_10920, groups = var_10509, pad = var_10922_pad_0, pad_type = var_10922_pad_type_0, strides = var_10918, weight = up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_469_cast);
            tensor<int32, [2]> var_10923_split_sizes_0 = const()[name = tensor<string, []>("op_10923_split_sizes_0"), val = tensor<int32, [2]>([1280, 1280])];
            tensor<int32, []> var_10923_axis_0 = const()[name = tensor<string, []>("op_10923_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 1280, 1, 4096]> var_10923_cast_0, tensor<fp16, [2, 1280, 1, 4096]> var_10923_cast_1 = split(axis = var_10923_axis_0, split_sizes = var_10923_split_sizes_0, x = var_10922_cast);
            tensor<string, []> var_10925_mode_0 = const()[name = tensor<string, []>("op_10925_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 1280, 1, 4096]> var_10925_cast = gelu(mode = var_10925_mode_0, x = var_10923_cast_1);
            tensor<fp16, [2, 1280, 1, 4096]> input_471_cast = mul(x = var_10923_cast_0, y = var_10925_cast);
            tensor<int32, [2]> var_10929 = const()[name = tensor<string, []>("op_10929"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10931 = const()[name = tensor<string, []>("op_10931"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_10933_pad_type_0 = const()[name = tensor<string, []>("op_10933_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_10933_pad_0 = const()[name = tensor<string, []>("op_10933_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1280, 1, 1]> up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [320, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1706687168)))];
            tensor<fp16, [320]> up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1707506432)))];
            tensor<fp16, [2, 320, 1, 4096]> var_10933_cast = conv(bias = up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_10931, groups = var_10509, pad = var_10933_pad_0, pad_type = var_10933_pad_type_0, strides = var_10929, weight = up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_471_cast);
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_293_cast = add(x = var_10933_cast, y = inputs_83_cast);
            tensor<int32, [4]> var_10935 = const()[name = tensor<string, []>("op_10935"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> input_473_cast = reshape(shape = var_10935, x = hidden_states_293_cast);
            tensor<int32, [2]> var_10939 = const()[name = tensor<string, []>("op_10939"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10941 = const()[name = tensor<string, []>("op_10941"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_295_pad_type_0 = const()[name = tensor<string, []>("hidden_states_295_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_295_pad_0 = const()[name = tensor<string, []>("hidden_states_295_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> up_blocks_3_attentions_0_proj_out_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_0_proj_out_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1707507136)))];
            tensor<fp16, [320]> up_blocks_3_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_0_proj_out_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1707712000)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_295_cast = conv(bias = up_blocks_3_attentions_0_proj_out_bias_to_fp16, dilations = var_10941, groups = var_10509, pad = hidden_states_295_pad_0, pad_type = hidden_states_295_pad_type_0, strides = var_10939, weight = up_blocks_3_attentions_0_proj_out_weight_to_fp16, x = input_473_cast);
            tensor<fp16, [2, 320, 64, 64]> hidden_states_297_cast = add(x = hidden_states_295_cast, y = hidden_states_283_cast);
            tensor<bool, []> input_475_interleave_0 = const()[name = tensor<string, []>("input_475_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 64, 64]> input_475_cast = concat(axis = var_10509, interleave = input_475_interleave_0, values = (hidden_states_297_cast, input_35_cast));
            tensor<int32, [5]> reshape_216_shape_0 = const()[name = tensor<string, []>("reshape_216_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 64, 64])];
            tensor<fp16, [2, 32, 20, 64, 64]> reshape_216_cast = reshape(shape = reshape_216_shape_0, x = input_475_cast);
            tensor<int32, [3]> reduce_mean_162_axes_0 = const()[name = tensor<string, []>("reduce_mean_162_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_162_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_162_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_162_cast = reduce_mean(axes = reduce_mean_162_axes_0, keep_dims = reduce_mean_162_keep_dims_0, x = reshape_216_cast);
            tensor<fp16, [2, 32, 20, 64, 64]> sub_108_cast = sub(x = reshape_216_cast, y = reduce_mean_162_cast);
            tensor<fp16, [2, 32, 20, 64, 64]> square_54_cast = square(x = sub_108_cast);
            tensor<int32, [3]> reduce_mean_164_axes_0 = const()[name = tensor<string, []>("reduce_mean_164_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_164_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_164_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_164_cast = reduce_mean(axes = reduce_mean_164_axes_0, keep_dims = reduce_mean_164_keep_dims_0, x = square_54_cast);
            tensor<fp16, []> add_108_y_0_to_fp16 = const()[name = tensor<string, []>("add_108_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_108_cast = add(x = reduce_mean_164_cast, y = add_108_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_54_cast = sqrt(x = add_108_cast);
            tensor<fp16, [2, 32, 20, 64, 64]> real_div_54_cast = real_div(x = sub_108_cast, y = sqrt_54_cast);
            tensor<int32, [4]> reshape_217_shape_0 = const()[name = tensor<string, []>("reshape_217_shape_0"), val = tensor<int32, [4]>([2, 640, 64, 64])];
            tensor<fp16, [2, 640, 64, 64]> reshape_217_cast = reshape(shape = reshape_217_shape_0, x = real_div_54_cast);
            tensor<fp16, [640]> add_109_mean_0_to_fp16 = const()[name = tensor<string, []>("add_109_mean_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1707712704)))];
            tensor<fp16, [640]> add_109_variance_0_to_fp16 = const()[name = tensor<string, []>("add_109_variance_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1707714048)))];
            tensor<fp16, [640]> add_109_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_109_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1707715392)))];
            tensor<fp16, [640]> add_109_beta_0_to_fp16 = const()[name = tensor<string, []>("add_109_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1707716736)))];
            tensor<fp16, []> add_109_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_109_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 64, 64]> add_109_cast = batch_norm(beta = add_109_beta_0_to_fp16, epsilon = add_109_epsilon_0_to_fp16, gamma = add_109_gamma_0_to_fp16, mean = add_109_mean_0_to_fp16, variance = add_109_variance_0_to_fp16, x = reshape_217_cast);
            tensor<fp16, [2, 640, 64, 64]> input_479_cast = silu(x = add_109_cast);
            tensor<int32, [2]> var_10959 = const()[name = tensor<string, []>("op_10959"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10961 = const()[name = tensor<string, []>("op_10961"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_299_pad_type_0 = const()[name = tensor<string, []>("hidden_states_299_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_299_pad_0 = const()[name = tensor<string, []>("hidden_states_299_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [320, 640, 3, 3]> up_blocks_3_resnets_1_conv1_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_resnets_1_conv1_weight_to_fp16"), val = tensor<fp16, [320, 640, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1707718080)))];
            tensor<fp16, [320]> up_blocks_3_resnets_1_conv1_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_resnets_1_conv1_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1711404544)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_299_cast = conv(bias = up_blocks_3_resnets_1_conv1_bias_to_fp16, dilations = var_10961, groups = var_10509, pad = hidden_states_299_pad_0, pad_type = hidden_states_299_pad_type_0, strides = var_10959, weight = up_blocks_3_resnets_1_conv1_weight_to_fp16, x = input_479_cast);
            tensor<int32, [2]> var_10967 = const()[name = tensor<string, []>("op_10967"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10969 = const()[name = tensor<string, []>("op_10969"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_41_pad_type_0 = const()[name = tensor<string, []>("temb_41_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_41_pad_0 = const()[name = tensor<string, []>("temb_41_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1280, 1, 1]> up_blocks_3_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [320, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1711405248)))];
            tensor<fp16, [320]> up_blocks_3_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1712224512)))];
            tensor<fp16, [2, 320, 1, 1]> temb_41_cast = conv(bias = up_blocks_3_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_10969, groups = var_10509, pad = temb_41_pad_0, pad_type = temb_41_pad_type_0, strides = var_10967, weight = up_blocks_3_resnets_1_time_emb_proj_weight_to_fp16, x = input_15_cast);
            tensor<fp16, [2, 320, 64, 64]> input_483_cast = add(x = hidden_states_299_cast, y = temb_41_cast);
            tensor<int32, [5]> reshape_220_shape_0 = const()[name = tensor<string, []>("reshape_220_shape_0"), val = tensor<int32, [5]>([2, 32, 10, 64, 64])];
            tensor<fp16, [2, 32, 10, 64, 64]> reshape_220_cast = reshape(shape = reshape_220_shape_0, x = input_483_cast);
            tensor<int32, [3]> reduce_mean_165_axes_0 = const()[name = tensor<string, []>("reduce_mean_165_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_165_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_165_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_165_cast = reduce_mean(axes = reduce_mean_165_axes_0, keep_dims = reduce_mean_165_keep_dims_0, x = reshape_220_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> sub_110_cast = sub(x = reshape_220_cast, y = reduce_mean_165_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> square_55_cast = square(x = sub_110_cast);
            tensor<int32, [3]> reduce_mean_167_axes_0 = const()[name = tensor<string, []>("reduce_mean_167_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_167_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_167_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_167_cast = reduce_mean(axes = reduce_mean_167_axes_0, keep_dims = reduce_mean_167_keep_dims_0, x = square_55_cast);
            tensor<fp16, []> add_110_y_0_to_fp16 = const()[name = tensor<string, []>("add_110_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_110_cast = add(x = reduce_mean_167_cast, y = add_110_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_55_cast = sqrt(x = add_110_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> real_div_55_cast = real_div(x = sub_110_cast, y = sqrt_55_cast);
            tensor<int32, [4]> reshape_221_shape_0 = const()[name = tensor<string, []>("reshape_221_shape_0"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> reshape_221_cast = reshape(shape = reshape_221_shape_0, x = real_div_55_cast);
            tensor<fp16, [320]> add_111_mean_0_to_fp16 = const()[name = tensor<string, []>("add_111_mean_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1712225216)))];
            tensor<fp16, [320]> add_111_variance_0_to_fp16 = const()[name = tensor<string, []>("add_111_variance_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1712225920)))];
            tensor<fp16, [320]> add_111_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_111_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1712226624)))];
            tensor<fp16, [320]> add_111_beta_0_to_fp16 = const()[name = tensor<string, []>("add_111_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1712227328)))];
            tensor<fp16, []> add_111_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_111_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 320, 64, 64]> add_111_cast = batch_norm(beta = add_111_beta_0_to_fp16, epsilon = add_111_epsilon_0_to_fp16, gamma = add_111_gamma_0_to_fp16, mean = add_111_mean_0_to_fp16, variance = add_111_variance_0_to_fp16, x = reshape_221_cast);
            tensor<fp16, [2, 320, 64, 64]> input_487_cast = silu(x = add_111_cast);
            tensor<int32, [2]> var_10979 = const()[name = tensor<string, []>("op_10979"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10981 = const()[name = tensor<string, []>("op_10981"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_301_pad_type_0 = const()[name = tensor<string, []>("hidden_states_301_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_301_pad_0 = const()[name = tensor<string, []>("hidden_states_301_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [320, 320, 3, 3]> up_blocks_3_resnets_1_conv2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_resnets_1_conv2_weight_to_fp16"), val = tensor<fp16, [320, 320, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1712228032)))];
            tensor<fp16, [320]> up_blocks_3_resnets_1_conv2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_resnets_1_conv2_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1714071296)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_301_cast = conv(bias = up_blocks_3_resnets_1_conv2_bias_to_fp16, dilations = var_10981, groups = var_10509, pad = hidden_states_301_pad_0, pad_type = hidden_states_301_pad_type_0, strides = var_10979, weight = up_blocks_3_resnets_1_conv2_weight_to_fp16, x = input_487_cast);
            tensor<int32, [2]> var_10986 = const()[name = tensor<string, []>("op_10986"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10988 = const()[name = tensor<string, []>("op_10988"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> x_25_pad_type_0 = const()[name = tensor<string, []>("x_25_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> x_25_pad_0 = const()[name = tensor<string, []>("x_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 640, 1, 1]> up_blocks_3_resnets_1_conv_shortcut_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_resnets_1_conv_shortcut_weight_to_fp16"), val = tensor<fp16, [320, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1714072000)))];
            tensor<fp16, [320]> up_blocks_3_resnets_1_conv_shortcut_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_resnets_1_conv_shortcut_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1714481664)))];
            tensor<fp16, [2, 320, 64, 64]> x_25_cast = conv(bias = up_blocks_3_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_10988, groups = var_10509, pad = x_25_pad_0, pad_type = x_25_pad_type_0, strides = var_10986, weight = up_blocks_3_resnets_1_conv_shortcut_weight_to_fp16, x = input_475_cast);
            tensor<fp16, [2, 320, 64, 64]> hidden_states_303_cast = add(x = x_25_cast, y = hidden_states_301_cast);
            tensor<int32, [5]> reshape_224_shape_0 = const()[name = tensor<string, []>("reshape_224_shape_0"), val = tensor<int32, [5]>([2, 32, 10, 64, 64])];
            tensor<fp16, [2, 32, 10, 64, 64]> reshape_224_cast = reshape(shape = reshape_224_shape_0, x = hidden_states_303_cast);
            tensor<int32, [3]> reduce_mean_168_axes_0 = const()[name = tensor<string, []>("reduce_mean_168_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_168_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_168_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_168_cast = reduce_mean(axes = reduce_mean_168_axes_0, keep_dims = reduce_mean_168_keep_dims_0, x = reshape_224_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> sub_112_cast = sub(x = reshape_224_cast, y = reduce_mean_168_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> square_56_cast = square(x = sub_112_cast);
            tensor<int32, [3]> reduce_mean_170_axes_0 = const()[name = tensor<string, []>("reduce_mean_170_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_170_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_170_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_170_cast = reduce_mean(axes = reduce_mean_170_axes_0, keep_dims = reduce_mean_170_keep_dims_0, x = square_56_cast);
            tensor<fp16, []> add_112_y_0_to_fp16 = const()[name = tensor<string, []>("add_112_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_112_cast = add(x = reduce_mean_170_cast, y = add_112_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_56_cast = sqrt(x = add_112_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> real_div_56_cast = real_div(x = sub_112_cast, y = sqrt_56_cast);
            tensor<int32, [4]> reshape_225_shape_0 = const()[name = tensor<string, []>("reshape_225_shape_0"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> reshape_225_cast = reshape(shape = reshape_225_shape_0, x = real_div_56_cast);
            tensor<fp16, [320]> add_113_mean_0_to_fp16 = const()[name = tensor<string, []>("add_113_mean_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1714482368)))];
            tensor<fp16, [320]> add_113_variance_0_to_fp16 = const()[name = tensor<string, []>("add_113_variance_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1714483072)))];
            tensor<fp16, [320]> add_113_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_113_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1714483776)))];
            tensor<fp16, [320]> add_113_beta_0_to_fp16 = const()[name = tensor<string, []>("add_113_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1714484480)))];
            tensor<fp16, []> add_113_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_113_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 320, 64, 64]> add_113_cast = batch_norm(beta = add_113_beta_0_to_fp16, epsilon = add_113_epsilon_0_to_fp16, gamma = add_113_gamma_0_to_fp16, mean = add_113_mean_0_to_fp16, variance = add_113_variance_0_to_fp16, x = reshape_225_cast);
            tensor<int32, [2]> var_11008 = const()[name = tensor<string, []>("op_11008"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11010 = const()[name = tensor<string, []>("op_11010"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_305_pad_type_0 = const()[name = tensor<string, []>("hidden_states_305_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_305_pad_0 = const()[name = tensor<string, []>("hidden_states_305_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> up_blocks_3_attentions_1_proj_in_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_1_proj_in_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1714485184)))];
            tensor<fp16, [320]> up_blocks_3_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_1_proj_in_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1714690048)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_305_cast = conv(bias = up_blocks_3_attentions_1_proj_in_bias_to_fp16, dilations = var_11010, groups = var_10509, pad = hidden_states_305_pad_0, pad_type = hidden_states_305_pad_type_0, strides = var_11008, weight = up_blocks_3_attentions_1_proj_in_weight_to_fp16, x = add_113_cast);
            tensor<int32, [4]> var_11015 = const()[name = tensor<string, []>("op_11015"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<fp16, [2, 320, 1, 4096]> inputs_85_cast = reshape(shape = var_11015, x = hidden_states_305_cast);
            tensor<int32, [1]> var_11025 = const()[name = tensor<string, []>("op_11025"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_85_cast = reduce_mean(axes = var_11025, keep_dims = var_10504, x = inputs_85_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_85_cast = sub(x = inputs_85_cast, y = channels_mean_85_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_sq_85_cast = mul(x = zero_mean_85_cast, y = zero_mean_85_cast);
            tensor<int32, [1]> var_11029 = const()[name = tensor<string, []>("op_11029"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_11030_cast = reduce_mean(axes = var_11029, keep_dims = var_10504, x = zero_mean_sq_85_cast);
            tensor<fp16, []> var_11031_to_fp16 = const()[name = tensor<string, []>("op_11031_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_11032_cast = add(x = var_11030_cast, y = var_11031_to_fp16);
            tensor<fp16, []> denom_85_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_85_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_85_cast = rsqrt(epsilon = denom_85_epsilon_0_to_fp16, x = var_11032_cast);
            tensor<fp16, [2, 320, 1, 4096]> out_85_cast = mul(x = zero_mean_85_cast, y = denom_85_cast);
            tensor<fp16, [1, 320, 1, 1]> var_11036_to_fp16 = const()[name = tensor<string, []>("op_11036_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1714690752)))];
            tensor<fp16, [2, 320, 1, 4096]> var_11037_cast = add(x = out_85_cast, y = var_11036_to_fp16);
            tensor<fp16, [1, 320, 1, 1]> var_11039_to_fp16 = const()[name = tensor<string, []>("op_11039_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1714691456)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_307_cast = mul(x = var_11037_cast, y = var_11039_to_fp16);
            tensor<int32, [2]> var_11046 = const()[name = tensor<string, []>("op_11046"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11048 = const()[name = tensor<string, []>("op_11048"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_57_pad_type_0 = const()[name = tensor<string, []>("q_57_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_57_pad_0 = const()[name = tensor<string, []>("q_57_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1714692160)))];
            tensor<fp16, [2, 320, 1, 4096]> q_57_cast = conv(dilations = var_11048, groups = var_10509, pad = q_57_pad_0, pad_type = q_57_pad_type_0, strides = var_11046, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_307_cast);
            tensor<int32, [2]> var_11052 = const()[name = tensor<string, []>("op_11052"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11054 = const()[name = tensor<string, []>("op_11054"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_113_pad_type_0 = const()[name = tensor<string, []>("k_113_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_113_pad_0 = const()[name = tensor<string, []>("k_113_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1714897024)))];
            tensor<fp16, [2, 320, 1, 4096]> k_113_cast = conv(dilations = var_11054, groups = var_10509, pad = k_113_pad_0, pad_type = k_113_pad_type_0, strides = var_11052, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_307_cast);
            tensor<int32, [2]> var_11058 = const()[name = tensor<string, []>("op_11058"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11060 = const()[name = tensor<string, []>("op_11060"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_57_pad_type_0 = const()[name = tensor<string, []>("v_57_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_57_pad_0 = const()[name = tensor<string, []>("v_57_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1715101888)))];
            tensor<fp16, [2, 320, 1, 4096]> v_57_cast = conv(dilations = var_11060, groups = var_10509, pad = v_57_pad_0, pad_type = v_57_pad_type_0, strides = var_11058, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_307_cast);
            tensor<int32, [4]> var_11064_begin_0 = const()[name = tensor<string, []>("op_11064_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11064_end_0 = const()[name = tensor<string, []>("op_11064_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_11064_end_mask_0 = const()[name = tensor<string, []>("op_11064_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11064_cast = slice_by_index(begin = var_11064_begin_0, end = var_11064_end_0, end_mask = var_11064_end_mask_0, x = q_57_cast);
            tensor<int32, [4]> var_11068_begin_0 = const()[name = tensor<string, []>("op_11068_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_11068_end_0 = const()[name = tensor<string, []>("op_11068_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_11068_end_mask_0 = const()[name = tensor<string, []>("op_11068_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11068_cast = slice_by_index(begin = var_11068_begin_0, end = var_11068_end_0, end_mask = var_11068_end_mask_0, x = q_57_cast);
            tensor<int32, [4]> var_11072_begin_0 = const()[name = tensor<string, []>("op_11072_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_11072_end_0 = const()[name = tensor<string, []>("op_11072_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_11072_end_mask_0 = const()[name = tensor<string, []>("op_11072_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11072_cast = slice_by_index(begin = var_11072_begin_0, end = var_11072_end_0, end_mask = var_11072_end_mask_0, x = q_57_cast);
            tensor<int32, [4]> var_11076_begin_0 = const()[name = tensor<string, []>("op_11076_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_11076_end_0 = const()[name = tensor<string, []>("op_11076_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_11076_end_mask_0 = const()[name = tensor<string, []>("op_11076_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11076_cast = slice_by_index(begin = var_11076_begin_0, end = var_11076_end_0, end_mask = var_11076_end_mask_0, x = q_57_cast);
            tensor<int32, [4]> var_11080_begin_0 = const()[name = tensor<string, []>("op_11080_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_11080_end_0 = const()[name = tensor<string, []>("op_11080_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_11080_end_mask_0 = const()[name = tensor<string, []>("op_11080_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11080_cast = slice_by_index(begin = var_11080_begin_0, end = var_11080_end_0, end_mask = var_11080_end_mask_0, x = q_57_cast);
            tensor<int32, [4]> k_115_perm_0 = const()[name = tensor<string, []>("k_115_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_11087_begin_0 = const()[name = tensor<string, []>("op_11087_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11087_end_0 = const()[name = tensor<string, []>("op_11087_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 64])];
            tensor<bool, [4]> var_11087_end_mask_0 = const()[name = tensor<string, []>("op_11087_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 320]> transpose_3 = transpose(perm = k_115_perm_0, x = k_113_cast);
            tensor<fp16, [2, 4096, 1, 64]> var_11087_cast = slice_by_index(begin = var_11087_begin_0, end = var_11087_end_0, end_mask = var_11087_end_mask_0, x = transpose_3);
            tensor<int32, [4]> var_11091_begin_0 = const()[name = tensor<string, []>("op_11091_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_11091_end_0 = const()[name = tensor<string, []>("op_11091_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 128])];
            tensor<bool, [4]> var_11091_end_mask_0 = const()[name = tensor<string, []>("op_11091_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_11091_cast = slice_by_index(begin = var_11091_begin_0, end = var_11091_end_0, end_mask = var_11091_end_mask_0, x = transpose_3);
            tensor<int32, [4]> var_11095_begin_0 = const()[name = tensor<string, []>("op_11095_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_11095_end_0 = const()[name = tensor<string, []>("op_11095_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 192])];
            tensor<bool, [4]> var_11095_end_mask_0 = const()[name = tensor<string, []>("op_11095_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_11095_cast = slice_by_index(begin = var_11095_begin_0, end = var_11095_end_0, end_mask = var_11095_end_mask_0, x = transpose_3);
            tensor<int32, [4]> var_11099_begin_0 = const()[name = tensor<string, []>("op_11099_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_11099_end_0 = const()[name = tensor<string, []>("op_11099_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 256])];
            tensor<bool, [4]> var_11099_end_mask_0 = const()[name = tensor<string, []>("op_11099_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_11099_cast = slice_by_index(begin = var_11099_begin_0, end = var_11099_end_0, end_mask = var_11099_end_mask_0, x = transpose_3);
            tensor<int32, [4]> var_11103_begin_0 = const()[name = tensor<string, []>("op_11103_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_11103_end_0 = const()[name = tensor<string, []>("op_11103_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_11103_end_mask_0 = const()[name = tensor<string, []>("op_11103_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_11103_cast = slice_by_index(begin = var_11103_begin_0, end = var_11103_end_0, end_mask = var_11103_end_mask_0, x = transpose_3);
            tensor<int32, [4]> var_11105_begin_0 = const()[name = tensor<string, []>("op_11105_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11105_end_0 = const()[name = tensor<string, []>("op_11105_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_11105_end_mask_0 = const()[name = tensor<string, []>("op_11105_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11105_cast = slice_by_index(begin = var_11105_begin_0, end = var_11105_end_0, end_mask = var_11105_end_mask_0, x = v_57_cast);
            tensor<int32, [4]> var_11109_begin_0 = const()[name = tensor<string, []>("op_11109_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_11109_end_0 = const()[name = tensor<string, []>("op_11109_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_11109_end_mask_0 = const()[name = tensor<string, []>("op_11109_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11109_cast = slice_by_index(begin = var_11109_begin_0, end = var_11109_end_0, end_mask = var_11109_end_mask_0, x = v_57_cast);
            tensor<int32, [4]> var_11113_begin_0 = const()[name = tensor<string, []>("op_11113_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_11113_end_0 = const()[name = tensor<string, []>("op_11113_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_11113_end_mask_0 = const()[name = tensor<string, []>("op_11113_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11113_cast = slice_by_index(begin = var_11113_begin_0, end = var_11113_end_0, end_mask = var_11113_end_mask_0, x = v_57_cast);
            tensor<int32, [4]> var_11117_begin_0 = const()[name = tensor<string, []>("op_11117_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_11117_end_0 = const()[name = tensor<string, []>("op_11117_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_11117_end_mask_0 = const()[name = tensor<string, []>("op_11117_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11117_cast = slice_by_index(begin = var_11117_begin_0, end = var_11117_end_0, end_mask = var_11117_end_mask_0, x = v_57_cast);
            tensor<int32, [4]> var_11121_begin_0 = const()[name = tensor<string, []>("op_11121_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_11121_end_0 = const()[name = tensor<string, []>("op_11121_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_11121_end_mask_0 = const()[name = tensor<string, []>("op_11121_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11121_cast = slice_by_index(begin = var_11121_begin_0, end = var_11121_end_0, end_mask = var_11121_end_mask_0, x = v_57_cast);
            tensor<string, []> var_11125_equation_0 = const()[name = tensor<string, []>("op_11125_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_11125_cast = einsum(equation = var_11125_equation_0, values = (var_11087_cast, var_11064_cast));
            tensor<fp16, []> var_11126_to_fp16 = const()[name = tensor<string, []>("op_11126_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_741_cast = mul(x = var_11125_cast, y = var_11126_to_fp16);
            tensor<string, []> var_11129_equation_0 = const()[name = tensor<string, []>("op_11129_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_11129_cast = einsum(equation = var_11129_equation_0, values = (var_11091_cast, var_11068_cast));
            tensor<fp16, []> var_11130_to_fp16 = const()[name = tensor<string, []>("op_11130_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_743_cast = mul(x = var_11129_cast, y = var_11130_to_fp16);
            tensor<string, []> var_11133_equation_0 = const()[name = tensor<string, []>("op_11133_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_11133_cast = einsum(equation = var_11133_equation_0, values = (var_11095_cast, var_11072_cast));
            tensor<fp16, []> var_11134_to_fp16 = const()[name = tensor<string, []>("op_11134_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_745_cast = mul(x = var_11133_cast, y = var_11134_to_fp16);
            tensor<string, []> var_11137_equation_0 = const()[name = tensor<string, []>("op_11137_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_11137_cast = einsum(equation = var_11137_equation_0, values = (var_11099_cast, var_11076_cast));
            tensor<fp16, []> var_11138_to_fp16 = const()[name = tensor<string, []>("op_11138_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_747_cast = mul(x = var_11137_cast, y = var_11138_to_fp16);
            tensor<string, []> var_11141_equation_0 = const()[name = tensor<string, []>("op_11141_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_11141_cast = einsum(equation = var_11141_equation_0, values = (var_11103_cast, var_11080_cast));
            tensor<fp16, []> var_11142_to_fp16 = const()[name = tensor<string, []>("op_11142_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_749_cast = mul(x = var_11141_cast, y = var_11142_to_fp16);
            tensor<fp16, [2, 4096, 1, 4096]> var_11144_cast = softmax(axis = var_10509, x = aw_741_cast);
            tensor<fp16, [2, 4096, 1, 4096]> var_11145_cast = softmax(axis = var_10509, x = aw_743_cast);
            tensor<fp16, [2, 4096, 1, 4096]> var_11146_cast = softmax(axis = var_10509, x = aw_745_cast);
            tensor<fp16, [2, 4096, 1, 4096]> var_11147_cast = softmax(axis = var_10509, x = aw_747_cast);
            tensor<fp16, [2, 4096, 1, 4096]> var_11148_cast = softmax(axis = var_10509, x = aw_749_cast);
            tensor<string, []> var_11150_equation_0 = const()[name = tensor<string, []>("op_11150_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_11150_cast = einsum(equation = var_11150_equation_0, values = (var_11105_cast, var_11144_cast));
            tensor<string, []> var_11152_equation_0 = const()[name = tensor<string, []>("op_11152_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_11152_cast = einsum(equation = var_11152_equation_0, values = (var_11109_cast, var_11145_cast));
            tensor<string, []> var_11154_equation_0 = const()[name = tensor<string, []>("op_11154_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_11154_cast = einsum(equation = var_11154_equation_0, values = (var_11113_cast, var_11146_cast));
            tensor<string, []> var_11156_equation_0 = const()[name = tensor<string, []>("op_11156_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_11156_cast = einsum(equation = var_11156_equation_0, values = (var_11117_cast, var_11147_cast));
            tensor<string, []> var_11158_equation_0 = const()[name = tensor<string, []>("op_11158_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_11158_cast = einsum(equation = var_11158_equation_0, values = (var_11121_cast, var_11148_cast));
            tensor<bool, []> input_491_interleave_0 = const()[name = tensor<string, []>("input_491_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 320, 1, 4096]> input_491_cast = concat(axis = var_10509, interleave = input_491_interleave_0, values = (var_11150_cast, var_11152_cast, var_11154_cast, var_11156_cast, var_11158_cast));
            tensor<int32, [2]> var_11164 = const()[name = tensor<string, []>("op_11164"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11166 = const()[name = tensor<string, []>("op_11166"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_11168_pad_type_0 = const()[name = tensor<string, []>("op_11168_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_11168_pad_0 = const()[name = tensor<string, []>("op_11168_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1715306752)))];
            tensor<fp16, [320]> up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1715511616)))];
            tensor<fp16, [2, 320, 1, 4096]> var_11168_cast = conv(bias = up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_11166, groups = var_10509, pad = var_11168_pad_0, pad_type = var_11168_pad_type_0, strides = var_11164, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_491_cast);
            tensor<fp16, [2, 320, 1, 4096]> inputs_87_cast = add(x = var_11168_cast, y = inputs_85_cast);
            tensor<int32, [1]> var_11172 = const()[name = tensor<string, []>("op_11172"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_87_cast = reduce_mean(axes = var_11172, keep_dims = var_10504, x = inputs_87_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_87_cast = sub(x = inputs_87_cast, y = channels_mean_87_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_sq_87_cast = mul(x = zero_mean_87_cast, y = zero_mean_87_cast);
            tensor<int32, [1]> var_11176 = const()[name = tensor<string, []>("op_11176"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_11177_cast = reduce_mean(axes = var_11176, keep_dims = var_10504, x = zero_mean_sq_87_cast);
            tensor<fp16, []> var_11178_to_fp16 = const()[name = tensor<string, []>("op_11178_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_11179_cast = add(x = var_11177_cast, y = var_11178_to_fp16);
            tensor<fp16, []> denom_87_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_87_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_87_cast = rsqrt(epsilon = denom_87_epsilon_0_to_fp16, x = var_11179_cast);
            tensor<fp16, [2, 320, 1, 4096]> out_87_cast = mul(x = zero_mean_87_cast, y = denom_87_cast);
            tensor<fp16, [1, 320, 1, 1]> var_11183_to_fp16 = const()[name = tensor<string, []>("op_11183_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1715512320)))];
            tensor<fp16, [2, 320, 1, 4096]> var_11184_cast = add(x = out_87_cast, y = var_11183_to_fp16);
            tensor<fp16, [1, 320, 1, 1]> var_11186_to_fp16 = const()[name = tensor<string, []>("op_11186_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1715513024)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_309_cast = mul(x = var_11184_cast, y = var_11186_to_fp16);
            tensor<int32, [2]> var_11193 = const()[name = tensor<string, []>("op_11193"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11195 = const()[name = tensor<string, []>("op_11195"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_59_pad_type_0 = const()[name = tensor<string, []>("q_59_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_59_pad_0 = const()[name = tensor<string, []>("q_59_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1715513728)))];
            tensor<fp16, [2, 320, 1, 4096]> q_59_cast = conv(dilations = var_11195, groups = var_10509, pad = q_59_pad_0, pad_type = q_59_pad_type_0, strides = var_11193, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_309_cast);
            tensor<int32, [2]> var_11199 = const()[name = tensor<string, []>("op_11199"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11201 = const()[name = tensor<string, []>("op_11201"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_117_pad_type_0 = const()[name = tensor<string, []>("k_117_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_117_pad_0 = const()[name = tensor<string, []>("k_117_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1024, 1, 1]> up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [320, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1715718592)))];
            tensor<fp16, [2, 320, 1, 77]> k_117_cast = conv(dilations = var_11201, groups = var_10509, pad = k_117_pad_0, pad_type = k_117_pad_type_0, strides = var_11199, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [2]> var_11205 = const()[name = tensor<string, []>("op_11205"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11207 = const()[name = tensor<string, []>("op_11207"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_59_pad_type_0 = const()[name = tensor<string, []>("v_59_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_59_pad_0 = const()[name = tensor<string, []>("v_59_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1024, 1, 1]> up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [320, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1716374016)))];
            tensor<fp16, [2, 320, 1, 77]> v_59_cast = conv(dilations = var_11207, groups = var_10509, pad = v_59_pad_0, pad_type = v_59_pad_type_0, strides = var_11205, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [4]> var_11211_begin_0 = const()[name = tensor<string, []>("op_11211_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11211_end_0 = const()[name = tensor<string, []>("op_11211_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_11211_end_mask_0 = const()[name = tensor<string, []>("op_11211_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11211_cast = slice_by_index(begin = var_11211_begin_0, end = var_11211_end_0, end_mask = var_11211_end_mask_0, x = q_59_cast);
            tensor<int32, [4]> var_11215_begin_0 = const()[name = tensor<string, []>("op_11215_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_11215_end_0 = const()[name = tensor<string, []>("op_11215_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_11215_end_mask_0 = const()[name = tensor<string, []>("op_11215_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11215_cast = slice_by_index(begin = var_11215_begin_0, end = var_11215_end_0, end_mask = var_11215_end_mask_0, x = q_59_cast);
            tensor<int32, [4]> var_11219_begin_0 = const()[name = tensor<string, []>("op_11219_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_11219_end_0 = const()[name = tensor<string, []>("op_11219_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_11219_end_mask_0 = const()[name = tensor<string, []>("op_11219_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11219_cast = slice_by_index(begin = var_11219_begin_0, end = var_11219_end_0, end_mask = var_11219_end_mask_0, x = q_59_cast);
            tensor<int32, [4]> var_11223_begin_0 = const()[name = tensor<string, []>("op_11223_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_11223_end_0 = const()[name = tensor<string, []>("op_11223_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_11223_end_mask_0 = const()[name = tensor<string, []>("op_11223_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11223_cast = slice_by_index(begin = var_11223_begin_0, end = var_11223_end_0, end_mask = var_11223_end_mask_0, x = q_59_cast);
            tensor<int32, [4]> var_11227_begin_0 = const()[name = tensor<string, []>("op_11227_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_11227_end_0 = const()[name = tensor<string, []>("op_11227_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_11227_end_mask_0 = const()[name = tensor<string, []>("op_11227_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11227_cast = slice_by_index(begin = var_11227_begin_0, end = var_11227_end_0, end_mask = var_11227_end_mask_0, x = q_59_cast);
            tensor<int32, [4]> k_119_perm_0 = const()[name = tensor<string, []>("k_119_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_11234_begin_0 = const()[name = tensor<string, []>("op_11234_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11234_end_0 = const()[name = tensor<string, []>("op_11234_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_11234_end_mask_0 = const()[name = tensor<string, []>("op_11234_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 320]> transpose_2 = transpose(perm = k_119_perm_0, x = k_117_cast);
            tensor<fp16, [2, 77, 1, 64]> var_11234_cast = slice_by_index(begin = var_11234_begin_0, end = var_11234_end_0, end_mask = var_11234_end_mask_0, x = transpose_2);
            tensor<int32, [4]> var_11238_begin_0 = const()[name = tensor<string, []>("op_11238_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_11238_end_0 = const()[name = tensor<string, []>("op_11238_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_11238_end_mask_0 = const()[name = tensor<string, []>("op_11238_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11238_cast = slice_by_index(begin = var_11238_begin_0, end = var_11238_end_0, end_mask = var_11238_end_mask_0, x = transpose_2);
            tensor<int32, [4]> var_11242_begin_0 = const()[name = tensor<string, []>("op_11242_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_11242_end_0 = const()[name = tensor<string, []>("op_11242_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_11242_end_mask_0 = const()[name = tensor<string, []>("op_11242_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11242_cast = slice_by_index(begin = var_11242_begin_0, end = var_11242_end_0, end_mask = var_11242_end_mask_0, x = transpose_2);
            tensor<int32, [4]> var_11246_begin_0 = const()[name = tensor<string, []>("op_11246_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_11246_end_0 = const()[name = tensor<string, []>("op_11246_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_11246_end_mask_0 = const()[name = tensor<string, []>("op_11246_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11246_cast = slice_by_index(begin = var_11246_begin_0, end = var_11246_end_0, end_mask = var_11246_end_mask_0, x = transpose_2);
            tensor<int32, [4]> var_11250_begin_0 = const()[name = tensor<string, []>("op_11250_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_11250_end_0 = const()[name = tensor<string, []>("op_11250_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_11250_end_mask_0 = const()[name = tensor<string, []>("op_11250_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11250_cast = slice_by_index(begin = var_11250_begin_0, end = var_11250_end_0, end_mask = var_11250_end_mask_0, x = transpose_2);
            tensor<int32, [4]> var_11252_begin_0 = const()[name = tensor<string, []>("op_11252_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11252_end_0 = const()[name = tensor<string, []>("op_11252_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_11252_end_mask_0 = const()[name = tensor<string, []>("op_11252_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11252_cast = slice_by_index(begin = var_11252_begin_0, end = var_11252_end_0, end_mask = var_11252_end_mask_0, x = v_59_cast);
            tensor<int32, [4]> var_11256_begin_0 = const()[name = tensor<string, []>("op_11256_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_11256_end_0 = const()[name = tensor<string, []>("op_11256_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_11256_end_mask_0 = const()[name = tensor<string, []>("op_11256_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11256_cast = slice_by_index(begin = var_11256_begin_0, end = var_11256_end_0, end_mask = var_11256_end_mask_0, x = v_59_cast);
            tensor<int32, [4]> var_11260_begin_0 = const()[name = tensor<string, []>("op_11260_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_11260_end_0 = const()[name = tensor<string, []>("op_11260_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_11260_end_mask_0 = const()[name = tensor<string, []>("op_11260_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11260_cast = slice_by_index(begin = var_11260_begin_0, end = var_11260_end_0, end_mask = var_11260_end_mask_0, x = v_59_cast);
            tensor<int32, [4]> var_11264_begin_0 = const()[name = tensor<string, []>("op_11264_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_11264_end_0 = const()[name = tensor<string, []>("op_11264_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_11264_end_mask_0 = const()[name = tensor<string, []>("op_11264_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11264_cast = slice_by_index(begin = var_11264_begin_0, end = var_11264_end_0, end_mask = var_11264_end_mask_0, x = v_59_cast);
            tensor<int32, [4]> var_11268_begin_0 = const()[name = tensor<string, []>("op_11268_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_11268_end_0 = const()[name = tensor<string, []>("op_11268_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_11268_end_mask_0 = const()[name = tensor<string, []>("op_11268_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11268_cast = slice_by_index(begin = var_11268_begin_0, end = var_11268_end_0, end_mask = var_11268_end_mask_0, x = v_59_cast);
            tensor<string, []> var_11272_equation_0 = const()[name = tensor<string, []>("op_11272_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_11272_cast = einsum(equation = var_11272_equation_0, values = (var_11234_cast, var_11211_cast));
            tensor<fp16, []> var_11273_to_fp16 = const()[name = tensor<string, []>("op_11273_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_751_cast = mul(x = var_11272_cast, y = var_11273_to_fp16);
            tensor<string, []> var_11276_equation_0 = const()[name = tensor<string, []>("op_11276_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_11276_cast = einsum(equation = var_11276_equation_0, values = (var_11238_cast, var_11215_cast));
            tensor<fp16, []> var_11277_to_fp16 = const()[name = tensor<string, []>("op_11277_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_753_cast = mul(x = var_11276_cast, y = var_11277_to_fp16);
            tensor<string, []> var_11280_equation_0 = const()[name = tensor<string, []>("op_11280_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_11280_cast = einsum(equation = var_11280_equation_0, values = (var_11242_cast, var_11219_cast));
            tensor<fp16, []> var_11281_to_fp16 = const()[name = tensor<string, []>("op_11281_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_755_cast = mul(x = var_11280_cast, y = var_11281_to_fp16);
            tensor<string, []> var_11284_equation_0 = const()[name = tensor<string, []>("op_11284_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_11284_cast = einsum(equation = var_11284_equation_0, values = (var_11246_cast, var_11223_cast));
            tensor<fp16, []> var_11285_to_fp16 = const()[name = tensor<string, []>("op_11285_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_757_cast = mul(x = var_11284_cast, y = var_11285_to_fp16);
            tensor<string, []> var_11288_equation_0 = const()[name = tensor<string, []>("op_11288_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_11288_cast = einsum(equation = var_11288_equation_0, values = (var_11250_cast, var_11227_cast));
            tensor<fp16, []> var_11289_to_fp16 = const()[name = tensor<string, []>("op_11289_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_759_cast = mul(x = var_11288_cast, y = var_11289_to_fp16);
            tensor<fp16, [2, 77, 1, 4096]> var_11291_cast = softmax(axis = var_10509, x = aw_751_cast);
            tensor<fp16, [2, 77, 1, 4096]> var_11292_cast = softmax(axis = var_10509, x = aw_753_cast);
            tensor<fp16, [2, 77, 1, 4096]> var_11293_cast = softmax(axis = var_10509, x = aw_755_cast);
            tensor<fp16, [2, 77, 1, 4096]> var_11294_cast = softmax(axis = var_10509, x = aw_757_cast);
            tensor<fp16, [2, 77, 1, 4096]> var_11295_cast = softmax(axis = var_10509, x = aw_759_cast);
            tensor<string, []> var_11297_equation_0 = const()[name = tensor<string, []>("op_11297_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_11297_cast = einsum(equation = var_11297_equation_0, values = (var_11252_cast, var_11291_cast));
            tensor<string, []> var_11299_equation_0 = const()[name = tensor<string, []>("op_11299_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_11299_cast = einsum(equation = var_11299_equation_0, values = (var_11256_cast, var_11292_cast));
            tensor<string, []> var_11301_equation_0 = const()[name = tensor<string, []>("op_11301_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_11301_cast = einsum(equation = var_11301_equation_0, values = (var_11260_cast, var_11293_cast));
            tensor<string, []> var_11303_equation_0 = const()[name = tensor<string, []>("op_11303_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_11303_cast = einsum(equation = var_11303_equation_0, values = (var_11264_cast, var_11294_cast));
            tensor<string, []> var_11305_equation_0 = const()[name = tensor<string, []>("op_11305_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_11305_cast = einsum(equation = var_11305_equation_0, values = (var_11268_cast, var_11295_cast));
            tensor<bool, []> input_493_interleave_0 = const()[name = tensor<string, []>("input_493_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 320, 1, 4096]> input_493_cast = concat(axis = var_10509, interleave = input_493_interleave_0, values = (var_11297_cast, var_11299_cast, var_11301_cast, var_11303_cast, var_11305_cast));
            tensor<int32, [2]> var_11311 = const()[name = tensor<string, []>("op_11311"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11313 = const()[name = tensor<string, []>("op_11313"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_11315_pad_type_0 = const()[name = tensor<string, []>("op_11315_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_11315_pad_0 = const()[name = tensor<string, []>("op_11315_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1717029440)))];
            tensor<fp16, [320]> up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1717234304)))];
            tensor<fp16, [2, 320, 1, 4096]> var_11315_cast = conv(bias = up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_11313, groups = var_10509, pad = var_11315_pad_0, pad_type = var_11315_pad_type_0, strides = var_11311, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_493_cast);
            tensor<fp16, [2, 320, 1, 4096]> inputs_89_cast = add(x = var_11315_cast, y = inputs_87_cast);
            tensor<int32, [1]> var_11319 = const()[name = tensor<string, []>("op_11319"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_89_cast = reduce_mean(axes = var_11319, keep_dims = var_10504, x = inputs_89_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_89_cast = sub(x = inputs_89_cast, y = channels_mean_89_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_sq_89_cast = mul(x = zero_mean_89_cast, y = zero_mean_89_cast);
            tensor<int32, [1]> var_11323 = const()[name = tensor<string, []>("op_11323"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_11324_cast = reduce_mean(axes = var_11323, keep_dims = var_10504, x = zero_mean_sq_89_cast);
            tensor<fp16, []> var_11325_to_fp16 = const()[name = tensor<string, []>("op_11325_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_11326_cast = add(x = var_11324_cast, y = var_11325_to_fp16);
            tensor<fp16, []> denom_89_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_89_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_89_cast = rsqrt(epsilon = denom_89_epsilon_0_to_fp16, x = var_11326_cast);
            tensor<fp16, [2, 320, 1, 4096]> out_89_cast = mul(x = zero_mean_89_cast, y = denom_89_cast);
            tensor<fp16, [1, 320, 1, 1]> var_11330_to_fp16 = const()[name = tensor<string, []>("op_11330_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1717235008)))];
            tensor<fp16, [2, 320, 1, 4096]> var_11331_cast = add(x = out_89_cast, y = var_11330_to_fp16);
            tensor<fp16, [1, 320, 1, 1]> var_11333_to_fp16 = const()[name = tensor<string, []>("op_11333_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1717235712)))];
            tensor<fp16, [2, 320, 1, 4096]> input_495_cast = mul(x = var_11331_cast, y = var_11333_to_fp16);
            tensor<int32, [2]> var_11341 = const()[name = tensor<string, []>("op_11341"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11343 = const()[name = tensor<string, []>("op_11343"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_11345_pad_type_0 = const()[name = tensor<string, []>("op_11345_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_11345_pad_0 = const()[name = tensor<string, []>("op_11345_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [2560, 320, 1, 1]> up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [2560, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1717236416)))];
            tensor<fp16, [2560]> up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1718874880)))];
            tensor<fp16, [2, 2560, 1, 4096]> var_11345_cast = conv(bias = up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_11343, groups = var_10509, pad = var_11345_pad_0, pad_type = var_11345_pad_type_0, strides = var_11341, weight = up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_495_cast);
            tensor<int32, [2]> var_11346_split_sizes_0 = const()[name = tensor<string, []>("op_11346_split_sizes_0"), val = tensor<int32, [2]>([1280, 1280])];
            tensor<int32, []> var_11346_axis_0 = const()[name = tensor<string, []>("op_11346_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 1280, 1, 4096]> var_11346_cast_0, tensor<fp16, [2, 1280, 1, 4096]> var_11346_cast_1 = split(axis = var_11346_axis_0, split_sizes = var_11346_split_sizes_0, x = var_11345_cast);
            tensor<string, []> var_11348_mode_0 = const()[name = tensor<string, []>("op_11348_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 1280, 1, 4096]> var_11348_cast = gelu(mode = var_11348_mode_0, x = var_11346_cast_1);
            tensor<fp16, [2, 1280, 1, 4096]> input_497_cast = mul(x = var_11346_cast_0, y = var_11348_cast);
            tensor<int32, [2]> var_11352 = const()[name = tensor<string, []>("op_11352"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11354 = const()[name = tensor<string, []>("op_11354"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_11356_pad_type_0 = const()[name = tensor<string, []>("op_11356_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_11356_pad_0 = const()[name = tensor<string, []>("op_11356_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1280, 1, 1]> up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [320, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1718880064)))];
            tensor<fp16, [320]> up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1719699328)))];
            tensor<fp16, [2, 320, 1, 4096]> var_11356_cast = conv(bias = up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_11354, groups = var_10509, pad = var_11356_pad_0, pad_type = var_11356_pad_type_0, strides = var_11352, weight = up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_497_cast);
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_313_cast = add(x = var_11356_cast, y = inputs_89_cast);
            tensor<int32, [4]> var_11358 = const()[name = tensor<string, []>("op_11358"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> input_499_cast = reshape(shape = var_11358, x = hidden_states_313_cast);
            tensor<int32, [2]> var_11362 = const()[name = tensor<string, []>("op_11362"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11364 = const()[name = tensor<string, []>("op_11364"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_315_pad_type_0 = const()[name = tensor<string, []>("hidden_states_315_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_315_pad_0 = const()[name = tensor<string, []>("hidden_states_315_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> up_blocks_3_attentions_1_proj_out_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_1_proj_out_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1719700032)))];
            tensor<fp16, [320]> up_blocks_3_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_1_proj_out_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1719904896)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_315_cast = conv(bias = up_blocks_3_attentions_1_proj_out_bias_to_fp16, dilations = var_11364, groups = var_10509, pad = hidden_states_315_pad_0, pad_type = hidden_states_315_pad_type_0, strides = var_11362, weight = up_blocks_3_attentions_1_proj_out_weight_to_fp16, x = input_499_cast);
            tensor<fp16, [2, 320, 64, 64]> hidden_states_317_cast = add(x = hidden_states_315_cast, y = hidden_states_303_cast);
            tensor<bool, []> input_501_interleave_0 = const()[name = tensor<string, []>("input_501_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 640, 64, 64]> input_501_cast = concat(axis = var_10509, interleave = input_501_interleave_0, values = (hidden_states_317_cast, input_7_cast));
            tensor<int32, [5]> reshape_228_shape_0 = const()[name = tensor<string, []>("reshape_228_shape_0"), val = tensor<int32, [5]>([2, 32, 20, 64, 64])];
            tensor<fp16, [2, 32, 20, 64, 64]> reshape_228_cast = reshape(shape = reshape_228_shape_0, x = input_501_cast);
            tensor<int32, [3]> reduce_mean_171_axes_0 = const()[name = tensor<string, []>("reduce_mean_171_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_171_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_171_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_171_cast = reduce_mean(axes = reduce_mean_171_axes_0, keep_dims = reduce_mean_171_keep_dims_0, x = reshape_228_cast);
            tensor<fp16, [2, 32, 20, 64, 64]> sub_114_cast = sub(x = reshape_228_cast, y = reduce_mean_171_cast);
            tensor<fp16, [2, 32, 20, 64, 64]> square_57_cast = square(x = sub_114_cast);
            tensor<int32, [3]> reduce_mean_173_axes_0 = const()[name = tensor<string, []>("reduce_mean_173_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_173_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_173_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_173_cast = reduce_mean(axes = reduce_mean_173_axes_0, keep_dims = reduce_mean_173_keep_dims_0, x = square_57_cast);
            tensor<fp16, []> add_114_y_0_to_fp16 = const()[name = tensor<string, []>("add_114_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_114_cast = add(x = reduce_mean_173_cast, y = add_114_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_57_cast = sqrt(x = add_114_cast);
            tensor<fp16, [2, 32, 20, 64, 64]> real_div_57_cast = real_div(x = sub_114_cast, y = sqrt_57_cast);
            tensor<int32, [4]> reshape_229_shape_0 = const()[name = tensor<string, []>("reshape_229_shape_0"), val = tensor<int32, [4]>([2, 640, 64, 64])];
            tensor<fp16, [2, 640, 64, 64]> reshape_229_cast = reshape(shape = reshape_229_shape_0, x = real_div_57_cast);
            tensor<fp16, [640]> add_115_mean_0_to_fp16 = const()[name = tensor<string, []>("add_115_mean_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1719905600)))];
            tensor<fp16, [640]> add_115_variance_0_to_fp16 = const()[name = tensor<string, []>("add_115_variance_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1719906944)))];
            tensor<fp16, [640]> add_115_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_115_gamma_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1719908288)))];
            tensor<fp16, [640]> add_115_beta_0_to_fp16 = const()[name = tensor<string, []>("add_115_beta_0_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1719909632)))];
            tensor<fp16, []> add_115_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_115_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 640, 64, 64]> add_115_cast = batch_norm(beta = add_115_beta_0_to_fp16, epsilon = add_115_epsilon_0_to_fp16, gamma = add_115_gamma_0_to_fp16, mean = add_115_mean_0_to_fp16, variance = add_115_variance_0_to_fp16, x = reshape_229_cast);
            tensor<fp16, [2, 640, 64, 64]> input_505_cast = silu(x = add_115_cast);
            tensor<int32, [2]> var_11382 = const()[name = tensor<string, []>("op_11382"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11384 = const()[name = tensor<string, []>("op_11384"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_319_pad_type_0 = const()[name = tensor<string, []>("hidden_states_319_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_319_pad_0 = const()[name = tensor<string, []>("hidden_states_319_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [320, 640, 3, 3]> up_blocks_3_resnets_2_conv1_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_resnets_2_conv1_weight_to_fp16"), val = tensor<fp16, [320, 640, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1719910976)))];
            tensor<fp16, [320]> up_blocks_3_resnets_2_conv1_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_resnets_2_conv1_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1723597440)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_319_cast = conv(bias = up_blocks_3_resnets_2_conv1_bias_to_fp16, dilations = var_11384, groups = var_10509, pad = hidden_states_319_pad_0, pad_type = hidden_states_319_pad_type_0, strides = var_11382, weight = up_blocks_3_resnets_2_conv1_weight_to_fp16, x = input_505_cast);
            tensor<int32, [2]> var_11390 = const()[name = tensor<string, []>("op_11390"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11392 = const()[name = tensor<string, []>("op_11392"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> temb_pad_type_0 = const()[name = tensor<string, []>("temb_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> temb_pad_0 = const()[name = tensor<string, []>("temb_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1280, 1, 1]> up_blocks_3_resnets_2_time_emb_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_resnets_2_time_emb_proj_weight_to_fp16"), val = tensor<fp16, [320, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1723598144)))];
            tensor<fp16, [320]> up_blocks_3_resnets_2_time_emb_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_resnets_2_time_emb_proj_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1724417408)))];
            tensor<fp16, [2, 320, 1, 1]> temb_cast = conv(bias = up_blocks_3_resnets_2_time_emb_proj_bias_to_fp16, dilations = var_11392, groups = var_10509, pad = temb_pad_0, pad_type = temb_pad_type_0, strides = var_11390, weight = up_blocks_3_resnets_2_time_emb_proj_weight_to_fp16, x = input_15_cast);
            tensor<fp16, [2, 320, 64, 64]> input_509_cast = add(x = hidden_states_319_cast, y = temb_cast);
            tensor<int32, [5]> reshape_232_shape_0 = const()[name = tensor<string, []>("reshape_232_shape_0"), val = tensor<int32, [5]>([2, 32, 10, 64, 64])];
            tensor<fp16, [2, 32, 10, 64, 64]> reshape_232_cast = reshape(shape = reshape_232_shape_0, x = input_509_cast);
            tensor<int32, [3]> reduce_mean_174_axes_0 = const()[name = tensor<string, []>("reduce_mean_174_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_174_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_174_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_174_cast = reduce_mean(axes = reduce_mean_174_axes_0, keep_dims = reduce_mean_174_keep_dims_0, x = reshape_232_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> sub_116_cast = sub(x = reshape_232_cast, y = reduce_mean_174_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> square_58_cast = square(x = sub_116_cast);
            tensor<int32, [3]> reduce_mean_176_axes_0 = const()[name = tensor<string, []>("reduce_mean_176_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_176_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_176_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_176_cast = reduce_mean(axes = reduce_mean_176_axes_0, keep_dims = reduce_mean_176_keep_dims_0, x = square_58_cast);
            tensor<fp16, []> add_116_y_0_to_fp16 = const()[name = tensor<string, []>("add_116_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_116_cast = add(x = reduce_mean_176_cast, y = add_116_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_58_cast = sqrt(x = add_116_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> real_div_58_cast = real_div(x = sub_116_cast, y = sqrt_58_cast);
            tensor<int32, [4]> reshape_233_shape_0 = const()[name = tensor<string, []>("reshape_233_shape_0"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> reshape_233_cast = reshape(shape = reshape_233_shape_0, x = real_div_58_cast);
            tensor<fp16, [320]> add_117_mean_0_to_fp16 = const()[name = tensor<string, []>("add_117_mean_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1724418112)))];
            tensor<fp16, [320]> add_117_variance_0_to_fp16 = const()[name = tensor<string, []>("add_117_variance_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1724418816)))];
            tensor<fp16, [320]> add_117_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_117_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1724419520)))];
            tensor<fp16, [320]> add_117_beta_0_to_fp16 = const()[name = tensor<string, []>("add_117_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1724420224)))];
            tensor<fp16, []> add_117_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_117_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 320, 64, 64]> add_117_cast = batch_norm(beta = add_117_beta_0_to_fp16, epsilon = add_117_epsilon_0_to_fp16, gamma = add_117_gamma_0_to_fp16, mean = add_117_mean_0_to_fp16, variance = add_117_variance_0_to_fp16, x = reshape_233_cast);
            tensor<fp16, [2, 320, 64, 64]> input_513_cast = silu(x = add_117_cast);
            tensor<int32, [2]> var_11402 = const()[name = tensor<string, []>("op_11402"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11404 = const()[name = tensor<string, []>("op_11404"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_321_pad_type_0 = const()[name = tensor<string, []>("hidden_states_321_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_321_pad_0 = const()[name = tensor<string, []>("hidden_states_321_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [320, 320, 3, 3]> up_blocks_3_resnets_2_conv2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_resnets_2_conv2_weight_to_fp16"), val = tensor<fp16, [320, 320, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1724420928)))];
            tensor<fp16, [320]> up_blocks_3_resnets_2_conv2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_resnets_2_conv2_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1726264192)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_321_cast = conv(bias = up_blocks_3_resnets_2_conv2_bias_to_fp16, dilations = var_11404, groups = var_10509, pad = hidden_states_321_pad_0, pad_type = hidden_states_321_pad_type_0, strides = var_11402, weight = up_blocks_3_resnets_2_conv2_weight_to_fp16, x = input_513_cast);
            tensor<int32, [2]> var_11409 = const()[name = tensor<string, []>("op_11409"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11411 = const()[name = tensor<string, []>("op_11411"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> x_pad_type_0 = const()[name = tensor<string, []>("x_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> x_pad_0 = const()[name = tensor<string, []>("x_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 640, 1, 1]> up_blocks_3_resnets_2_conv_shortcut_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_resnets_2_conv_shortcut_weight_to_fp16"), val = tensor<fp16, [320, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1726264896)))];
            tensor<fp16, [320]> up_blocks_3_resnets_2_conv_shortcut_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_resnets_2_conv_shortcut_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1726674560)))];
            tensor<fp16, [2, 320, 64, 64]> x_cast = conv(bias = up_blocks_3_resnets_2_conv_shortcut_bias_to_fp16, dilations = var_11411, groups = var_10509, pad = x_pad_0, pad_type = x_pad_type_0, strides = var_11409, weight = up_blocks_3_resnets_2_conv_shortcut_weight_to_fp16, x = input_501_cast);
            tensor<fp16, [2, 320, 64, 64]> hidden_states_323_cast = add(x = x_cast, y = hidden_states_321_cast);
            tensor<int32, [5]> reshape_236_shape_0 = const()[name = tensor<string, []>("reshape_236_shape_0"), val = tensor<int32, [5]>([2, 32, 10, 64, 64])];
            tensor<fp16, [2, 32, 10, 64, 64]> reshape_236_cast = reshape(shape = reshape_236_shape_0, x = hidden_states_323_cast);
            tensor<int32, [3]> reduce_mean_177_axes_0 = const()[name = tensor<string, []>("reduce_mean_177_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_177_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_177_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_177_cast = reduce_mean(axes = reduce_mean_177_axes_0, keep_dims = reduce_mean_177_keep_dims_0, x = reshape_236_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> sub_118_cast = sub(x = reshape_236_cast, y = reduce_mean_177_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> square_59_cast = square(x = sub_118_cast);
            tensor<int32, [3]> reduce_mean_179_axes_0 = const()[name = tensor<string, []>("reduce_mean_179_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_179_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_179_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_179_cast = reduce_mean(axes = reduce_mean_179_axes_0, keep_dims = reduce_mean_179_keep_dims_0, x = square_59_cast);
            tensor<fp16, []> add_118_y_0_to_fp16 = const()[name = tensor<string, []>("add_118_y_0_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_118_cast = add(x = reduce_mean_179_cast, y = add_118_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_59_cast = sqrt(x = add_118_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> real_div_59_cast = real_div(x = sub_118_cast, y = sqrt_59_cast);
            tensor<int32, [4]> reshape_237_shape_0 = const()[name = tensor<string, []>("reshape_237_shape_0"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> reshape_237_cast = reshape(shape = reshape_237_shape_0, x = real_div_59_cast);
            tensor<fp16, [320]> add_119_mean_0_to_fp16 = const()[name = tensor<string, []>("add_119_mean_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1726675264)))];
            tensor<fp16, [320]> add_119_variance_0_to_fp16 = const()[name = tensor<string, []>("add_119_variance_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1726675968)))];
            tensor<fp16, [320]> add_119_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_119_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1726676672)))];
            tensor<fp16, [320]> add_119_beta_0_to_fp16 = const()[name = tensor<string, []>("add_119_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1726677376)))];
            tensor<fp16, []> add_119_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_119_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 320, 64, 64]> add_119_cast = batch_norm(beta = add_119_beta_0_to_fp16, epsilon = add_119_epsilon_0_to_fp16, gamma = add_119_gamma_0_to_fp16, mean = add_119_mean_0_to_fp16, variance = add_119_variance_0_to_fp16, x = reshape_237_cast);
            tensor<int32, [2]> var_11431 = const()[name = tensor<string, []>("op_11431"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11433 = const()[name = tensor<string, []>("op_11433"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_325_pad_type_0 = const()[name = tensor<string, []>("hidden_states_325_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_325_pad_0 = const()[name = tensor<string, []>("hidden_states_325_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> up_blocks_3_attentions_2_proj_in_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_2_proj_in_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1726678080)))];
            tensor<fp16, [320]> up_blocks_3_attentions_2_proj_in_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_2_proj_in_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1726882944)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_325_cast = conv(bias = up_blocks_3_attentions_2_proj_in_bias_to_fp16, dilations = var_11433, groups = var_10509, pad = hidden_states_325_pad_0, pad_type = hidden_states_325_pad_type_0, strides = var_11431, weight = up_blocks_3_attentions_2_proj_in_weight_to_fp16, x = add_119_cast);
            tensor<int32, [4]> var_11438 = const()[name = tensor<string, []>("op_11438"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<fp16, [2, 320, 1, 4096]> inputs_91_cast = reshape(shape = var_11438, x = hidden_states_325_cast);
            tensor<int32, [1]> var_11448 = const()[name = tensor<string, []>("op_11448"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_91_cast = reduce_mean(axes = var_11448, keep_dims = var_10504, x = inputs_91_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_91_cast = sub(x = inputs_91_cast, y = channels_mean_91_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_sq_91_cast = mul(x = zero_mean_91_cast, y = zero_mean_91_cast);
            tensor<int32, [1]> var_11452 = const()[name = tensor<string, []>("op_11452"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_11453_cast = reduce_mean(axes = var_11452, keep_dims = var_10504, x = zero_mean_sq_91_cast);
            tensor<fp16, []> var_11454_to_fp16 = const()[name = tensor<string, []>("op_11454_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_11455_cast = add(x = var_11453_cast, y = var_11454_to_fp16);
            tensor<fp16, []> denom_91_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_91_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_91_cast = rsqrt(epsilon = denom_91_epsilon_0_to_fp16, x = var_11455_cast);
            tensor<fp16, [2, 320, 1, 4096]> out_91_cast = mul(x = zero_mean_91_cast, y = denom_91_cast);
            tensor<fp16, [1, 320, 1, 1]> var_11459_to_fp16 = const()[name = tensor<string, []>("op_11459_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1726883648)))];
            tensor<fp16, [2, 320, 1, 4096]> var_11460_cast = add(x = out_91_cast, y = var_11459_to_fp16);
            tensor<fp16, [1, 320, 1, 1]> var_11462_to_fp16 = const()[name = tensor<string, []>("op_11462_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1726884352)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_327_cast = mul(x = var_11460_cast, y = var_11462_to_fp16);
            tensor<int32, [2]> var_11469 = const()[name = tensor<string, []>("op_11469"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11471 = const()[name = tensor<string, []>("op_11471"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_61_pad_type_0 = const()[name = tensor<string, []>("q_61_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_61_pad_0 = const()[name = tensor<string, []>("q_61_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1726885056)))];
            tensor<fp16, [2, 320, 1, 4096]> q_61_cast = conv(dilations = var_11471, groups = var_10509, pad = q_61_pad_0, pad_type = q_61_pad_type_0, strides = var_11469, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_327_cast);
            tensor<int32, [2]> var_11475 = const()[name = tensor<string, []>("op_11475"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11477 = const()[name = tensor<string, []>("op_11477"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_121_pad_type_0 = const()[name = tensor<string, []>("k_121_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_121_pad_0 = const()[name = tensor<string, []>("k_121_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1727089920)))];
            tensor<fp16, [2, 320, 1, 4096]> k_121_cast = conv(dilations = var_11477, groups = var_10509, pad = k_121_pad_0, pad_type = k_121_pad_type_0, strides = var_11475, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_327_cast);
            tensor<int32, [2]> var_11481 = const()[name = tensor<string, []>("op_11481"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11483 = const()[name = tensor<string, []>("op_11483"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_61_pad_type_0 = const()[name = tensor<string, []>("v_61_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_61_pad_0 = const()[name = tensor<string, []>("v_61_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1727294784)))];
            tensor<fp16, [2, 320, 1, 4096]> v_61_cast = conv(dilations = var_11483, groups = var_10509, pad = v_61_pad_0, pad_type = v_61_pad_type_0, strides = var_11481, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_327_cast);
            tensor<int32, [4]> var_11487_begin_0 = const()[name = tensor<string, []>("op_11487_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11487_end_0 = const()[name = tensor<string, []>("op_11487_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_11487_end_mask_0 = const()[name = tensor<string, []>("op_11487_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11487_cast = slice_by_index(begin = var_11487_begin_0, end = var_11487_end_0, end_mask = var_11487_end_mask_0, x = q_61_cast);
            tensor<int32, [4]> var_11491_begin_0 = const()[name = tensor<string, []>("op_11491_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_11491_end_0 = const()[name = tensor<string, []>("op_11491_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_11491_end_mask_0 = const()[name = tensor<string, []>("op_11491_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11491_cast = slice_by_index(begin = var_11491_begin_0, end = var_11491_end_0, end_mask = var_11491_end_mask_0, x = q_61_cast);
            tensor<int32, [4]> var_11495_begin_0 = const()[name = tensor<string, []>("op_11495_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_11495_end_0 = const()[name = tensor<string, []>("op_11495_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_11495_end_mask_0 = const()[name = tensor<string, []>("op_11495_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11495_cast = slice_by_index(begin = var_11495_begin_0, end = var_11495_end_0, end_mask = var_11495_end_mask_0, x = q_61_cast);
            tensor<int32, [4]> var_11499_begin_0 = const()[name = tensor<string, []>("op_11499_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_11499_end_0 = const()[name = tensor<string, []>("op_11499_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_11499_end_mask_0 = const()[name = tensor<string, []>("op_11499_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11499_cast = slice_by_index(begin = var_11499_begin_0, end = var_11499_end_0, end_mask = var_11499_end_mask_0, x = q_61_cast);
            tensor<int32, [4]> var_11503_begin_0 = const()[name = tensor<string, []>("op_11503_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_11503_end_0 = const()[name = tensor<string, []>("op_11503_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_11503_end_mask_0 = const()[name = tensor<string, []>("op_11503_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11503_cast = slice_by_index(begin = var_11503_begin_0, end = var_11503_end_0, end_mask = var_11503_end_mask_0, x = q_61_cast);
            tensor<int32, [4]> k_123_perm_0 = const()[name = tensor<string, []>("k_123_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_11510_begin_0 = const()[name = tensor<string, []>("op_11510_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11510_end_0 = const()[name = tensor<string, []>("op_11510_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 64])];
            tensor<bool, [4]> var_11510_end_mask_0 = const()[name = tensor<string, []>("op_11510_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 320]> transpose_1 = transpose(perm = k_123_perm_0, x = k_121_cast);
            tensor<fp16, [2, 4096, 1, 64]> var_11510_cast = slice_by_index(begin = var_11510_begin_0, end = var_11510_end_0, end_mask = var_11510_end_mask_0, x = transpose_1);
            tensor<int32, [4]> var_11514_begin_0 = const()[name = tensor<string, []>("op_11514_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_11514_end_0 = const()[name = tensor<string, []>("op_11514_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 128])];
            tensor<bool, [4]> var_11514_end_mask_0 = const()[name = tensor<string, []>("op_11514_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_11514_cast = slice_by_index(begin = var_11514_begin_0, end = var_11514_end_0, end_mask = var_11514_end_mask_0, x = transpose_1);
            tensor<int32, [4]> var_11518_begin_0 = const()[name = tensor<string, []>("op_11518_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_11518_end_0 = const()[name = tensor<string, []>("op_11518_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 192])];
            tensor<bool, [4]> var_11518_end_mask_0 = const()[name = tensor<string, []>("op_11518_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_11518_cast = slice_by_index(begin = var_11518_begin_0, end = var_11518_end_0, end_mask = var_11518_end_mask_0, x = transpose_1);
            tensor<int32, [4]> var_11522_begin_0 = const()[name = tensor<string, []>("op_11522_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_11522_end_0 = const()[name = tensor<string, []>("op_11522_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 256])];
            tensor<bool, [4]> var_11522_end_mask_0 = const()[name = tensor<string, []>("op_11522_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_11522_cast = slice_by_index(begin = var_11522_begin_0, end = var_11522_end_0, end_mask = var_11522_end_mask_0, x = transpose_1);
            tensor<int32, [4]> var_11526_begin_0 = const()[name = tensor<string, []>("op_11526_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_11526_end_0 = const()[name = tensor<string, []>("op_11526_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_11526_end_mask_0 = const()[name = tensor<string, []>("op_11526_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_11526_cast = slice_by_index(begin = var_11526_begin_0, end = var_11526_end_0, end_mask = var_11526_end_mask_0, x = transpose_1);
            tensor<int32, [4]> var_11528_begin_0 = const()[name = tensor<string, []>("op_11528_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11528_end_0 = const()[name = tensor<string, []>("op_11528_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_11528_end_mask_0 = const()[name = tensor<string, []>("op_11528_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11528_cast = slice_by_index(begin = var_11528_begin_0, end = var_11528_end_0, end_mask = var_11528_end_mask_0, x = v_61_cast);
            tensor<int32, [4]> var_11532_begin_0 = const()[name = tensor<string, []>("op_11532_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_11532_end_0 = const()[name = tensor<string, []>("op_11532_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_11532_end_mask_0 = const()[name = tensor<string, []>("op_11532_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11532_cast = slice_by_index(begin = var_11532_begin_0, end = var_11532_end_0, end_mask = var_11532_end_mask_0, x = v_61_cast);
            tensor<int32, [4]> var_11536_begin_0 = const()[name = tensor<string, []>("op_11536_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_11536_end_0 = const()[name = tensor<string, []>("op_11536_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_11536_end_mask_0 = const()[name = tensor<string, []>("op_11536_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11536_cast = slice_by_index(begin = var_11536_begin_0, end = var_11536_end_0, end_mask = var_11536_end_mask_0, x = v_61_cast);
            tensor<int32, [4]> var_11540_begin_0 = const()[name = tensor<string, []>("op_11540_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_11540_end_0 = const()[name = tensor<string, []>("op_11540_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_11540_end_mask_0 = const()[name = tensor<string, []>("op_11540_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11540_cast = slice_by_index(begin = var_11540_begin_0, end = var_11540_end_0, end_mask = var_11540_end_mask_0, x = v_61_cast);
            tensor<int32, [4]> var_11544_begin_0 = const()[name = tensor<string, []>("op_11544_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_11544_end_0 = const()[name = tensor<string, []>("op_11544_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_11544_end_mask_0 = const()[name = tensor<string, []>("op_11544_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11544_cast = slice_by_index(begin = var_11544_begin_0, end = var_11544_end_0, end_mask = var_11544_end_mask_0, x = v_61_cast);
            tensor<string, []> var_11548_equation_0 = const()[name = tensor<string, []>("op_11548_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_11548_cast = einsum(equation = var_11548_equation_0, values = (var_11510_cast, var_11487_cast));
            tensor<fp16, []> var_11549_to_fp16 = const()[name = tensor<string, []>("op_11549_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_761_cast = mul(x = var_11548_cast, y = var_11549_to_fp16);
            tensor<string, []> var_11552_equation_0 = const()[name = tensor<string, []>("op_11552_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_11552_cast = einsum(equation = var_11552_equation_0, values = (var_11514_cast, var_11491_cast));
            tensor<fp16, []> var_11553_to_fp16 = const()[name = tensor<string, []>("op_11553_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_763_cast = mul(x = var_11552_cast, y = var_11553_to_fp16);
            tensor<string, []> var_11556_equation_0 = const()[name = tensor<string, []>("op_11556_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_11556_cast = einsum(equation = var_11556_equation_0, values = (var_11518_cast, var_11495_cast));
            tensor<fp16, []> var_11557_to_fp16 = const()[name = tensor<string, []>("op_11557_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_765_cast = mul(x = var_11556_cast, y = var_11557_to_fp16);
            tensor<string, []> var_11560_equation_0 = const()[name = tensor<string, []>("op_11560_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_11560_cast = einsum(equation = var_11560_equation_0, values = (var_11522_cast, var_11499_cast));
            tensor<fp16, []> var_11561_to_fp16 = const()[name = tensor<string, []>("op_11561_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_767_cast = mul(x = var_11560_cast, y = var_11561_to_fp16);
            tensor<string, []> var_11564_equation_0 = const()[name = tensor<string, []>("op_11564_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 4096]> var_11564_cast = einsum(equation = var_11564_equation_0, values = (var_11526_cast, var_11503_cast));
            tensor<fp16, []> var_11565_to_fp16 = const()[name = tensor<string, []>("op_11565_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 4096]> aw_769_cast = mul(x = var_11564_cast, y = var_11565_to_fp16);
            tensor<fp16, [2, 4096, 1, 4096]> var_11567_cast = softmax(axis = var_10509, x = aw_761_cast);
            tensor<fp16, [2, 4096, 1, 4096]> var_11568_cast = softmax(axis = var_10509, x = aw_763_cast);
            tensor<fp16, [2, 4096, 1, 4096]> var_11569_cast = softmax(axis = var_10509, x = aw_765_cast);
            tensor<fp16, [2, 4096, 1, 4096]> var_11570_cast = softmax(axis = var_10509, x = aw_767_cast);
            tensor<fp16, [2, 4096, 1, 4096]> var_11571_cast = softmax(axis = var_10509, x = aw_769_cast);
            tensor<string, []> var_11573_equation_0 = const()[name = tensor<string, []>("op_11573_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_11573_cast = einsum(equation = var_11573_equation_0, values = (var_11528_cast, var_11567_cast));
            tensor<string, []> var_11575_equation_0 = const()[name = tensor<string, []>("op_11575_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_11575_cast = einsum(equation = var_11575_equation_0, values = (var_11532_cast, var_11568_cast));
            tensor<string, []> var_11577_equation_0 = const()[name = tensor<string, []>("op_11577_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_11577_cast = einsum(equation = var_11577_equation_0, values = (var_11536_cast, var_11569_cast));
            tensor<string, []> var_11579_equation_0 = const()[name = tensor<string, []>("op_11579_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_11579_cast = einsum(equation = var_11579_equation_0, values = (var_11540_cast, var_11570_cast));
            tensor<string, []> var_11581_equation_0 = const()[name = tensor<string, []>("op_11581_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_11581_cast = einsum(equation = var_11581_equation_0, values = (var_11544_cast, var_11571_cast));
            tensor<bool, []> input_517_interleave_0 = const()[name = tensor<string, []>("input_517_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 320, 1, 4096]> input_517_cast = concat(axis = var_10509, interleave = input_517_interleave_0, values = (var_11573_cast, var_11575_cast, var_11577_cast, var_11579_cast, var_11581_cast));
            tensor<int32, [2]> var_11587 = const()[name = tensor<string, []>("op_11587"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11589 = const()[name = tensor<string, []>("op_11589"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_11591_pad_type_0 = const()[name = tensor<string, []>("op_11591_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_11591_pad_0 = const()[name = tensor<string, []>("op_11591_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1727499648)))];
            tensor<fp16, [320]> up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1727704512)))];
            tensor<fp16, [2, 320, 1, 4096]> var_11591_cast = conv(bias = up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_11589, groups = var_10509, pad = var_11591_pad_0, pad_type = var_11591_pad_type_0, strides = var_11587, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_517_cast);
            tensor<fp16, [2, 320, 1, 4096]> inputs_93_cast = add(x = var_11591_cast, y = inputs_91_cast);
            tensor<int32, [1]> var_11595 = const()[name = tensor<string, []>("op_11595"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_93_cast = reduce_mean(axes = var_11595, keep_dims = var_10504, x = inputs_93_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_93_cast = sub(x = inputs_93_cast, y = channels_mean_93_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_sq_93_cast = mul(x = zero_mean_93_cast, y = zero_mean_93_cast);
            tensor<int32, [1]> var_11599 = const()[name = tensor<string, []>("op_11599"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_11600_cast = reduce_mean(axes = var_11599, keep_dims = var_10504, x = zero_mean_sq_93_cast);
            tensor<fp16, []> var_11601_to_fp16 = const()[name = tensor<string, []>("op_11601_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_11602_cast = add(x = var_11600_cast, y = var_11601_to_fp16);
            tensor<fp16, []> denom_93_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_93_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_93_cast = rsqrt(epsilon = denom_93_epsilon_0_to_fp16, x = var_11602_cast);
            tensor<fp16, [2, 320, 1, 4096]> out_93_cast = mul(x = zero_mean_93_cast, y = denom_93_cast);
            tensor<fp16, [1, 320, 1, 1]> var_11606_to_fp16 = const()[name = tensor<string, []>("op_11606_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1727705216)))];
            tensor<fp16, [2, 320, 1, 4096]> var_11607_cast = add(x = out_93_cast, y = var_11606_to_fp16);
            tensor<fp16, [1, 320, 1, 1]> var_11609_to_fp16 = const()[name = tensor<string, []>("op_11609_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1727705920)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_329_cast = mul(x = var_11607_cast, y = var_11609_to_fp16);
            tensor<int32, [2]> var_11616 = const()[name = tensor<string, []>("op_11616"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11618 = const()[name = tensor<string, []>("op_11618"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> q_pad_type_0 = const()[name = tensor<string, []>("q_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> q_pad_0 = const()[name = tensor<string, []>("q_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1727706624)))];
            tensor<fp16, [2, 320, 1, 4096]> q_cast = conv(dilations = var_11618, groups = var_10509, pad = q_pad_0, pad_type = q_pad_type_0, strides = var_11616, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_329_cast);
            tensor<int32, [2]> var_11622 = const()[name = tensor<string, []>("op_11622"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11624 = const()[name = tensor<string, []>("op_11624"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_125_pad_type_0 = const()[name = tensor<string, []>("k_125_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_125_pad_0 = const()[name = tensor<string, []>("k_125_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1024, 1, 1]> up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor<fp16, [320, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1727911488)))];
            tensor<fp16, [2, 320, 1, 77]> k_125_cast = conv(dilations = var_11624, groups = var_10509, pad = k_125_pad_0, pad_type = k_125_pad_type_0, strides = var_11622, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [2]> var_11628 = const()[name = tensor<string, []>("op_11628"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11630 = const()[name = tensor<string, []>("op_11630"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_pad_type_0 = const()[name = tensor<string, []>("v_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_pad_0 = const()[name = tensor<string, []>("v_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1024, 1, 1]> up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor<fp16, [320, 1024, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1728566912)))];
            tensor<fp16, [2, 320, 1, 77]> v_cast = conv(dilations = var_11630, groups = var_10509, pad = v_pad_0, pad_type = v_pad_type_0, strides = var_11628, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states);
            tensor<int32, [4]> var_11634_begin_0 = const()[name = tensor<string, []>("op_11634_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11634_end_0 = const()[name = tensor<string, []>("op_11634_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_11634_end_mask_0 = const()[name = tensor<string, []>("op_11634_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11634_cast = slice_by_index(begin = var_11634_begin_0, end = var_11634_end_0, end_mask = var_11634_end_mask_0, x = q_cast);
            tensor<int32, [4]> var_11638_begin_0 = const()[name = tensor<string, []>("op_11638_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_11638_end_0 = const()[name = tensor<string, []>("op_11638_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_11638_end_mask_0 = const()[name = tensor<string, []>("op_11638_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11638_cast = slice_by_index(begin = var_11638_begin_0, end = var_11638_end_0, end_mask = var_11638_end_mask_0, x = q_cast);
            tensor<int32, [4]> var_11642_begin_0 = const()[name = tensor<string, []>("op_11642_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_11642_end_0 = const()[name = tensor<string, []>("op_11642_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_11642_end_mask_0 = const()[name = tensor<string, []>("op_11642_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11642_cast = slice_by_index(begin = var_11642_begin_0, end = var_11642_end_0, end_mask = var_11642_end_mask_0, x = q_cast);
            tensor<int32, [4]> var_11646_begin_0 = const()[name = tensor<string, []>("op_11646_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_11646_end_0 = const()[name = tensor<string, []>("op_11646_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_11646_end_mask_0 = const()[name = tensor<string, []>("op_11646_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11646_cast = slice_by_index(begin = var_11646_begin_0, end = var_11646_end_0, end_mask = var_11646_end_mask_0, x = q_cast);
            tensor<int32, [4]> var_11650_begin_0 = const()[name = tensor<string, []>("op_11650_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_11650_end_0 = const()[name = tensor<string, []>("op_11650_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_11650_end_mask_0 = const()[name = tensor<string, []>("op_11650_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_11650_cast = slice_by_index(begin = var_11650_begin_0, end = var_11650_end_0, end_mask = var_11650_end_mask_0, x = q_cast);
            tensor<int32, [4]> k_perm_0 = const()[name = tensor<string, []>("k_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_11657_begin_0 = const()[name = tensor<string, []>("op_11657_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11657_end_0 = const()[name = tensor<string, []>("op_11657_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_11657_end_mask_0 = const()[name = tensor<string, []>("op_11657_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 320]> transpose_0 = transpose(perm = k_perm_0, x = k_125_cast);
            tensor<fp16, [2, 77, 1, 64]> var_11657_cast = slice_by_index(begin = var_11657_begin_0, end = var_11657_end_0, end_mask = var_11657_end_mask_0, x = transpose_0);
            tensor<int32, [4]> var_11661_begin_0 = const()[name = tensor<string, []>("op_11661_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_11661_end_0 = const()[name = tensor<string, []>("op_11661_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_11661_end_mask_0 = const()[name = tensor<string, []>("op_11661_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11661_cast = slice_by_index(begin = var_11661_begin_0, end = var_11661_end_0, end_mask = var_11661_end_mask_0, x = transpose_0);
            tensor<int32, [4]> var_11665_begin_0 = const()[name = tensor<string, []>("op_11665_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_11665_end_0 = const()[name = tensor<string, []>("op_11665_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_11665_end_mask_0 = const()[name = tensor<string, []>("op_11665_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11665_cast = slice_by_index(begin = var_11665_begin_0, end = var_11665_end_0, end_mask = var_11665_end_mask_0, x = transpose_0);
            tensor<int32, [4]> var_11669_begin_0 = const()[name = tensor<string, []>("op_11669_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_11669_end_0 = const()[name = tensor<string, []>("op_11669_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_11669_end_mask_0 = const()[name = tensor<string, []>("op_11669_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11669_cast = slice_by_index(begin = var_11669_begin_0, end = var_11669_end_0, end_mask = var_11669_end_mask_0, x = transpose_0);
            tensor<int32, [4]> var_11673_begin_0 = const()[name = tensor<string, []>("op_11673_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_11673_end_0 = const()[name = tensor<string, []>("op_11673_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_11673_end_mask_0 = const()[name = tensor<string, []>("op_11673_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11673_cast = slice_by_index(begin = var_11673_begin_0, end = var_11673_end_0, end_mask = var_11673_end_mask_0, x = transpose_0);
            tensor<int32, [4]> var_11675_begin_0 = const()[name = tensor<string, []>("op_11675_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11675_end_0 = const()[name = tensor<string, []>("op_11675_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_11675_end_mask_0 = const()[name = tensor<string, []>("op_11675_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11675_cast = slice_by_index(begin = var_11675_begin_0, end = var_11675_end_0, end_mask = var_11675_end_mask_0, x = v_cast);
            tensor<int32, [4]> var_11679_begin_0 = const()[name = tensor<string, []>("op_11679_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_11679_end_0 = const()[name = tensor<string, []>("op_11679_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_11679_end_mask_0 = const()[name = tensor<string, []>("op_11679_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11679_cast = slice_by_index(begin = var_11679_begin_0, end = var_11679_end_0, end_mask = var_11679_end_mask_0, x = v_cast);
            tensor<int32, [4]> var_11683_begin_0 = const()[name = tensor<string, []>("op_11683_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_11683_end_0 = const()[name = tensor<string, []>("op_11683_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_11683_end_mask_0 = const()[name = tensor<string, []>("op_11683_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11683_cast = slice_by_index(begin = var_11683_begin_0, end = var_11683_end_0, end_mask = var_11683_end_mask_0, x = v_cast);
            tensor<int32, [4]> var_11687_begin_0 = const()[name = tensor<string, []>("op_11687_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_11687_end_0 = const()[name = tensor<string, []>("op_11687_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_11687_end_mask_0 = const()[name = tensor<string, []>("op_11687_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11687_cast = slice_by_index(begin = var_11687_begin_0, end = var_11687_end_0, end_mask = var_11687_end_mask_0, x = v_cast);
            tensor<int32, [4]> var_11691_begin_0 = const()[name = tensor<string, []>("op_11691_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_11691_end_0 = const()[name = tensor<string, []>("op_11691_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_11691_end_mask_0 = const()[name = tensor<string, []>("op_11691_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11691_cast = slice_by_index(begin = var_11691_begin_0, end = var_11691_end_0, end_mask = var_11691_end_mask_0, x = v_cast);
            tensor<string, []> var_11695_equation_0 = const()[name = tensor<string, []>("op_11695_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_11695_cast = einsum(equation = var_11695_equation_0, values = (var_11657_cast, var_11634_cast));
            tensor<fp16, []> var_11696_to_fp16 = const()[name = tensor<string, []>("op_11696_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_771_cast = mul(x = var_11695_cast, y = var_11696_to_fp16);
            tensor<string, []> var_11699_equation_0 = const()[name = tensor<string, []>("op_11699_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_11699_cast = einsum(equation = var_11699_equation_0, values = (var_11661_cast, var_11638_cast));
            tensor<fp16, []> var_11700_to_fp16 = const()[name = tensor<string, []>("op_11700_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_773_cast = mul(x = var_11699_cast, y = var_11700_to_fp16);
            tensor<string, []> var_11703_equation_0 = const()[name = tensor<string, []>("op_11703_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_11703_cast = einsum(equation = var_11703_equation_0, values = (var_11665_cast, var_11642_cast));
            tensor<fp16, []> var_11704_to_fp16 = const()[name = tensor<string, []>("op_11704_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_775_cast = mul(x = var_11703_cast, y = var_11704_to_fp16);
            tensor<string, []> var_11707_equation_0 = const()[name = tensor<string, []>("op_11707_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_11707_cast = einsum(equation = var_11707_equation_0, values = (var_11669_cast, var_11646_cast));
            tensor<fp16, []> var_11708_to_fp16 = const()[name = tensor<string, []>("op_11708_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_777_cast = mul(x = var_11707_cast, y = var_11708_to_fp16);
            tensor<string, []> var_11711_equation_0 = const()[name = tensor<string, []>("op_11711_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 4096]> var_11711_cast = einsum(equation = var_11711_equation_0, values = (var_11673_cast, var_11650_cast));
            tensor<fp16, []> var_11712_to_fp16 = const()[name = tensor<string, []>("op_11712_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 4096]> aw_cast = mul(x = var_11711_cast, y = var_11712_to_fp16);
            tensor<fp16, [2, 77, 1, 4096]> var_11714_cast = softmax(axis = var_10509, x = aw_771_cast);
            tensor<fp16, [2, 77, 1, 4096]> var_11715_cast = softmax(axis = var_10509, x = aw_773_cast);
            tensor<fp16, [2, 77, 1, 4096]> var_11716_cast = softmax(axis = var_10509, x = aw_775_cast);
            tensor<fp16, [2, 77, 1, 4096]> var_11717_cast = softmax(axis = var_10509, x = aw_777_cast);
            tensor<fp16, [2, 77, 1, 4096]> var_11718_cast = softmax(axis = var_10509, x = aw_cast);
            tensor<string, []> var_11720_equation_0 = const()[name = tensor<string, []>("op_11720_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_11720_cast = einsum(equation = var_11720_equation_0, values = (var_11675_cast, var_11714_cast));
            tensor<string, []> var_11722_equation_0 = const()[name = tensor<string, []>("op_11722_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_11722_cast = einsum(equation = var_11722_equation_0, values = (var_11679_cast, var_11715_cast));
            tensor<string, []> var_11724_equation_0 = const()[name = tensor<string, []>("op_11724_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_11724_cast = einsum(equation = var_11724_equation_0, values = (var_11683_cast, var_11716_cast));
            tensor<string, []> var_11726_equation_0 = const()[name = tensor<string, []>("op_11726_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_11726_cast = einsum(equation = var_11726_equation_0, values = (var_11687_cast, var_11717_cast));
            tensor<string, []> var_11728_equation_0 = const()[name = tensor<string, []>("op_11728_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 4096]> var_11728_cast = einsum(equation = var_11728_equation_0, values = (var_11691_cast, var_11718_cast));
            tensor<bool, []> input_519_interleave_0 = const()[name = tensor<string, []>("input_519_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 320, 1, 4096]> input_519_cast = concat(axis = var_10509, interleave = input_519_interleave_0, values = (var_11720_cast, var_11722_cast, var_11724_cast, var_11726_cast, var_11728_cast));
            tensor<int32, [2]> var_11734 = const()[name = tensor<string, []>("op_11734"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11736 = const()[name = tensor<string, []>("op_11736"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_11738_pad_type_0 = const()[name = tensor<string, []>("op_11738_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_11738_pad_0 = const()[name = tensor<string, []>("op_11738_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1729222336)))];
            tensor<fp16, [320]> up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1729427200)))];
            tensor<fp16, [2, 320, 1, 4096]> var_11738_cast = conv(bias = up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_11736, groups = var_10509, pad = var_11738_pad_0, pad_type = var_11738_pad_type_0, strides = var_11734, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_519_cast);
            tensor<fp16, [2, 320, 1, 4096]> inputs_cast = add(x = var_11738_cast, y = inputs_93_cast);
            tensor<int32, [1]> var_11742 = const()[name = tensor<string, []>("op_11742"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_cast = reduce_mean(axes = var_11742, keep_dims = var_10504, x = inputs_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_cast = sub(x = inputs_cast, y = channels_mean_cast);
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_sq_cast = mul(x = zero_mean_cast, y = zero_mean_cast);
            tensor<int32, [1]> var_11746 = const()[name = tensor<string, []>("op_11746"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_11747_cast = reduce_mean(axes = var_11746, keep_dims = var_10504, x = zero_mean_sq_cast);
            tensor<fp16, []> var_11748_to_fp16 = const()[name = tensor<string, []>("op_11748_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_11749_cast = add(x = var_11747_cast, y = var_11748_to_fp16);
            tensor<fp16, []> denom_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [2, 1, 1, 4096]> denom_cast = rsqrt(epsilon = denom_epsilon_0_to_fp16, x = var_11749_cast);
            tensor<fp16, [2, 320, 1, 4096]> out_cast = mul(x = zero_mean_cast, y = denom_cast);
            tensor<fp16, [1, 320, 1, 1]> var_11753_to_fp16 = const()[name = tensor<string, []>("op_11753_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1729427904)))];
            tensor<fp16, [2, 320, 1, 4096]> var_11754_cast = add(x = out_cast, y = var_11753_to_fp16);
            tensor<fp16, [1, 320, 1, 1]> var_11756_to_fp16 = const()[name = tensor<string, []>("op_11756_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1729428608)))];
            tensor<fp16, [2, 320, 1, 4096]> input_521_cast = mul(x = var_11754_cast, y = var_11756_to_fp16);
            tensor<int32, [2]> var_11764 = const()[name = tensor<string, []>("op_11764"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11766 = const()[name = tensor<string, []>("op_11766"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_11768_pad_type_0 = const()[name = tensor<string, []>("op_11768_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_11768_pad_0 = const()[name = tensor<string, []>("op_11768_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [2560, 320, 1, 1]> up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor<fp16, [2560, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1729429312)))];
            tensor<fp16, [2560]> up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1731067776)))];
            tensor<fp16, [2, 2560, 1, 4096]> var_11768_cast = conv(bias = up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_11766, groups = var_10509, pad = var_11768_pad_0, pad_type = var_11768_pad_type_0, strides = var_11764, weight = up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_521_cast);
            tensor<int32, [2]> var_11769_split_sizes_0 = const()[name = tensor<string, []>("op_11769_split_sizes_0"), val = tensor<int32, [2]>([1280, 1280])];
            tensor<int32, []> var_11769_axis_0 = const()[name = tensor<string, []>("op_11769_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 1280, 1, 4096]> var_11769_cast_0, tensor<fp16, [2, 1280, 1, 4096]> var_11769_cast_1 = split(axis = var_11769_axis_0, split_sizes = var_11769_split_sizes_0, x = var_11768_cast);
            tensor<string, []> var_11771_mode_0 = const()[name = tensor<string, []>("op_11771_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 1280, 1, 4096]> var_11771_cast = gelu(mode = var_11771_mode_0, x = var_11769_cast_1);
            tensor<fp16, [2, 1280, 1, 4096]> input_523_cast = mul(x = var_11769_cast_0, y = var_11771_cast);
            tensor<int32, [2]> var_11775 = const()[name = tensor<string, []>("op_11775"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11777 = const()[name = tensor<string, []>("op_11777"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_11779_pad_type_0 = const()[name = tensor<string, []>("op_11779_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_11779_pad_0 = const()[name = tensor<string, []>("op_11779_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1280, 1, 1]> up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor<fp16, [320, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1731072960)))];
            tensor<fp16, [320]> up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1731892224)))];
            tensor<fp16, [2, 320, 1, 4096]> var_11779_cast = conv(bias = up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_11777, groups = var_10509, pad = var_11779_pad_0, pad_type = var_11779_pad_type_0, strides = var_11775, weight = up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_523_cast);
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_333_cast = add(x = var_11779_cast, y = inputs_cast);
            tensor<int32, [4]> var_11781 = const()[name = tensor<string, []>("op_11781"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> input_525_cast = reshape(shape = var_11781, x = hidden_states_333_cast);
            tensor<int32, [2]> var_11785 = const()[name = tensor<string, []>("op_11785"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11787 = const()[name = tensor<string, []>("op_11787"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> hidden_states_pad_type_0 = const()[name = tensor<string, []>("hidden_states_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> hidden_states_pad_0 = const()[name = tensor<string, []>("hidden_states_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 320, 1, 1]> up_blocks_3_attentions_2_proj_out_weight_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_2_proj_out_weight_to_fp16"), val = tensor<fp16, [320, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1731892928)))];
            tensor<fp16, [320]> up_blocks_3_attentions_2_proj_out_bias_to_fp16 = const()[name = tensor<string, []>("up_blocks_3_attentions_2_proj_out_bias_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1732097792)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_cast = conv(bias = up_blocks_3_attentions_2_proj_out_bias_to_fp16, dilations = var_11787, groups = var_10509, pad = hidden_states_pad_0, pad_type = hidden_states_pad_type_0, strides = var_11785, weight = up_blocks_3_attentions_2_proj_out_weight_to_fp16, x = input_525_cast);
            tensor<fp16, [2, 320, 64, 64]> input_527_cast = add(x = hidden_states_cast, y = hidden_states_323_cast);
            tensor<int32, [5]> reshape_240_shape_0 = const()[name = tensor<string, []>("reshape_240_shape_0"), val = tensor<int32, [5]>([2, 32, 10, 64, 64])];
            tensor<fp16, [2, 32, 10, 64, 64]> reshape_240_cast = reshape(shape = reshape_240_shape_0, x = input_527_cast);
            tensor<int32, [3]> reduce_mean_180_axes_0 = const()[name = tensor<string, []>("reduce_mean_180_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_180_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_180_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_180_cast = reduce_mean(axes = reduce_mean_180_axes_0, keep_dims = reduce_mean_180_keep_dims_0, x = reshape_240_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> sub_120_cast = sub(x = reshape_240_cast, y = reduce_mean_180_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> square_60_cast = square(x = sub_120_cast);
            tensor<int32, [3]> reduce_mean_182_axes_0 = const()[name = tensor<string, []>("reduce_mean_182_axes_0"), val = tensor<int32, [3]>([2, 3, 4])];
            tensor<bool, []> reduce_mean_182_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_182_keep_dims_0"), val = tensor<bool, []>(true)];
            tensor<fp16, [2, 32, 1, 1, 1]> reduce_mean_182_cast = reduce_mean(axes = reduce_mean_182_axes_0, keep_dims = reduce_mean_182_keep_dims_0, x = square_60_cast);
            tensor<fp16, []> add_120_y_0_to_fp16 = const()[name = tensor<string, []>("add_120_y_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 32, 1, 1, 1]> add_120_cast = add(x = reduce_mean_182_cast, y = add_120_y_0_to_fp16);
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_60_cast = sqrt(x = add_120_cast);
            tensor<fp16, [2, 32, 10, 64, 64]> real_div_60_cast = real_div(x = sub_120_cast, y = sqrt_60_cast);
            tensor<int32, [4]> reshape_241_shape_0 = const()[name = tensor<string, []>("reshape_241_shape_0"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> reshape_241_cast = reshape(shape = reshape_241_shape_0, x = real_div_60_cast);
            tensor<fp16, [320]> add_121_mean_0_to_fp16 = const()[name = tensor<string, []>("add_121_mean_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1732098496)))];
            tensor<fp16, [320]> add_121_variance_0_to_fp16 = const()[name = tensor<string, []>("add_121_variance_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1732099200)))];
            tensor<fp16, [320]> add_121_gamma_0_to_fp16 = const()[name = tensor<string, []>("add_121_gamma_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1732099904)))];
            tensor<fp16, [320]> add_121_beta_0_to_fp16 = const()[name = tensor<string, []>("add_121_beta_0_to_fp16"), val = tensor<fp16, [320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1732100608)))];
            tensor<fp16, []> add_121_epsilon_0_to_fp16 = const()[name = tensor<string, []>("add_121_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 320, 64, 64]> add_121_cast = batch_norm(beta = add_121_beta_0_to_fp16, epsilon = add_121_epsilon_0_to_fp16, gamma = add_121_gamma_0_to_fp16, mean = add_121_mean_0_to_fp16, variance = add_121_variance_0_to_fp16, x = reshape_241_cast);
            tensor<fp16, [2, 320, 64, 64]> input_cast = silu(x = add_121_cast);
            tensor<int32, []> var_11801 = const()[name = tensor<string, []>("op_11801"), val = tensor<int32, []>(1)];
            tensor<int32, [2]> var_11804 = const()[name = tensor<string, []>("op_11804"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11806 = const()[name = tensor<string, []>("op_11806"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_11808_pad_type_0 = const()[name = tensor<string, []>("op_11808_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_11808_pad_0 = const()[name = tensor<string, []>("op_11808_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [4, 320, 3, 3]> conv_out_weight_to_fp16 = const()[name = tensor<string, []>("conv_out_weight_to_fp16"), val = tensor<fp16, [4, 320, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1732101312)))];
            tensor<fp16, [4]> conv_out_bias_to_fp16 = const()[name = tensor<string, []>("conv_out_bias_to_fp16"), val = tensor<fp16, [4]>([-0x1.498p-9, 0x1.e34p-10, 0x1.694p-12, 0x1.b84p-10])];
            tensor<fp16, [2, 4, 64, 64]> var_11808_cast = conv(bias = conv_out_bias_to_fp16, dilations = var_11806, groups = var_11801, pad = var_11808_pad_0, pad_type = var_11808_pad_type_0, strides = var_11804, weight = conv_out_weight_to_fp16, x = input_cast);
            tensor<string, []> var_11808_cast_to_fp32_dtype_0 = const()[name = tensor<string, []>("op_11808_cast_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
            tensor<fp32, [2, 4, 64, 64]> noise_pred = cast(dtype = var_11808_cast_to_fp32_dtype_0, x = var_11808_cast);
        } -> (noise_pred);
}