program(1.0)
[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "5.30.0"}, {"coremlc-version", "1839.0.0"}})]
{
    func main<ios16>(tensor<fp16, [2, 768, 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)[name = tensor<string, []>("op_42_cast")];
            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)[name = tensor<string, []>("emb_3_cast")];
            tensor<fp16, [2, 160]> var_49_cast = sin(x = emb_3_cast)[name = tensor<string, []>("op_49_cast")];
            tensor<fp16, [2, 160]> var_50_cast = cos(x = emb_3_cast)[name = tensor<string, []>("op_50_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))[name = tensor<string, []>("emb_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)[name = tensor<string, []>("op_54_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)[name = tensor<string, []>("op_56_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))[name = tensor<string, []>("sample_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)[name = tensor<string, []>("op_66_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)[name = tensor<string, []>("input_1_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(448))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(307712))), name = tensor<string, []>("time_embedding_linear_1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 320, 1, 1])];
            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, []>(307904)))];
            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_palettized, x = input_1_cast)[name = tensor<string, []>("input_3_cast")];
            tensor<fp16, [2, 1280, 1, 1]> input_5_cast = silu(x = input_3_cast)[name = tensor<string, []>("input_5_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(310528))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1539392))), name = tensor<string, []>("time_embedding_linear_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(1539584)))];
            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_palettized, x = input_5_cast)[name = tensor<string, []>("input_13_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [8640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1542208))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1550912))), name = tensor<string, []>("conv_in_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 4, 3, 3])];
            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, []>(1551104)))];
            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_palettized, x = sample)[name = tensor<string, []>("input_7_cast")];
            tensor<int32, []> var_95 = const()[name = tensor<string, []>("op_95"), val = tensor<int32, []>(3)];
            tensor<bool, []> var_118 = const()[name = tensor<string, []>("op_118"), val = tensor<bool, []>(true)];
            tensor<int32, []> var_123 = const()[name = tensor<string, []>("op_123"), 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)[name = tensor<string, []>("reshape_0_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)[name = tensor<string, []>("reduce_mean_0_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> sub_0_cast = sub(x = reshape_0_cast, y = reduce_mean_0_cast)[name = tensor<string, []>("sub_0_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> square_0_cast = square(x = sub_0_cast)[name = tensor<string, []>("square_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)[name = tensor<string, []>("reduce_mean_2_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)[name = tensor<string, []>("add_0_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_0_cast = sqrt(x = add_0_cast)[name = tensor<string, []>("sqrt_0_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> real_div_0_cast = real_div(x = sub_0_cast, y = sqrt_0_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_1_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, []>(1551808)))];
            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, []>(1552512)))];
            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, []>(1553216)))];
            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, []>(1553920)))];
            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)[name = tensor<string, []>("add_1_cast")];
            tensor<fp16, [2, 320, 64, 64]> input_11_cast = silu(x = add_1_cast)[name = tensor<string, []>("input_11_cast")];
            tensor<int32, [2]> var_145 = const()[name = tensor<string, []>("op_145"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_147 = const()[name = tensor<string, []>("op_147"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [691200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1554624))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2245888))), name = tensor<string, []>("down_blocks_0_resnets_0_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 3, 3])];
            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, []>(2246080)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_1_cast = conv(bias = down_blocks_0_resnets_0_conv1_bias_to_fp16, dilations = var_147, groups = var_123, pad = hidden_states_1_pad_0, pad_type = hidden_states_1_pad_type_0, strides = var_145, weight = down_blocks_0_resnets_0_conv1_weight_to_fp16_palettized, x = input_11_cast)[name = tensor<string, []>("hidden_states_1_cast")];
            tensor<fp16, [2, 1280, 1, 1]> input_15_cast = silu(x = input_13_cast)[name = tensor<string, []>("input_15_cast")];
            tensor<int32, [2]> var_153 = const()[name = tensor<string, []>("op_153"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_155 = const()[name = tensor<string, []>("op_155"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2246784))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2554048))), name = tensor<string, []>("down_blocks_0_resnets_0_time_emb_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 1280, 1, 1])];
            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, []>(2554240)))];
            tensor<fp16, [2, 320, 1, 1]> temb_1_cast = conv(bias = down_blocks_0_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_155, groups = var_123, pad = temb_1_pad_0, pad_type = temb_1_pad_type_0, strides = var_153, weight = down_blocks_0_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor<string, []>("temb_1_cast")];
            tensor<fp16, [2, 320, 64, 64]> input_17_cast = add(x = hidden_states_1_cast, y = temb_1_cast)[name = tensor<string, []>("input_17_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)[name = tensor<string, []>("reshape_4_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)[name = tensor<string, []>("reduce_mean_3_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> sub_2_cast = sub(x = reshape_4_cast, y = reduce_mean_3_cast)[name = tensor<string, []>("sub_2_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> square_1_cast = square(x = sub_2_cast)[name = tensor<string, []>("square_1_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)[name = tensor<string, []>("reduce_mean_5_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)[name = tensor<string, []>("add_2_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_1_cast = sqrt(x = add_2_cast)[name = tensor<string, []>("sqrt_1_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> real_div_1_cast = real_div(x = sub_2_cast, y = sqrt_1_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_5_cast")];
            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, []>(2554944)))];
            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, []>(2555648)))];
            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_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_5_cast)[name = tensor<string, []>("add_3_cast")];
            tensor<fp16, [2, 320, 64, 64]> input_21_cast = silu(x = add_3_cast)[name = tensor<string, []>("input_21_cast")];
            tensor<int32, [2]> var_165 = const()[name = tensor<string, []>("op_165"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_167 = const()[name = tensor<string, []>("op_167"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [691200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2556352))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3247616))), name = tensor<string, []>("down_blocks_0_resnets_0_conv2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 3, 3])];
            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, []>(3247808)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_3_cast = conv(bias = down_blocks_0_resnets_0_conv2_bias_to_fp16, dilations = var_167, groups = var_123, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = var_165, weight = down_blocks_0_resnets_0_conv2_weight_to_fp16_palettized, x = input_21_cast)[name = tensor<string, []>("hidden_states_3_cast")];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_5_cast = add(x = input_7_cast, y = hidden_states_3_cast)[name = tensor<string, []>("hidden_states_5_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)[name = tensor<string, []>("reshape_8_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)[name = tensor<string, []>("reduce_mean_6_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> sub_4_cast = sub(x = reshape_8_cast, y = reduce_mean_6_cast)[name = tensor<string, []>("sub_4_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> square_2_cast = square(x = sub_4_cast)[name = tensor<string, []>("square_2_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)[name = tensor<string, []>("reduce_mean_8_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)[name = tensor<string, []>("add_4_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_2_cast = sqrt(x = add_4_cast)[name = tensor<string, []>("sqrt_2_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> real_div_2_cast = real_div(x = sub_4_cast, y = sqrt_2_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_9_cast")];
            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, []>(3248512)))];
            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, []>(3249216)))];
            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_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_9_cast)[name = tensor<string, []>("add_5_cast")];
            tensor<int32, [2]> var_187 = const()[name = tensor<string, []>("op_187"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_189 = const()[name = tensor<string, []>("op_189"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3249920))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3326784))), name = tensor<string, []>("down_blocks_0_attentions_0_proj_in_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            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, []>(3326976)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_7_cast = conv(bias = down_blocks_0_attentions_0_proj_in_bias_to_fp16, dilations = var_189, groups = var_123, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = var_187, weight = down_blocks_0_attentions_0_proj_in_weight_to_fp16_palettized, x = add_5_cast)[name = tensor<string, []>("hidden_states_7_cast")];
            tensor<int32, [4]> var_194 = const()[name = tensor<string, []>("op_194"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<fp16, [2, 320, 1, 4096]> inputs_1_cast = reshape(shape = var_194, x = hidden_states_7_cast)[name = tensor<string, []>("inputs_1_cast")];
            tensor<int32, [1]> var_204 = const()[name = tensor<string, []>("op_204"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_1_cast = reduce_mean(axes = var_204, keep_dims = var_118, x = inputs_1_cast)[name = tensor<string, []>("channels_mean_1_cast")];
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_1_cast = sub(x = inputs_1_cast, y = channels_mean_1_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_1_cast")];
            tensor<int32, [1]> var_208 = const()[name = tensor<string, []>("op_208"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_209_cast = reduce_mean(axes = var_208, keep_dims = var_118, x = zero_mean_sq_1_cast)[name = tensor<string, []>("op_209_cast")];
            tensor<fp16, []> var_210_to_fp16 = const()[name = tensor<string, []>("op_210_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_211_cast = add(x = var_209_cast, y = var_210_to_fp16)[name = tensor<string, []>("op_211_cast")];
            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_211_cast)[name = tensor<string, []>("denom_1_cast")];
            tensor<fp16, [2, 320, 1, 4096]> out_1_cast = mul(x = zero_mean_1_cast, y = denom_1_cast)[name = tensor<string, []>("out_1_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_215_to_fp16 = const()[name = tensor<string, []>("op_215_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3327680)))];
            tensor<fp16, [2, 320, 1, 4096]> var_216_cast = add(x = out_1_cast, y = var_215_to_fp16)[name = tensor<string, []>("op_216_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_218_to_fp16 = const()[name = tensor<string, []>("op_218_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3328384)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_9_cast = mul(x = var_216_cast, y = var_218_to_fp16)[name = tensor<string, []>("hidden_states_9_cast")];
            tensor<int32, [2]> var_225 = const()[name = tensor<string, []>("op_225"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_227 = const()[name = tensor<string, []>("op_227"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3329088))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3405952))), name = tensor<string, []>("down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            tensor<fp16, [2, 320, 1, 4096]> q_1_cast = conv(dilations = var_227, groups = var_123, pad = q_1_pad_0, pad_type = q_1_pad_type_0, strides = var_225, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_9_cast)[name = tensor<string, []>("q_1_cast")];
            tensor<int32, [2]> var_231 = const()[name = tensor<string, []>("op_231"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_233 = const()[name = tensor<string, []>("op_233"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3406144))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3483008))), name = tensor<string, []>("down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            tensor<fp16, [2, 320, 1, 4096]> k_1_cast = conv(dilations = var_233, groups = var_123, pad = k_1_pad_0, pad_type = k_1_pad_type_0, strides = var_231, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_9_cast)[name = tensor<string, []>("k_1_cast")];
            tensor<int32, [2]> var_237 = const()[name = tensor<string, []>("op_237"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_239 = const()[name = tensor<string, []>("op_239"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3483200))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3560064))), name = tensor<string, []>("down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            tensor<fp16, [2, 320, 1, 4096]> v_1_cast = conv(dilations = var_239, groups = var_123, pad = v_1_pad_0, pad_type = v_1_pad_type_0, strides = var_237, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_9_cast)[name = tensor<string, []>("v_1_cast")];
            tensor<int32, [4]> var_243_begin_0 = const()[name = tensor<string, []>("op_243_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_243_end_0 = const()[name = tensor<string, []>("op_243_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_243_end_mask_0 = const()[name = tensor<string, []>("op_243_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_243_cast = slice_by_index(begin = var_243_begin_0, end = var_243_end_0, end_mask = var_243_end_mask_0, x = q_1_cast)[name = tensor<string, []>("op_243_cast")];
            tensor<int32, [4]> var_247_begin_0 = const()[name = tensor<string, []>("op_247_begin_0"), val = tensor<int32, [4]>([0, 40, 0, 0])];
            tensor<int32, [4]> var_247_end_0 = const()[name = tensor<string, []>("op_247_end_0"), val = tensor<int32, [4]>([2, 80, 1, 4096])];
            tensor<bool, [4]> var_247_end_mask_0 = const()[name = tensor<string, []>("op_247_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_247_cast = slice_by_index(begin = var_247_begin_0, end = var_247_end_0, end_mask = var_247_end_mask_0, x = q_1_cast)[name = tensor<string, []>("op_247_cast")];
            tensor<int32, [4]> var_251_begin_0 = const()[name = tensor<string, []>("op_251_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_251_end_0 = const()[name = tensor<string, []>("op_251_end_0"), val = tensor<int32, [4]>([2, 120, 1, 4096])];
            tensor<bool, [4]> var_251_end_mask_0 = const()[name = tensor<string, []>("op_251_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_251_cast = slice_by_index(begin = var_251_begin_0, end = var_251_end_0, end_mask = var_251_end_mask_0, x = q_1_cast)[name = tensor<string, []>("op_251_cast")];
            tensor<int32, [4]> var_255_begin_0 = const()[name = tensor<string, []>("op_255_begin_0"), val = tensor<int32, [4]>([0, 120, 0, 0])];
            tensor<int32, [4]> var_255_end_0 = const()[name = tensor<string, []>("op_255_end_0"), val = tensor<int32, [4]>([2, 160, 1, 4096])];
            tensor<bool, [4]> var_255_end_mask_0 = const()[name = tensor<string, []>("op_255_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_255_cast = slice_by_index(begin = var_255_begin_0, end = var_255_end_0, end_mask = var_255_end_mask_0, x = q_1_cast)[name = tensor<string, []>("op_255_cast")];
            tensor<int32, [4]> var_259_begin_0 = const()[name = tensor<string, []>("op_259_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_259_end_0 = const()[name = tensor<string, []>("op_259_end_0"), val = tensor<int32, [4]>([2, 200, 1, 4096])];
            tensor<bool, [4]> var_259_end_mask_0 = const()[name = tensor<string, []>("op_259_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_259_cast = slice_by_index(begin = var_259_begin_0, end = var_259_end_0, end_mask = var_259_end_mask_0, x = q_1_cast)[name = tensor<string, []>("op_259_cast")];
            tensor<int32, [4]> var_263_begin_0 = const()[name = tensor<string, []>("op_263_begin_0"), val = tensor<int32, [4]>([0, 200, 0, 0])];
            tensor<int32, [4]> var_263_end_0 = const()[name = tensor<string, []>("op_263_end_0"), val = tensor<int32, [4]>([2, 240, 1, 4096])];
            tensor<bool, [4]> var_263_end_mask_0 = const()[name = tensor<string, []>("op_263_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_263_cast = slice_by_index(begin = var_263_begin_0, end = var_263_end_0, end_mask = var_263_end_mask_0, x = q_1_cast)[name = tensor<string, []>("op_263_cast")];
            tensor<int32, [4]> var_267_begin_0 = const()[name = tensor<string, []>("op_267_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_267_end_0 = const()[name = tensor<string, []>("op_267_end_0"), val = tensor<int32, [4]>([2, 280, 1, 4096])];
            tensor<bool, [4]> var_267_end_mask_0 = const()[name = tensor<string, []>("op_267_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_267_cast = slice_by_index(begin = var_267_begin_0, end = var_267_end_0, end_mask = var_267_end_mask_0, x = q_1_cast)[name = tensor<string, []>("op_267_cast")];
            tensor<int32, [4]> var_271_begin_0 = const()[name = tensor<string, []>("op_271_begin_0"), val = tensor<int32, [4]>([0, 280, 0, 0])];
            tensor<int32, [4]> var_271_end_0 = const()[name = tensor<string, []>("op_271_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_271_end_mask_0 = const()[name = tensor<string, []>("op_271_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_271_cast = slice_by_index(begin = var_271_begin_0, end = var_271_end_0, end_mask = var_271_end_mask_0, x = q_1_cast)[name = tensor<string, []>("op_271_cast")];
            tensor<int32, [4]> var_274_begin_0 = const()[name = tensor<string, []>("op_274_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_274_end_0 = const()[name = tensor<string, []>("op_274_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_274_end_mask_0 = const()[name = tensor<string, []>("op_274_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_274_cast = slice_by_index(begin = var_274_begin_0, end = var_274_end_0, end_mask = var_274_end_mask_0, x = var_243_cast)[name = tensor<string, []>("op_274_cast")];
            tensor<int32, [4]> var_275_begin_0 = const()[name = tensor<string, []>("op_275_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_275_end_0 = const()[name = tensor<string, []>("op_275_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_275_end_mask_0 = const()[name = tensor<string, []>("op_275_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_275_cast = slice_by_index(begin = var_275_begin_0, end = var_275_end_0, end_mask = var_275_end_mask_0, x = var_243_cast)[name = tensor<string, []>("op_275_cast")];
            tensor<int32, [4]> var_276_begin_0 = const()[name = tensor<string, []>("op_276_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_276_end_0 = const()[name = tensor<string, []>("op_276_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_276_end_mask_0 = const()[name = tensor<string, []>("op_276_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_276_cast = slice_by_index(begin = var_276_begin_0, end = var_276_end_0, end_mask = var_276_end_mask_0, x = var_243_cast)[name = tensor<string, []>("op_276_cast")];
            tensor<int32, [4]> var_277_begin_0 = const()[name = tensor<string, []>("op_277_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_277_end_0 = const()[name = tensor<string, []>("op_277_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_277_end_mask_0 = const()[name = tensor<string, []>("op_277_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_277_cast = slice_by_index(begin = var_277_begin_0, end = var_277_end_0, end_mask = var_277_end_mask_0, x = var_243_cast)[name = tensor<string, []>("op_277_cast")];
            tensor<int32, [4]> var_278_begin_0 = const()[name = tensor<string, []>("op_278_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_278_end_0 = const()[name = tensor<string, []>("op_278_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_278_end_mask_0 = const()[name = tensor<string, []>("op_278_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_278_cast = slice_by_index(begin = var_278_begin_0, end = var_278_end_0, end_mask = var_278_end_mask_0, x = var_243_cast)[name = tensor<string, []>("op_278_cast")];
            tensor<int32, [4]> var_279_begin_0 = const()[name = tensor<string, []>("op_279_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_279_end_0 = const()[name = tensor<string, []>("op_279_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_279_end_mask_0 = const()[name = tensor<string, []>("op_279_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_279_cast = slice_by_index(begin = var_279_begin_0, end = var_279_end_0, end_mask = var_279_end_mask_0, x = var_243_cast)[name = tensor<string, []>("op_279_cast")];
            tensor<int32, [4]> var_280_begin_0 = const()[name = tensor<string, []>("op_280_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_280_end_0 = const()[name = tensor<string, []>("op_280_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_280_end_mask_0 = const()[name = tensor<string, []>("op_280_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_280_cast = slice_by_index(begin = var_280_begin_0, end = var_280_end_0, end_mask = var_280_end_mask_0, x = var_243_cast)[name = tensor<string, []>("op_280_cast")];
            tensor<int32, [4]> var_281_begin_0 = const()[name = tensor<string, []>("op_281_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_281_end_0 = const()[name = tensor<string, []>("op_281_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_281_end_mask_0 = const()[name = tensor<string, []>("op_281_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_281_cast = slice_by_index(begin = var_281_begin_0, end = var_281_end_0, end_mask = var_281_end_mask_0, x = var_243_cast)[name = tensor<string, []>("op_281_cast")];
            tensor<int32, [4]> var_282_begin_0 = const()[name = tensor<string, []>("op_282_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_282_end_0 = const()[name = tensor<string, []>("op_282_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_282_end_mask_0 = const()[name = tensor<string, []>("op_282_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_282_cast = slice_by_index(begin = var_282_begin_0, end = var_282_end_0, end_mask = var_282_end_mask_0, x = var_247_cast)[name = tensor<string, []>("op_282_cast")];
            tensor<int32, [4]> var_283_begin_0 = const()[name = tensor<string, []>("op_283_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_283_end_0 = const()[name = tensor<string, []>("op_283_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_283_end_mask_0 = const()[name = tensor<string, []>("op_283_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_283_cast = slice_by_index(begin = var_283_begin_0, end = var_283_end_0, end_mask = var_283_end_mask_0, x = var_247_cast)[name = tensor<string, []>("op_283_cast")];
            tensor<int32, [4]> var_284_begin_0 = const()[name = tensor<string, []>("op_284_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_284_end_0 = const()[name = tensor<string, []>("op_284_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_284_end_mask_0 = const()[name = tensor<string, []>("op_284_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_284_cast = slice_by_index(begin = var_284_begin_0, end = var_284_end_0, end_mask = var_284_end_mask_0, x = var_247_cast)[name = tensor<string, []>("op_284_cast")];
            tensor<int32, [4]> var_285_begin_0 = const()[name = tensor<string, []>("op_285_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_285_end_0 = const()[name = tensor<string, []>("op_285_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_285_end_mask_0 = const()[name = tensor<string, []>("op_285_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_285_cast = slice_by_index(begin = var_285_begin_0, end = var_285_end_0, end_mask = var_285_end_mask_0, x = var_247_cast)[name = tensor<string, []>("op_285_cast")];
            tensor<int32, [4]> var_286_begin_0 = const()[name = tensor<string, []>("op_286_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_286_end_0 = const()[name = tensor<string, []>("op_286_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_286_end_mask_0 = const()[name = tensor<string, []>("op_286_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_286_cast = slice_by_index(begin = var_286_begin_0, end = var_286_end_0, end_mask = var_286_end_mask_0, x = var_247_cast)[name = tensor<string, []>("op_286_cast")];
            tensor<int32, [4]> var_287_begin_0 = const()[name = tensor<string, []>("op_287_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_287_end_0 = const()[name = tensor<string, []>("op_287_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_287_end_mask_0 = const()[name = tensor<string, []>("op_287_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_287_cast = slice_by_index(begin = var_287_begin_0, end = var_287_end_0, end_mask = var_287_end_mask_0, x = var_247_cast)[name = tensor<string, []>("op_287_cast")];
            tensor<int32, [4]> var_288_begin_0 = const()[name = tensor<string, []>("op_288_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_288_end_0 = const()[name = tensor<string, []>("op_288_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_288_end_mask_0 = const()[name = tensor<string, []>("op_288_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_288_cast = slice_by_index(begin = var_288_begin_0, end = var_288_end_0, end_mask = var_288_end_mask_0, x = var_247_cast)[name = tensor<string, []>("op_288_cast")];
            tensor<int32, [4]> var_289_begin_0 = const()[name = tensor<string, []>("op_289_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_289_end_0 = const()[name = tensor<string, []>("op_289_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_289_end_mask_0 = const()[name = tensor<string, []>("op_289_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_289_cast = slice_by_index(begin = var_289_begin_0, end = var_289_end_0, end_mask = var_289_end_mask_0, x = var_247_cast)[name = tensor<string, []>("op_289_cast")];
            tensor<int32, [4]> var_290_begin_0 = const()[name = tensor<string, []>("op_290_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_290_end_0 = const()[name = tensor<string, []>("op_290_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_290_end_mask_0 = const()[name = tensor<string, []>("op_290_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_290_cast = slice_by_index(begin = var_290_begin_0, end = var_290_end_0, end_mask = var_290_end_mask_0, x = var_251_cast)[name = tensor<string, []>("op_290_cast")];
            tensor<int32, [4]> var_291_begin_0 = const()[name = tensor<string, []>("op_291_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_291_end_0 = const()[name = tensor<string, []>("op_291_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_291_end_mask_0 = const()[name = tensor<string, []>("op_291_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_291_cast = slice_by_index(begin = var_291_begin_0, end = var_291_end_0, end_mask = var_291_end_mask_0, x = var_251_cast)[name = tensor<string, []>("op_291_cast")];
            tensor<int32, [4]> var_292_begin_0 = const()[name = tensor<string, []>("op_292_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_292_end_0 = const()[name = tensor<string, []>("op_292_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_292_end_mask_0 = const()[name = tensor<string, []>("op_292_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_292_cast = slice_by_index(begin = var_292_begin_0, end = var_292_end_0, end_mask = var_292_end_mask_0, x = var_251_cast)[name = tensor<string, []>("op_292_cast")];
            tensor<int32, [4]> var_293_begin_0 = const()[name = tensor<string, []>("op_293_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_293_end_0 = const()[name = tensor<string, []>("op_293_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_293_end_mask_0 = const()[name = tensor<string, []>("op_293_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_293_cast = slice_by_index(begin = var_293_begin_0, end = var_293_end_0, end_mask = var_293_end_mask_0, x = var_251_cast)[name = tensor<string, []>("op_293_cast")];
            tensor<int32, [4]> var_294_begin_0 = const()[name = tensor<string, []>("op_294_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_294_end_0 = const()[name = tensor<string, []>("op_294_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_294_end_mask_0 = const()[name = tensor<string, []>("op_294_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_294_cast = slice_by_index(begin = var_294_begin_0, end = var_294_end_0, end_mask = var_294_end_mask_0, x = var_251_cast)[name = tensor<string, []>("op_294_cast")];
            tensor<int32, [4]> var_295_begin_0 = const()[name = tensor<string, []>("op_295_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_295_end_0 = const()[name = tensor<string, []>("op_295_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_295_end_mask_0 = const()[name = tensor<string, []>("op_295_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_295_cast = slice_by_index(begin = var_295_begin_0, end = var_295_end_0, end_mask = var_295_end_mask_0, x = var_251_cast)[name = tensor<string, []>("op_295_cast")];
            tensor<int32, [4]> var_296_begin_0 = const()[name = tensor<string, []>("op_296_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_296_end_0 = const()[name = tensor<string, []>("op_296_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_296_end_mask_0 = const()[name = tensor<string, []>("op_296_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_296_cast = slice_by_index(begin = var_296_begin_0, end = var_296_end_0, end_mask = var_296_end_mask_0, x = var_251_cast)[name = tensor<string, []>("op_296_cast")];
            tensor<int32, [4]> var_297_begin_0 = const()[name = tensor<string, []>("op_297_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_297_end_0 = const()[name = tensor<string, []>("op_297_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_297_end_mask_0 = const()[name = tensor<string, []>("op_297_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_297_cast = slice_by_index(begin = var_297_begin_0, end = var_297_end_0, end_mask = var_297_end_mask_0, x = var_251_cast)[name = tensor<string, []>("op_297_cast")];
            tensor<int32, [4]> var_298_begin_0 = const()[name = tensor<string, []>("op_298_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_298_end_0 = const()[name = tensor<string, []>("op_298_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_298_end_mask_0 = const()[name = tensor<string, []>("op_298_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_298_cast = slice_by_index(begin = var_298_begin_0, end = var_298_end_0, end_mask = var_298_end_mask_0, x = var_255_cast)[name = tensor<string, []>("op_298_cast")];
            tensor<int32, [4]> var_299_begin_0 = const()[name = tensor<string, []>("op_299_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_299_end_0 = const()[name = tensor<string, []>("op_299_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_299_end_mask_0 = const()[name = tensor<string, []>("op_299_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_299_cast = slice_by_index(begin = var_299_begin_0, end = var_299_end_0, end_mask = var_299_end_mask_0, x = var_255_cast)[name = tensor<string, []>("op_299_cast")];
            tensor<int32, [4]> var_300_begin_0 = const()[name = tensor<string, []>("op_300_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_300_end_0 = const()[name = tensor<string, []>("op_300_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_300_end_mask_0 = const()[name = tensor<string, []>("op_300_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_300_cast = slice_by_index(begin = var_300_begin_0, end = var_300_end_0, end_mask = var_300_end_mask_0, x = var_255_cast)[name = tensor<string, []>("op_300_cast")];
            tensor<int32, [4]> var_301_begin_0 = const()[name = tensor<string, []>("op_301_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_301_end_0 = const()[name = tensor<string, []>("op_301_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_301_end_mask_0 = const()[name = tensor<string, []>("op_301_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_301_cast = slice_by_index(begin = var_301_begin_0, end = var_301_end_0, end_mask = var_301_end_mask_0, x = var_255_cast)[name = tensor<string, []>("op_301_cast")];
            tensor<int32, [4]> var_302_begin_0 = const()[name = tensor<string, []>("op_302_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_302_end_0 = const()[name = tensor<string, []>("op_302_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_302_end_mask_0 = const()[name = tensor<string, []>("op_302_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_302_cast = slice_by_index(begin = var_302_begin_0, end = var_302_end_0, end_mask = var_302_end_mask_0, x = var_255_cast)[name = tensor<string, []>("op_302_cast")];
            tensor<int32, [4]> var_303_begin_0 = const()[name = tensor<string, []>("op_303_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_303_end_0 = const()[name = tensor<string, []>("op_303_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_303_end_mask_0 = const()[name = tensor<string, []>("op_303_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_303_cast = slice_by_index(begin = var_303_begin_0, end = var_303_end_0, end_mask = var_303_end_mask_0, x = var_255_cast)[name = tensor<string, []>("op_303_cast")];
            tensor<int32, [4]> var_304_begin_0 = const()[name = tensor<string, []>("op_304_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_304_end_0 = const()[name = tensor<string, []>("op_304_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_304_end_mask_0 = const()[name = tensor<string, []>("op_304_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_304_cast = slice_by_index(begin = var_304_begin_0, end = var_304_end_0, end_mask = var_304_end_mask_0, x = var_255_cast)[name = tensor<string, []>("op_304_cast")];
            tensor<int32, [4]> var_305_begin_0 = const()[name = tensor<string, []>("op_305_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_305_end_0 = const()[name = tensor<string, []>("op_305_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_305_end_mask_0 = const()[name = tensor<string, []>("op_305_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_305_cast = slice_by_index(begin = var_305_begin_0, end = var_305_end_0, end_mask = var_305_end_mask_0, x = var_255_cast)[name = tensor<string, []>("op_305_cast")];
            tensor<int32, [4]> var_306_begin_0 = const()[name = tensor<string, []>("op_306_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_306_end_0 = const()[name = tensor<string, []>("op_306_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_306_end_mask_0 = const()[name = tensor<string, []>("op_306_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_306_cast = slice_by_index(begin = var_306_begin_0, end = var_306_end_0, end_mask = var_306_end_mask_0, x = var_259_cast)[name = tensor<string, []>("op_306_cast")];
            tensor<int32, [4]> var_307_begin_0 = const()[name = tensor<string, []>("op_307_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_307_end_0 = const()[name = tensor<string, []>("op_307_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_307_end_mask_0 = const()[name = tensor<string, []>("op_307_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_307_cast = slice_by_index(begin = var_307_begin_0, end = var_307_end_0, end_mask = var_307_end_mask_0, x = var_259_cast)[name = tensor<string, []>("op_307_cast")];
            tensor<int32, [4]> var_308_begin_0 = const()[name = tensor<string, []>("op_308_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_308_end_0 = const()[name = tensor<string, []>("op_308_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_308_end_mask_0 = const()[name = tensor<string, []>("op_308_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_308_cast = slice_by_index(begin = var_308_begin_0, end = var_308_end_0, end_mask = var_308_end_mask_0, x = var_259_cast)[name = tensor<string, []>("op_308_cast")];
            tensor<int32, [4]> var_309_begin_0 = const()[name = tensor<string, []>("op_309_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_309_end_0 = const()[name = tensor<string, []>("op_309_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_309_end_mask_0 = const()[name = tensor<string, []>("op_309_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_309_cast = slice_by_index(begin = var_309_begin_0, end = var_309_end_0, end_mask = var_309_end_mask_0, x = var_259_cast)[name = tensor<string, []>("op_309_cast")];
            tensor<int32, [4]> var_310_begin_0 = const()[name = tensor<string, []>("op_310_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_310_end_0 = const()[name = tensor<string, []>("op_310_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_310_end_mask_0 = const()[name = tensor<string, []>("op_310_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_310_cast = slice_by_index(begin = var_310_begin_0, end = var_310_end_0, end_mask = var_310_end_mask_0, x = var_259_cast)[name = tensor<string, []>("op_310_cast")];
            tensor<int32, [4]> var_311_begin_0 = const()[name = tensor<string, []>("op_311_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_311_end_0 = const()[name = tensor<string, []>("op_311_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_311_end_mask_0 = const()[name = tensor<string, []>("op_311_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_311_cast = slice_by_index(begin = var_311_begin_0, end = var_311_end_0, end_mask = var_311_end_mask_0, x = var_259_cast)[name = tensor<string, []>("op_311_cast")];
            tensor<int32, [4]> var_312_begin_0 = const()[name = tensor<string, []>("op_312_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_312_end_0 = const()[name = tensor<string, []>("op_312_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_312_end_mask_0 = const()[name = tensor<string, []>("op_312_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_312_cast = slice_by_index(begin = var_312_begin_0, end = var_312_end_0, end_mask = var_312_end_mask_0, x = var_259_cast)[name = tensor<string, []>("op_312_cast")];
            tensor<int32, [4]> var_313_begin_0 = const()[name = tensor<string, []>("op_313_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_313_end_0 = const()[name = tensor<string, []>("op_313_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_313_end_mask_0 = const()[name = tensor<string, []>("op_313_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_313_cast = slice_by_index(begin = var_313_begin_0, end = var_313_end_0, end_mask = var_313_end_mask_0, x = var_259_cast)[name = tensor<string, []>("op_313_cast")];
            tensor<int32, [4]> var_314_begin_0 = const()[name = tensor<string, []>("op_314_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_314_end_0 = const()[name = tensor<string, []>("op_314_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_314_end_mask_0 = const()[name = tensor<string, []>("op_314_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_314_cast = slice_by_index(begin = var_314_begin_0, end = var_314_end_0, end_mask = var_314_end_mask_0, x = var_263_cast)[name = tensor<string, []>("op_314_cast")];
            tensor<int32, [4]> var_315_begin_0 = const()[name = tensor<string, []>("op_315_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_315_end_0 = const()[name = tensor<string, []>("op_315_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_315_end_mask_0 = const()[name = tensor<string, []>("op_315_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_315_cast = slice_by_index(begin = var_315_begin_0, end = var_315_end_0, end_mask = var_315_end_mask_0, x = var_263_cast)[name = tensor<string, []>("op_315_cast")];
            tensor<int32, [4]> var_316_begin_0 = const()[name = tensor<string, []>("op_316_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_316_end_0 = const()[name = tensor<string, []>("op_316_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_316_end_mask_0 = const()[name = tensor<string, []>("op_316_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_316_cast = slice_by_index(begin = var_316_begin_0, end = var_316_end_0, end_mask = var_316_end_mask_0, x = var_263_cast)[name = tensor<string, []>("op_316_cast")];
            tensor<int32, [4]> var_317_begin_0 = const()[name = tensor<string, []>("op_317_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_317_end_0 = const()[name = tensor<string, []>("op_317_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_317_end_mask_0 = const()[name = tensor<string, []>("op_317_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_317_cast = slice_by_index(begin = var_317_begin_0, end = var_317_end_0, end_mask = var_317_end_mask_0, x = var_263_cast)[name = tensor<string, []>("op_317_cast")];
            tensor<int32, [4]> var_318_begin_0 = const()[name = tensor<string, []>("op_318_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_318_end_0 = const()[name = tensor<string, []>("op_318_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_318_end_mask_0 = const()[name = tensor<string, []>("op_318_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_318_cast = slice_by_index(begin = var_318_begin_0, end = var_318_end_0, end_mask = var_318_end_mask_0, x = var_263_cast)[name = tensor<string, []>("op_318_cast")];
            tensor<int32, [4]> var_319_begin_0 = const()[name = tensor<string, []>("op_319_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_319_end_0 = const()[name = tensor<string, []>("op_319_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_319_end_mask_0 = const()[name = tensor<string, []>("op_319_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_319_cast = slice_by_index(begin = var_319_begin_0, end = var_319_end_0, end_mask = var_319_end_mask_0, x = var_263_cast)[name = tensor<string, []>("op_319_cast")];
            tensor<int32, [4]> var_320_begin_0 = const()[name = tensor<string, []>("op_320_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_320_end_0 = const()[name = tensor<string, []>("op_320_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_320_end_mask_0 = const()[name = tensor<string, []>("op_320_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_320_cast = slice_by_index(begin = var_320_begin_0, end = var_320_end_0, end_mask = var_320_end_mask_0, x = var_263_cast)[name = tensor<string, []>("op_320_cast")];
            tensor<int32, [4]> var_321_begin_0 = const()[name = tensor<string, []>("op_321_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_321_end_0 = const()[name = tensor<string, []>("op_321_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_321_end_mask_0 = const()[name = tensor<string, []>("op_321_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_321_cast = slice_by_index(begin = var_321_begin_0, end = var_321_end_0, end_mask = var_321_end_mask_0, x = var_263_cast)[name = tensor<string, []>("op_321_cast")];
            tensor<int32, [4]> var_322_begin_0 = const()[name = tensor<string, []>("op_322_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_322_end_0 = const()[name = tensor<string, []>("op_322_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_322_end_mask_0 = const()[name = tensor<string, []>("op_322_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_322_cast = slice_by_index(begin = var_322_begin_0, end = var_322_end_0, end_mask = var_322_end_mask_0, x = var_267_cast)[name = tensor<string, []>("op_322_cast")];
            tensor<int32, [4]> var_323_begin_0 = const()[name = tensor<string, []>("op_323_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_323_end_0 = const()[name = tensor<string, []>("op_323_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_323_end_mask_0 = const()[name = tensor<string, []>("op_323_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_323_cast = slice_by_index(begin = var_323_begin_0, end = var_323_end_0, end_mask = var_323_end_mask_0, x = var_267_cast)[name = tensor<string, []>("op_323_cast")];
            tensor<int32, [4]> var_324_begin_0 = const()[name = tensor<string, []>("op_324_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_324_end_0 = const()[name = tensor<string, []>("op_324_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_324_end_mask_0 = const()[name = tensor<string, []>("op_324_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_324_cast = slice_by_index(begin = var_324_begin_0, end = var_324_end_0, end_mask = var_324_end_mask_0, x = var_267_cast)[name = tensor<string, []>("op_324_cast")];
            tensor<int32, [4]> var_325_begin_0 = const()[name = tensor<string, []>("op_325_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_325_end_0 = const()[name = tensor<string, []>("op_325_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_325_end_mask_0 = const()[name = tensor<string, []>("op_325_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_325_cast = slice_by_index(begin = var_325_begin_0, end = var_325_end_0, end_mask = var_325_end_mask_0, x = var_267_cast)[name = tensor<string, []>("op_325_cast")];
            tensor<int32, [4]> var_326_begin_0 = const()[name = tensor<string, []>("op_326_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_326_end_0 = const()[name = tensor<string, []>("op_326_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_326_end_mask_0 = const()[name = tensor<string, []>("op_326_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_326_cast = slice_by_index(begin = var_326_begin_0, end = var_326_end_0, end_mask = var_326_end_mask_0, x = var_267_cast)[name = tensor<string, []>("op_326_cast")];
            tensor<int32, [4]> var_327_begin_0 = const()[name = tensor<string, []>("op_327_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_327_end_0 = const()[name = tensor<string, []>("op_327_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_327_end_mask_0 = const()[name = tensor<string, []>("op_327_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_327_cast = slice_by_index(begin = var_327_begin_0, end = var_327_end_0, end_mask = var_327_end_mask_0, x = var_267_cast)[name = tensor<string, []>("op_327_cast")];
            tensor<int32, [4]> var_328_begin_0 = const()[name = tensor<string, []>("op_328_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_328_end_0 = const()[name = tensor<string, []>("op_328_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_328_end_mask_0 = const()[name = tensor<string, []>("op_328_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_328_cast = slice_by_index(begin = var_328_begin_0, end = var_328_end_0, end_mask = var_328_end_mask_0, x = var_267_cast)[name = tensor<string, []>("op_328_cast")];
            tensor<int32, [4]> var_329_begin_0 = const()[name = tensor<string, []>("op_329_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_329_end_0 = const()[name = tensor<string, []>("op_329_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_329_end_mask_0 = const()[name = tensor<string, []>("op_329_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_329_cast = slice_by_index(begin = var_329_begin_0, end = var_329_end_0, end_mask = var_329_end_mask_0, x = var_267_cast)[name = tensor<string, []>("op_329_cast")];
            tensor<int32, [4]> var_330_begin_0 = const()[name = tensor<string, []>("op_330_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_330_end_0 = const()[name = tensor<string, []>("op_330_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_330_end_mask_0 = const()[name = tensor<string, []>("op_330_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_330_cast = slice_by_index(begin = var_330_begin_0, end = var_330_end_0, end_mask = var_330_end_mask_0, x = var_271_cast)[name = tensor<string, []>("op_330_cast")];
            tensor<int32, [4]> var_331_begin_0 = const()[name = tensor<string, []>("op_331_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_331_end_0 = const()[name = tensor<string, []>("op_331_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_331_end_mask_0 = const()[name = tensor<string, []>("op_331_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_331_cast = slice_by_index(begin = var_331_begin_0, end = var_331_end_0, end_mask = var_331_end_mask_0, x = var_271_cast)[name = tensor<string, []>("op_331_cast")];
            tensor<int32, [4]> var_332_begin_0 = const()[name = tensor<string, []>("op_332_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_332_end_0 = const()[name = tensor<string, []>("op_332_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_332_end_mask_0 = const()[name = tensor<string, []>("op_332_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_332_cast = slice_by_index(begin = var_332_begin_0, end = var_332_end_0, end_mask = var_332_end_mask_0, x = var_271_cast)[name = tensor<string, []>("op_332_cast")];
            tensor<int32, [4]> var_333_begin_0 = const()[name = tensor<string, []>("op_333_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_333_end_0 = const()[name = tensor<string, []>("op_333_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_333_end_mask_0 = const()[name = tensor<string, []>("op_333_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_333_cast = slice_by_index(begin = var_333_begin_0, end = var_333_end_0, end_mask = var_333_end_mask_0, x = var_271_cast)[name = tensor<string, []>("op_333_cast")];
            tensor<int32, [4]> var_334_begin_0 = const()[name = tensor<string, []>("op_334_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_334_end_0 = const()[name = tensor<string, []>("op_334_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_334_end_mask_0 = const()[name = tensor<string, []>("op_334_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_334_cast = slice_by_index(begin = var_334_begin_0, end = var_334_end_0, end_mask = var_334_end_mask_0, x = var_271_cast)[name = tensor<string, []>("op_334_cast")];
            tensor<int32, [4]> var_335_begin_0 = const()[name = tensor<string, []>("op_335_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_335_end_0 = const()[name = tensor<string, []>("op_335_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_335_end_mask_0 = const()[name = tensor<string, []>("op_335_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_335_cast = slice_by_index(begin = var_335_begin_0, end = var_335_end_0, end_mask = var_335_end_mask_0, x = var_271_cast)[name = tensor<string, []>("op_335_cast")];
            tensor<int32, [4]> var_336_begin_0 = const()[name = tensor<string, []>("op_336_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_336_end_0 = const()[name = tensor<string, []>("op_336_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_336_end_mask_0 = const()[name = tensor<string, []>("op_336_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_336_cast = slice_by_index(begin = var_336_begin_0, end = var_336_end_0, end_mask = var_336_end_mask_0, x = var_271_cast)[name = tensor<string, []>("op_336_cast")];
            tensor<int32, [4]> var_337_begin_0 = const()[name = tensor<string, []>("op_337_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_337_end_0 = const()[name = tensor<string, []>("op_337_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_337_end_mask_0 = const()[name = tensor<string, []>("op_337_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_337_cast = slice_by_index(begin = var_337_begin_0, end = var_337_end_0, end_mask = var_337_end_mask_0, x = var_271_cast)[name = tensor<string, []>("op_337_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_342_begin_0 = const()[name = tensor<string, []>("op_342_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_342_end_0 = const()[name = tensor<string, []>("op_342_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 40])];
            tensor<bool, [4]> var_342_end_mask_0 = const()[name = tensor<string, []>("op_342_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)[name = tensor<string, []>("transpose_31")];
            tensor<fp16, [2, 4096, 1, 40]> var_342_cast = slice_by_index(begin = var_342_begin_0, end = var_342_end_0, end_mask = var_342_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_342_cast")];
            tensor<int32, [4]> var_346_begin_0 = const()[name = tensor<string, []>("op_346_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 40])];
            tensor<int32, [4]> var_346_end_0 = const()[name = tensor<string, []>("op_346_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 80])];
            tensor<bool, [4]> var_346_end_mask_0 = const()[name = tensor<string, []>("op_346_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_346_cast = slice_by_index(begin = var_346_begin_0, end = var_346_end_0, end_mask = var_346_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_346_cast")];
            tensor<int32, [4]> var_350_begin_0 = const()[name = tensor<string, []>("op_350_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 80])];
            tensor<int32, [4]> var_350_end_0 = const()[name = tensor<string, []>("op_350_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 120])];
            tensor<bool, [4]> var_350_end_mask_0 = const()[name = tensor<string, []>("op_350_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_350_cast = slice_by_index(begin = var_350_begin_0, end = var_350_end_0, end_mask = var_350_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_350_cast")];
            tensor<int32, [4]> var_354_begin_0 = const()[name = tensor<string, []>("op_354_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 120])];
            tensor<int32, [4]> var_354_end_0 = const()[name = tensor<string, []>("op_354_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 160])];
            tensor<bool, [4]> var_354_end_mask_0 = const()[name = tensor<string, []>("op_354_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_354_cast = slice_by_index(begin = var_354_begin_0, end = var_354_end_0, end_mask = var_354_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_354_cast")];
            tensor<int32, [4]> var_358_begin_0 = const()[name = tensor<string, []>("op_358_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_358_end_0 = const()[name = tensor<string, []>("op_358_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 200])];
            tensor<bool, [4]> var_358_end_mask_0 = const()[name = tensor<string, []>("op_358_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_358_cast = slice_by_index(begin = var_358_begin_0, end = var_358_end_0, end_mask = var_358_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_358_cast")];
            tensor<int32, [4]> var_362_begin_0 = const()[name = tensor<string, []>("op_362_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 200])];
            tensor<int32, [4]> var_362_end_0 = const()[name = tensor<string, []>("op_362_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 240])];
            tensor<bool, [4]> var_362_end_mask_0 = const()[name = tensor<string, []>("op_362_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_362_cast = slice_by_index(begin = var_362_begin_0, end = var_362_end_0, end_mask = var_362_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_362_cast")];
            tensor<int32, [4]> var_366_begin_0 = const()[name = tensor<string, []>("op_366_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 240])];
            tensor<int32, [4]> var_366_end_0 = const()[name = tensor<string, []>("op_366_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 280])];
            tensor<bool, [4]> var_366_end_mask_0 = const()[name = tensor<string, []>("op_366_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_366_cast = slice_by_index(begin = var_366_begin_0, end = var_366_end_0, end_mask = var_366_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_366_cast")];
            tensor<int32, [4]> var_370_begin_0 = const()[name = tensor<string, []>("op_370_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 280])];
            tensor<int32, [4]> var_370_end_0 = const()[name = tensor<string, []>("op_370_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_370_end_mask_0 = const()[name = tensor<string, []>("op_370_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_370_cast = slice_by_index(begin = var_370_begin_0, end = var_370_end_0, end_mask = var_370_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_370_cast")];
            tensor<int32, [4]> var_372_begin_0 = const()[name = tensor<string, []>("op_372_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_372_end_0 = const()[name = tensor<string, []>("op_372_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_372_end_mask_0 = const()[name = tensor<string, []>("op_372_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_372_cast = slice_by_index(begin = var_372_begin_0, end = var_372_end_0, end_mask = var_372_end_mask_0, x = v_1_cast)[name = tensor<string, []>("op_372_cast")];
            tensor<int32, [4]> var_376_begin_0 = const()[name = tensor<string, []>("op_376_begin_0"), val = tensor<int32, [4]>([0, 40, 0, 0])];
            tensor<int32, [4]> var_376_end_0 = const()[name = tensor<string, []>("op_376_end_0"), val = tensor<int32, [4]>([2, 80, 1, 4096])];
            tensor<bool, [4]> var_376_end_mask_0 = const()[name = tensor<string, []>("op_376_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_376_cast = slice_by_index(begin = var_376_begin_0, end = var_376_end_0, end_mask = var_376_end_mask_0, x = v_1_cast)[name = tensor<string, []>("op_376_cast")];
            tensor<int32, [4]> var_380_begin_0 = const()[name = tensor<string, []>("op_380_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_380_end_0 = const()[name = tensor<string, []>("op_380_end_0"), val = tensor<int32, [4]>([2, 120, 1, 4096])];
            tensor<bool, [4]> var_380_end_mask_0 = const()[name = tensor<string, []>("op_380_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_380_cast = slice_by_index(begin = var_380_begin_0, end = var_380_end_0, end_mask = var_380_end_mask_0, x = v_1_cast)[name = tensor<string, []>("op_380_cast")];
            tensor<int32, [4]> var_384_begin_0 = const()[name = tensor<string, []>("op_384_begin_0"), val = tensor<int32, [4]>([0, 120, 0, 0])];
            tensor<int32, [4]> var_384_end_0 = const()[name = tensor<string, []>("op_384_end_0"), val = tensor<int32, [4]>([2, 160, 1, 4096])];
            tensor<bool, [4]> var_384_end_mask_0 = const()[name = tensor<string, []>("op_384_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_384_cast = slice_by_index(begin = var_384_begin_0, end = var_384_end_0, end_mask = var_384_end_mask_0, x = v_1_cast)[name = tensor<string, []>("op_384_cast")];
            tensor<int32, [4]> var_388_begin_0 = const()[name = tensor<string, []>("op_388_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_388_end_0 = const()[name = tensor<string, []>("op_388_end_0"), val = tensor<int32, [4]>([2, 200, 1, 4096])];
            tensor<bool, [4]> var_388_end_mask_0 = const()[name = tensor<string, []>("op_388_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_388_cast = slice_by_index(begin = var_388_begin_0, end = var_388_end_0, end_mask = var_388_end_mask_0, x = v_1_cast)[name = tensor<string, []>("op_388_cast")];
            tensor<int32, [4]> var_392_begin_0 = const()[name = tensor<string, []>("op_392_begin_0"), val = tensor<int32, [4]>([0, 200, 0, 0])];
            tensor<int32, [4]> var_392_end_0 = const()[name = tensor<string, []>("op_392_end_0"), val = tensor<int32, [4]>([2, 240, 1, 4096])];
            tensor<bool, [4]> var_392_end_mask_0 = const()[name = tensor<string, []>("op_392_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_392_cast = slice_by_index(begin = var_392_begin_0, end = var_392_end_0, end_mask = var_392_end_mask_0, x = v_1_cast)[name = tensor<string, []>("op_392_cast")];
            tensor<int32, [4]> var_396_begin_0 = const()[name = tensor<string, []>("op_396_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_396_end_0 = const()[name = tensor<string, []>("op_396_end_0"), val = tensor<int32, [4]>([2, 280, 1, 4096])];
            tensor<bool, [4]> var_396_end_mask_0 = const()[name = tensor<string, []>("op_396_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_396_cast = slice_by_index(begin = var_396_begin_0, end = var_396_end_0, end_mask = var_396_end_mask_0, x = v_1_cast)[name = tensor<string, []>("op_396_cast")];
            tensor<int32, [4]> var_400_begin_0 = const()[name = tensor<string, []>("op_400_begin_0"), val = tensor<int32, [4]>([0, 280, 0, 0])];
            tensor<int32, [4]> var_400_end_0 = const()[name = tensor<string, []>("op_400_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_400_end_mask_0 = const()[name = tensor<string, []>("op_400_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_400_cast = slice_by_index(begin = var_400_begin_0, end = var_400_end_0, end_mask = var_400_end_mask_0, x = v_1_cast)[name = tensor<string, []>("op_400_cast")];
            tensor<string, []> var_404_equation_0 = const()[name = tensor<string, []>("op_404_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_404_cast = einsum(equation = var_404_equation_0, values = (var_342_cast, var_274_cast))[name = tensor<string, []>("op_404_cast")];
            tensor<fp16, []> var_405_to_fp16 = const()[name = tensor<string, []>("op_405_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1_cast = mul(x = var_404_cast, y = var_405_to_fp16)[name = tensor<string, []>("aw_chunk_1_cast")];
            tensor<string, []> var_408_equation_0 = const()[name = tensor<string, []>("op_408_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_408_cast = einsum(equation = var_408_equation_0, values = (var_342_cast, var_275_cast))[name = tensor<string, []>("op_408_cast")];
            tensor<fp16, []> var_409_to_fp16 = const()[name = tensor<string, []>("op_409_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_3_cast = mul(x = var_408_cast, y = var_409_to_fp16)[name = tensor<string, []>("aw_chunk_3_cast")];
            tensor<string, []> var_412_equation_0 = const()[name = tensor<string, []>("op_412_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_412_cast = einsum(equation = var_412_equation_0, values = (var_342_cast, var_276_cast))[name = tensor<string, []>("op_412_cast")];
            tensor<fp16, []> var_413_to_fp16 = const()[name = tensor<string, []>("op_413_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_5_cast = mul(x = var_412_cast, y = var_413_to_fp16)[name = tensor<string, []>("aw_chunk_5_cast")];
            tensor<string, []> var_416_equation_0 = const()[name = tensor<string, []>("op_416_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_416_cast = einsum(equation = var_416_equation_0, values = (var_342_cast, var_277_cast))[name = tensor<string, []>("op_416_cast")];
            tensor<fp16, []> var_417_to_fp16 = const()[name = tensor<string, []>("op_417_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_7_cast = mul(x = var_416_cast, y = var_417_to_fp16)[name = tensor<string, []>("aw_chunk_7_cast")];
            tensor<string, []> var_420_equation_0 = const()[name = tensor<string, []>("op_420_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_420_cast = einsum(equation = var_420_equation_0, values = (var_342_cast, var_278_cast))[name = tensor<string, []>("op_420_cast")];
            tensor<fp16, []> var_421_to_fp16 = const()[name = tensor<string, []>("op_421_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_9_cast = mul(x = var_420_cast, y = var_421_to_fp16)[name = tensor<string, []>("aw_chunk_9_cast")];
            tensor<string, []> var_424_equation_0 = const()[name = tensor<string, []>("op_424_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_424_cast = einsum(equation = var_424_equation_0, values = (var_342_cast, var_279_cast))[name = tensor<string, []>("op_424_cast")];
            tensor<fp16, []> var_425_to_fp16 = const()[name = tensor<string, []>("op_425_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_11_cast = mul(x = var_424_cast, y = var_425_to_fp16)[name = tensor<string, []>("aw_chunk_11_cast")];
            tensor<string, []> var_428_equation_0 = const()[name = tensor<string, []>("op_428_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_428_cast = einsum(equation = var_428_equation_0, values = (var_342_cast, var_280_cast))[name = tensor<string, []>("op_428_cast")];
            tensor<fp16, []> var_429_to_fp16 = const()[name = tensor<string, []>("op_429_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_13_cast = mul(x = var_428_cast, y = var_429_to_fp16)[name = tensor<string, []>("aw_chunk_13_cast")];
            tensor<string, []> var_432_equation_0 = const()[name = tensor<string, []>("op_432_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_432_cast = einsum(equation = var_432_equation_0, values = (var_342_cast, var_281_cast))[name = tensor<string, []>("op_432_cast")];
            tensor<fp16, []> var_433_to_fp16 = const()[name = tensor<string, []>("op_433_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_15_cast = mul(x = var_432_cast, y = var_433_to_fp16)[name = tensor<string, []>("aw_chunk_15_cast")];
            tensor<string, []> var_436_equation_0 = const()[name = tensor<string, []>("op_436_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_436_cast = einsum(equation = var_436_equation_0, values = (var_346_cast, var_282_cast))[name = tensor<string, []>("op_436_cast")];
            tensor<fp16, []> var_437_to_fp16 = const()[name = tensor<string, []>("op_437_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_17_cast = mul(x = var_436_cast, y = var_437_to_fp16)[name = tensor<string, []>("aw_chunk_17_cast")];
            tensor<string, []> var_440_equation_0 = const()[name = tensor<string, []>("op_440_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_440_cast = einsum(equation = var_440_equation_0, values = (var_346_cast, var_283_cast))[name = tensor<string, []>("op_440_cast")];
            tensor<fp16, []> var_441_to_fp16 = const()[name = tensor<string, []>("op_441_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_19_cast = mul(x = var_440_cast, y = var_441_to_fp16)[name = tensor<string, []>("aw_chunk_19_cast")];
            tensor<string, []> var_444_equation_0 = const()[name = tensor<string, []>("op_444_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_444_cast = einsum(equation = var_444_equation_0, values = (var_346_cast, var_284_cast))[name = tensor<string, []>("op_444_cast")];
            tensor<fp16, []> var_445_to_fp16 = const()[name = tensor<string, []>("op_445_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_21_cast = mul(x = var_444_cast, y = var_445_to_fp16)[name = tensor<string, []>("aw_chunk_21_cast")];
            tensor<string, []> var_448_equation_0 = const()[name = tensor<string, []>("op_448_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_448_cast = einsum(equation = var_448_equation_0, values = (var_346_cast, var_285_cast))[name = tensor<string, []>("op_448_cast")];
            tensor<fp16, []> var_449_to_fp16 = const()[name = tensor<string, []>("op_449_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_23_cast = mul(x = var_448_cast, y = var_449_to_fp16)[name = tensor<string, []>("aw_chunk_23_cast")];
            tensor<string, []> var_452_equation_0 = const()[name = tensor<string, []>("op_452_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_452_cast = einsum(equation = var_452_equation_0, values = (var_346_cast, var_286_cast))[name = tensor<string, []>("op_452_cast")];
            tensor<fp16, []> var_453_to_fp16 = const()[name = tensor<string, []>("op_453_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_25_cast = mul(x = var_452_cast, y = var_453_to_fp16)[name = tensor<string, []>("aw_chunk_25_cast")];
            tensor<string, []> var_456_equation_0 = const()[name = tensor<string, []>("op_456_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_456_cast = einsum(equation = var_456_equation_0, values = (var_346_cast, var_287_cast))[name = tensor<string, []>("op_456_cast")];
            tensor<fp16, []> var_457_to_fp16 = const()[name = tensor<string, []>("op_457_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_27_cast = mul(x = var_456_cast, y = var_457_to_fp16)[name = tensor<string, []>("aw_chunk_27_cast")];
            tensor<string, []> var_460_equation_0 = const()[name = tensor<string, []>("op_460_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_460_cast = einsum(equation = var_460_equation_0, values = (var_346_cast, var_288_cast))[name = tensor<string, []>("op_460_cast")];
            tensor<fp16, []> var_461_to_fp16 = const()[name = tensor<string, []>("op_461_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_29_cast = mul(x = var_460_cast, y = var_461_to_fp16)[name = tensor<string, []>("aw_chunk_29_cast")];
            tensor<string, []> var_464_equation_0 = const()[name = tensor<string, []>("op_464_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_464_cast = einsum(equation = var_464_equation_0, values = (var_346_cast, var_289_cast))[name = tensor<string, []>("op_464_cast")];
            tensor<fp16, []> var_465_to_fp16 = const()[name = tensor<string, []>("op_465_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_31_cast = mul(x = var_464_cast, y = var_465_to_fp16)[name = tensor<string, []>("aw_chunk_31_cast")];
            tensor<string, []> var_468_equation_0 = const()[name = tensor<string, []>("op_468_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_468_cast = einsum(equation = var_468_equation_0, values = (var_350_cast, var_290_cast))[name = tensor<string, []>("op_468_cast")];
            tensor<fp16, []> var_469_to_fp16 = const()[name = tensor<string, []>("op_469_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_33_cast = mul(x = var_468_cast, y = var_469_to_fp16)[name = tensor<string, []>("aw_chunk_33_cast")];
            tensor<string, []> var_472_equation_0 = const()[name = tensor<string, []>("op_472_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_472_cast = einsum(equation = var_472_equation_0, values = (var_350_cast, var_291_cast))[name = tensor<string, []>("op_472_cast")];
            tensor<fp16, []> var_473_to_fp16 = const()[name = tensor<string, []>("op_473_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_35_cast = mul(x = var_472_cast, y = var_473_to_fp16)[name = tensor<string, []>("aw_chunk_35_cast")];
            tensor<string, []> var_476_equation_0 = const()[name = tensor<string, []>("op_476_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_476_cast = einsum(equation = var_476_equation_0, values = (var_350_cast, var_292_cast))[name = tensor<string, []>("op_476_cast")];
            tensor<fp16, []> var_477_to_fp16 = const()[name = tensor<string, []>("op_477_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_37_cast = mul(x = var_476_cast, y = var_477_to_fp16)[name = tensor<string, []>("aw_chunk_37_cast")];
            tensor<string, []> var_480_equation_0 = const()[name = tensor<string, []>("op_480_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_480_cast = einsum(equation = var_480_equation_0, values = (var_350_cast, var_293_cast))[name = tensor<string, []>("op_480_cast")];
            tensor<fp16, []> var_481_to_fp16 = const()[name = tensor<string, []>("op_481_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_39_cast = mul(x = var_480_cast, y = var_481_to_fp16)[name = tensor<string, []>("aw_chunk_39_cast")];
            tensor<string, []> var_484_equation_0 = const()[name = tensor<string, []>("op_484_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_484_cast = einsum(equation = var_484_equation_0, values = (var_350_cast, var_294_cast))[name = tensor<string, []>("op_484_cast")];
            tensor<fp16, []> var_485_to_fp16 = const()[name = tensor<string, []>("op_485_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_41_cast = mul(x = var_484_cast, y = var_485_to_fp16)[name = tensor<string, []>("aw_chunk_41_cast")];
            tensor<string, []> var_488_equation_0 = const()[name = tensor<string, []>("op_488_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_488_cast = einsum(equation = var_488_equation_0, values = (var_350_cast, var_295_cast))[name = tensor<string, []>("op_488_cast")];
            tensor<fp16, []> var_489_to_fp16 = const()[name = tensor<string, []>("op_489_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_43_cast = mul(x = var_488_cast, y = var_489_to_fp16)[name = tensor<string, []>("aw_chunk_43_cast")];
            tensor<string, []> var_492_equation_0 = const()[name = tensor<string, []>("op_492_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_492_cast = einsum(equation = var_492_equation_0, values = (var_350_cast, var_296_cast))[name = tensor<string, []>("op_492_cast")];
            tensor<fp16, []> var_493_to_fp16 = const()[name = tensor<string, []>("op_493_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_45_cast = mul(x = var_492_cast, y = var_493_to_fp16)[name = tensor<string, []>("aw_chunk_45_cast")];
            tensor<string, []> var_496_equation_0 = const()[name = tensor<string, []>("op_496_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_496_cast = einsum(equation = var_496_equation_0, values = (var_350_cast, var_297_cast))[name = tensor<string, []>("op_496_cast")];
            tensor<fp16, []> var_497_to_fp16 = const()[name = tensor<string, []>("op_497_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_47_cast = mul(x = var_496_cast, y = var_497_to_fp16)[name = tensor<string, []>("aw_chunk_47_cast")];
            tensor<string, []> var_500_equation_0 = const()[name = tensor<string, []>("op_500_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_500_cast = einsum(equation = var_500_equation_0, values = (var_354_cast, var_298_cast))[name = tensor<string, []>("op_500_cast")];
            tensor<fp16, []> var_501_to_fp16 = const()[name = tensor<string, []>("op_501_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_49_cast = mul(x = var_500_cast, y = var_501_to_fp16)[name = tensor<string, []>("aw_chunk_49_cast")];
            tensor<string, []> var_504_equation_0 = const()[name = tensor<string, []>("op_504_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_504_cast = einsum(equation = var_504_equation_0, values = (var_354_cast, var_299_cast))[name = tensor<string, []>("op_504_cast")];
            tensor<fp16, []> var_505_to_fp16 = const()[name = tensor<string, []>("op_505_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_51_cast = mul(x = var_504_cast, y = var_505_to_fp16)[name = tensor<string, []>("aw_chunk_51_cast")];
            tensor<string, []> var_508_equation_0 = const()[name = tensor<string, []>("op_508_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_508_cast = einsum(equation = var_508_equation_0, values = (var_354_cast, var_300_cast))[name = tensor<string, []>("op_508_cast")];
            tensor<fp16, []> var_509_to_fp16 = const()[name = tensor<string, []>("op_509_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_53_cast = mul(x = var_508_cast, y = var_509_to_fp16)[name = tensor<string, []>("aw_chunk_53_cast")];
            tensor<string, []> var_512_equation_0 = const()[name = tensor<string, []>("op_512_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_512_cast = einsum(equation = var_512_equation_0, values = (var_354_cast, var_301_cast))[name = tensor<string, []>("op_512_cast")];
            tensor<fp16, []> var_513_to_fp16 = const()[name = tensor<string, []>("op_513_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_55_cast = mul(x = var_512_cast, y = var_513_to_fp16)[name = tensor<string, []>("aw_chunk_55_cast")];
            tensor<string, []> var_516_equation_0 = const()[name = tensor<string, []>("op_516_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_516_cast = einsum(equation = var_516_equation_0, values = (var_354_cast, var_302_cast))[name = tensor<string, []>("op_516_cast")];
            tensor<fp16, []> var_517_to_fp16 = const()[name = tensor<string, []>("op_517_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_57_cast = mul(x = var_516_cast, y = var_517_to_fp16)[name = tensor<string, []>("aw_chunk_57_cast")];
            tensor<string, []> var_520_equation_0 = const()[name = tensor<string, []>("op_520_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_520_cast = einsum(equation = var_520_equation_0, values = (var_354_cast, var_303_cast))[name = tensor<string, []>("op_520_cast")];
            tensor<fp16, []> var_521_to_fp16 = const()[name = tensor<string, []>("op_521_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_59_cast = mul(x = var_520_cast, y = var_521_to_fp16)[name = tensor<string, []>("aw_chunk_59_cast")];
            tensor<string, []> var_524_equation_0 = const()[name = tensor<string, []>("op_524_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_524_cast = einsum(equation = var_524_equation_0, values = (var_354_cast, var_304_cast))[name = tensor<string, []>("op_524_cast")];
            tensor<fp16, []> var_525_to_fp16 = const()[name = tensor<string, []>("op_525_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_61_cast = mul(x = var_524_cast, y = var_525_to_fp16)[name = tensor<string, []>("aw_chunk_61_cast")];
            tensor<string, []> var_528_equation_0 = const()[name = tensor<string, []>("op_528_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_528_cast = einsum(equation = var_528_equation_0, values = (var_354_cast, var_305_cast))[name = tensor<string, []>("op_528_cast")];
            tensor<fp16, []> var_529_to_fp16 = const()[name = tensor<string, []>("op_529_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_63_cast = mul(x = var_528_cast, y = var_529_to_fp16)[name = tensor<string, []>("aw_chunk_63_cast")];
            tensor<string, []> var_532_equation_0 = const()[name = tensor<string, []>("op_532_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_532_cast = einsum(equation = var_532_equation_0, values = (var_358_cast, var_306_cast))[name = tensor<string, []>("op_532_cast")];
            tensor<fp16, []> var_533_to_fp16 = const()[name = tensor<string, []>("op_533_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_65_cast = mul(x = var_532_cast, y = var_533_to_fp16)[name = tensor<string, []>("aw_chunk_65_cast")];
            tensor<string, []> var_536_equation_0 = const()[name = tensor<string, []>("op_536_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_536_cast = einsum(equation = var_536_equation_0, values = (var_358_cast, var_307_cast))[name = tensor<string, []>("op_536_cast")];
            tensor<fp16, []> var_537_to_fp16 = const()[name = tensor<string, []>("op_537_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_67_cast = mul(x = var_536_cast, y = var_537_to_fp16)[name = tensor<string, []>("aw_chunk_67_cast")];
            tensor<string, []> var_540_equation_0 = const()[name = tensor<string, []>("op_540_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_540_cast = einsum(equation = var_540_equation_0, values = (var_358_cast, var_308_cast))[name = tensor<string, []>("op_540_cast")];
            tensor<fp16, []> var_541_to_fp16 = const()[name = tensor<string, []>("op_541_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_69_cast = mul(x = var_540_cast, y = var_541_to_fp16)[name = tensor<string, []>("aw_chunk_69_cast")];
            tensor<string, []> var_544_equation_0 = const()[name = tensor<string, []>("op_544_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_544_cast = einsum(equation = var_544_equation_0, values = (var_358_cast, var_309_cast))[name = tensor<string, []>("op_544_cast")];
            tensor<fp16, []> var_545_to_fp16 = const()[name = tensor<string, []>("op_545_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_71_cast = mul(x = var_544_cast, y = var_545_to_fp16)[name = tensor<string, []>("aw_chunk_71_cast")];
            tensor<string, []> var_548_equation_0 = const()[name = tensor<string, []>("op_548_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_548_cast = einsum(equation = var_548_equation_0, values = (var_358_cast, var_310_cast))[name = tensor<string, []>("op_548_cast")];
            tensor<fp16, []> var_549_to_fp16 = const()[name = tensor<string, []>("op_549_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_73_cast = mul(x = var_548_cast, y = var_549_to_fp16)[name = tensor<string, []>("aw_chunk_73_cast")];
            tensor<string, []> var_552_equation_0 = const()[name = tensor<string, []>("op_552_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_552_cast = einsum(equation = var_552_equation_0, values = (var_358_cast, var_311_cast))[name = tensor<string, []>("op_552_cast")];
            tensor<fp16, []> var_553_to_fp16 = const()[name = tensor<string, []>("op_553_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_75_cast = mul(x = var_552_cast, y = var_553_to_fp16)[name = tensor<string, []>("aw_chunk_75_cast")];
            tensor<string, []> var_556_equation_0 = const()[name = tensor<string, []>("op_556_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_556_cast = einsum(equation = var_556_equation_0, values = (var_358_cast, var_312_cast))[name = tensor<string, []>("op_556_cast")];
            tensor<fp16, []> var_557_to_fp16 = const()[name = tensor<string, []>("op_557_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_77_cast = mul(x = var_556_cast, y = var_557_to_fp16)[name = tensor<string, []>("aw_chunk_77_cast")];
            tensor<string, []> var_560_equation_0 = const()[name = tensor<string, []>("op_560_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_560_cast = einsum(equation = var_560_equation_0, values = (var_358_cast, var_313_cast))[name = tensor<string, []>("op_560_cast")];
            tensor<fp16, []> var_561_to_fp16 = const()[name = tensor<string, []>("op_561_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_79_cast = mul(x = var_560_cast, y = var_561_to_fp16)[name = tensor<string, []>("aw_chunk_79_cast")];
            tensor<string, []> var_564_equation_0 = const()[name = tensor<string, []>("op_564_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_564_cast = einsum(equation = var_564_equation_0, values = (var_362_cast, var_314_cast))[name = tensor<string, []>("op_564_cast")];
            tensor<fp16, []> var_565_to_fp16 = const()[name = tensor<string, []>("op_565_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_81_cast = mul(x = var_564_cast, y = var_565_to_fp16)[name = tensor<string, []>("aw_chunk_81_cast")];
            tensor<string, []> var_568_equation_0 = const()[name = tensor<string, []>("op_568_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_568_cast = einsum(equation = var_568_equation_0, values = (var_362_cast, var_315_cast))[name = tensor<string, []>("op_568_cast")];
            tensor<fp16, []> var_569_to_fp16 = const()[name = tensor<string, []>("op_569_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_83_cast = mul(x = var_568_cast, y = var_569_to_fp16)[name = tensor<string, []>("aw_chunk_83_cast")];
            tensor<string, []> var_572_equation_0 = const()[name = tensor<string, []>("op_572_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_572_cast = einsum(equation = var_572_equation_0, values = (var_362_cast, var_316_cast))[name = tensor<string, []>("op_572_cast")];
            tensor<fp16, []> var_573_to_fp16 = const()[name = tensor<string, []>("op_573_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_85_cast = mul(x = var_572_cast, y = var_573_to_fp16)[name = tensor<string, []>("aw_chunk_85_cast")];
            tensor<string, []> var_576_equation_0 = const()[name = tensor<string, []>("op_576_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_576_cast = einsum(equation = var_576_equation_0, values = (var_362_cast, var_317_cast))[name = tensor<string, []>("op_576_cast")];
            tensor<fp16, []> var_577_to_fp16 = const()[name = tensor<string, []>("op_577_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_87_cast = mul(x = var_576_cast, y = var_577_to_fp16)[name = tensor<string, []>("aw_chunk_87_cast")];
            tensor<string, []> var_580_equation_0 = const()[name = tensor<string, []>("op_580_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_580_cast = einsum(equation = var_580_equation_0, values = (var_362_cast, var_318_cast))[name = tensor<string, []>("op_580_cast")];
            tensor<fp16, []> var_581_to_fp16 = const()[name = tensor<string, []>("op_581_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_89_cast = mul(x = var_580_cast, y = var_581_to_fp16)[name = tensor<string, []>("aw_chunk_89_cast")];
            tensor<string, []> var_584_equation_0 = const()[name = tensor<string, []>("op_584_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_584_cast = einsum(equation = var_584_equation_0, values = (var_362_cast, var_319_cast))[name = tensor<string, []>("op_584_cast")];
            tensor<fp16, []> var_585_to_fp16 = const()[name = tensor<string, []>("op_585_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_91_cast = mul(x = var_584_cast, y = var_585_to_fp16)[name = tensor<string, []>("aw_chunk_91_cast")];
            tensor<string, []> var_588_equation_0 = const()[name = tensor<string, []>("op_588_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_588_cast = einsum(equation = var_588_equation_0, values = (var_362_cast, var_320_cast))[name = tensor<string, []>("op_588_cast")];
            tensor<fp16, []> var_589_to_fp16 = const()[name = tensor<string, []>("op_589_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_93_cast = mul(x = var_588_cast, y = var_589_to_fp16)[name = tensor<string, []>("aw_chunk_93_cast")];
            tensor<string, []> var_592_equation_0 = const()[name = tensor<string, []>("op_592_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_592_cast = einsum(equation = var_592_equation_0, values = (var_362_cast, var_321_cast))[name = tensor<string, []>("op_592_cast")];
            tensor<fp16, []> var_593_to_fp16 = const()[name = tensor<string, []>("op_593_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_95_cast = mul(x = var_592_cast, y = var_593_to_fp16)[name = tensor<string, []>("aw_chunk_95_cast")];
            tensor<string, []> var_596_equation_0 = const()[name = tensor<string, []>("op_596_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_596_cast = einsum(equation = var_596_equation_0, values = (var_366_cast, var_322_cast))[name = tensor<string, []>("op_596_cast")];
            tensor<fp16, []> var_597_to_fp16 = const()[name = tensor<string, []>("op_597_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_97_cast = mul(x = var_596_cast, y = var_597_to_fp16)[name = tensor<string, []>("aw_chunk_97_cast")];
            tensor<string, []> var_600_equation_0 = const()[name = tensor<string, []>("op_600_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_600_cast = einsum(equation = var_600_equation_0, values = (var_366_cast, var_323_cast))[name = tensor<string, []>("op_600_cast")];
            tensor<fp16, []> var_601_to_fp16 = const()[name = tensor<string, []>("op_601_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_99_cast = mul(x = var_600_cast, y = var_601_to_fp16)[name = tensor<string, []>("aw_chunk_99_cast")];
            tensor<string, []> var_604_equation_0 = const()[name = tensor<string, []>("op_604_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_604_cast = einsum(equation = var_604_equation_0, values = (var_366_cast, var_324_cast))[name = tensor<string, []>("op_604_cast")];
            tensor<fp16, []> var_605_to_fp16 = const()[name = tensor<string, []>("op_605_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_101_cast = mul(x = var_604_cast, y = var_605_to_fp16)[name = tensor<string, []>("aw_chunk_101_cast")];
            tensor<string, []> var_608_equation_0 = const()[name = tensor<string, []>("op_608_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_608_cast = einsum(equation = var_608_equation_0, values = (var_366_cast, var_325_cast))[name = tensor<string, []>("op_608_cast")];
            tensor<fp16, []> var_609_to_fp16 = const()[name = tensor<string, []>("op_609_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_103_cast = mul(x = var_608_cast, y = var_609_to_fp16)[name = tensor<string, []>("aw_chunk_103_cast")];
            tensor<string, []> var_612_equation_0 = const()[name = tensor<string, []>("op_612_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_612_cast = einsum(equation = var_612_equation_0, values = (var_366_cast, var_326_cast))[name = tensor<string, []>("op_612_cast")];
            tensor<fp16, []> var_613_to_fp16 = const()[name = tensor<string, []>("op_613_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_105_cast = mul(x = var_612_cast, y = var_613_to_fp16)[name = tensor<string, []>("aw_chunk_105_cast")];
            tensor<string, []> var_616_equation_0 = const()[name = tensor<string, []>("op_616_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_616_cast = einsum(equation = var_616_equation_0, values = (var_366_cast, var_327_cast))[name = tensor<string, []>("op_616_cast")];
            tensor<fp16, []> var_617_to_fp16 = const()[name = tensor<string, []>("op_617_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_107_cast = mul(x = var_616_cast, y = var_617_to_fp16)[name = tensor<string, []>("aw_chunk_107_cast")];
            tensor<string, []> var_620_equation_0 = const()[name = tensor<string, []>("op_620_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_620_cast = einsum(equation = var_620_equation_0, values = (var_366_cast, var_328_cast))[name = tensor<string, []>("op_620_cast")];
            tensor<fp16, []> var_621_to_fp16 = const()[name = tensor<string, []>("op_621_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_109_cast = mul(x = var_620_cast, y = var_621_to_fp16)[name = tensor<string, []>("aw_chunk_109_cast")];
            tensor<string, []> var_624_equation_0 = const()[name = tensor<string, []>("op_624_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_624_cast = einsum(equation = var_624_equation_0, values = (var_366_cast, var_329_cast))[name = tensor<string, []>("op_624_cast")];
            tensor<fp16, []> var_625_to_fp16 = const()[name = tensor<string, []>("op_625_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_111_cast = mul(x = var_624_cast, y = var_625_to_fp16)[name = tensor<string, []>("aw_chunk_111_cast")];
            tensor<string, []> var_628_equation_0 = const()[name = tensor<string, []>("op_628_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_628_cast = einsum(equation = var_628_equation_0, values = (var_370_cast, var_330_cast))[name = tensor<string, []>("op_628_cast")];
            tensor<fp16, []> var_629_to_fp16 = const()[name = tensor<string, []>("op_629_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_113_cast = mul(x = var_628_cast, y = var_629_to_fp16)[name = tensor<string, []>("aw_chunk_113_cast")];
            tensor<string, []> var_632_equation_0 = const()[name = tensor<string, []>("op_632_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_632_cast = einsum(equation = var_632_equation_0, values = (var_370_cast, var_331_cast))[name = tensor<string, []>("op_632_cast")];
            tensor<fp16, []> var_633_to_fp16 = const()[name = tensor<string, []>("op_633_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_115_cast = mul(x = var_632_cast, y = var_633_to_fp16)[name = tensor<string, []>("aw_chunk_115_cast")];
            tensor<string, []> var_636_equation_0 = const()[name = tensor<string, []>("op_636_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_636_cast = einsum(equation = var_636_equation_0, values = (var_370_cast, var_332_cast))[name = tensor<string, []>("op_636_cast")];
            tensor<fp16, []> var_637_to_fp16 = const()[name = tensor<string, []>("op_637_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_117_cast = mul(x = var_636_cast, y = var_637_to_fp16)[name = tensor<string, []>("aw_chunk_117_cast")];
            tensor<string, []> var_640_equation_0 = const()[name = tensor<string, []>("op_640_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_640_cast = einsum(equation = var_640_equation_0, values = (var_370_cast, var_333_cast))[name = tensor<string, []>("op_640_cast")];
            tensor<fp16, []> var_641_to_fp16 = const()[name = tensor<string, []>("op_641_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_119_cast = mul(x = var_640_cast, y = var_641_to_fp16)[name = tensor<string, []>("aw_chunk_119_cast")];
            tensor<string, []> var_644_equation_0 = const()[name = tensor<string, []>("op_644_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_644_cast = einsum(equation = var_644_equation_0, values = (var_370_cast, var_334_cast))[name = tensor<string, []>("op_644_cast")];
            tensor<fp16, []> var_645_to_fp16 = const()[name = tensor<string, []>("op_645_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_121_cast = mul(x = var_644_cast, y = var_645_to_fp16)[name = tensor<string, []>("aw_chunk_121_cast")];
            tensor<string, []> var_648_equation_0 = const()[name = tensor<string, []>("op_648_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_648_cast = einsum(equation = var_648_equation_0, values = (var_370_cast, var_335_cast))[name = tensor<string, []>("op_648_cast")];
            tensor<fp16, []> var_649_to_fp16 = const()[name = tensor<string, []>("op_649_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_123_cast = mul(x = var_648_cast, y = var_649_to_fp16)[name = tensor<string, []>("aw_chunk_123_cast")];
            tensor<string, []> var_652_equation_0 = const()[name = tensor<string, []>("op_652_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_652_cast = einsum(equation = var_652_equation_0, values = (var_370_cast, var_336_cast))[name = tensor<string, []>("op_652_cast")];
            tensor<fp16, []> var_653_to_fp16 = const()[name = tensor<string, []>("op_653_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_125_cast = mul(x = var_652_cast, y = var_653_to_fp16)[name = tensor<string, []>("aw_chunk_125_cast")];
            tensor<string, []> var_656_equation_0 = const()[name = tensor<string, []>("op_656_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_656_cast = einsum(equation = var_656_equation_0, values = (var_370_cast, var_337_cast))[name = tensor<string, []>("op_656_cast")];
            tensor<fp16, []> var_657_to_fp16 = const()[name = tensor<string, []>("op_657_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_127_cast = mul(x = var_656_cast, y = var_657_to_fp16)[name = tensor<string, []>("aw_chunk_127_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_659_cast = softmax(axis = var_123, x = aw_chunk_1_cast)[name = tensor<string, []>("op_659_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_660_cast = softmax(axis = var_123, x = aw_chunk_3_cast)[name = tensor<string, []>("op_660_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_661_cast = softmax(axis = var_123, x = aw_chunk_5_cast)[name = tensor<string, []>("op_661_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_662_cast = softmax(axis = var_123, x = aw_chunk_7_cast)[name = tensor<string, []>("op_662_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_663_cast = softmax(axis = var_123, x = aw_chunk_9_cast)[name = tensor<string, []>("op_663_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_664_cast = softmax(axis = var_123, x = aw_chunk_11_cast)[name = tensor<string, []>("op_664_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_665_cast = softmax(axis = var_123, x = aw_chunk_13_cast)[name = tensor<string, []>("op_665_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_666_cast = softmax(axis = var_123, x = aw_chunk_15_cast)[name = tensor<string, []>("op_666_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_667_cast = softmax(axis = var_123, x = aw_chunk_17_cast)[name = tensor<string, []>("op_667_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_668_cast = softmax(axis = var_123, x = aw_chunk_19_cast)[name = tensor<string, []>("op_668_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_669_cast = softmax(axis = var_123, x = aw_chunk_21_cast)[name = tensor<string, []>("op_669_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_670_cast = softmax(axis = var_123, x = aw_chunk_23_cast)[name = tensor<string, []>("op_670_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_671_cast = softmax(axis = var_123, x = aw_chunk_25_cast)[name = tensor<string, []>("op_671_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_672_cast = softmax(axis = var_123, x = aw_chunk_27_cast)[name = tensor<string, []>("op_672_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_673_cast = softmax(axis = var_123, x = aw_chunk_29_cast)[name = tensor<string, []>("op_673_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_674_cast = softmax(axis = var_123, x = aw_chunk_31_cast)[name = tensor<string, []>("op_674_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_675_cast = softmax(axis = var_123, x = aw_chunk_33_cast)[name = tensor<string, []>("op_675_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_676_cast = softmax(axis = var_123, x = aw_chunk_35_cast)[name = tensor<string, []>("op_676_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_677_cast = softmax(axis = var_123, x = aw_chunk_37_cast)[name = tensor<string, []>("op_677_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_678_cast = softmax(axis = var_123, x = aw_chunk_39_cast)[name = tensor<string, []>("op_678_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_679_cast = softmax(axis = var_123, x = aw_chunk_41_cast)[name = tensor<string, []>("op_679_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_680_cast = softmax(axis = var_123, x = aw_chunk_43_cast)[name = tensor<string, []>("op_680_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_681_cast = softmax(axis = var_123, x = aw_chunk_45_cast)[name = tensor<string, []>("op_681_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_682_cast = softmax(axis = var_123, x = aw_chunk_47_cast)[name = tensor<string, []>("op_682_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_683_cast = softmax(axis = var_123, x = aw_chunk_49_cast)[name = tensor<string, []>("op_683_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_684_cast = softmax(axis = var_123, x = aw_chunk_51_cast)[name = tensor<string, []>("op_684_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_685_cast = softmax(axis = var_123, x = aw_chunk_53_cast)[name = tensor<string, []>("op_685_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_686_cast = softmax(axis = var_123, x = aw_chunk_55_cast)[name = tensor<string, []>("op_686_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_687_cast = softmax(axis = var_123, x = aw_chunk_57_cast)[name = tensor<string, []>("op_687_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_688_cast = softmax(axis = var_123, x = aw_chunk_59_cast)[name = tensor<string, []>("op_688_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_689_cast = softmax(axis = var_123, x = aw_chunk_61_cast)[name = tensor<string, []>("op_689_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_690_cast = softmax(axis = var_123, x = aw_chunk_63_cast)[name = tensor<string, []>("op_690_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_691_cast = softmax(axis = var_123, x = aw_chunk_65_cast)[name = tensor<string, []>("op_691_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_692_cast = softmax(axis = var_123, x = aw_chunk_67_cast)[name = tensor<string, []>("op_692_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_693_cast = softmax(axis = var_123, x = aw_chunk_69_cast)[name = tensor<string, []>("op_693_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_694_cast = softmax(axis = var_123, x = aw_chunk_71_cast)[name = tensor<string, []>("op_694_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_695_cast = softmax(axis = var_123, x = aw_chunk_73_cast)[name = tensor<string, []>("op_695_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_696_cast = softmax(axis = var_123, x = aw_chunk_75_cast)[name = tensor<string, []>("op_696_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_697_cast = softmax(axis = var_123, x = aw_chunk_77_cast)[name = tensor<string, []>("op_697_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_698_cast = softmax(axis = var_123, x = aw_chunk_79_cast)[name = tensor<string, []>("op_698_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_699_cast = softmax(axis = var_123, x = aw_chunk_81_cast)[name = tensor<string, []>("op_699_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_700_cast = softmax(axis = var_123, x = aw_chunk_83_cast)[name = tensor<string, []>("op_700_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_701_cast = softmax(axis = var_123, x = aw_chunk_85_cast)[name = tensor<string, []>("op_701_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_702_cast = softmax(axis = var_123, x = aw_chunk_87_cast)[name = tensor<string, []>("op_702_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_703_cast = softmax(axis = var_123, x = aw_chunk_89_cast)[name = tensor<string, []>("op_703_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_704_cast = softmax(axis = var_123, x = aw_chunk_91_cast)[name = tensor<string, []>("op_704_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_705_cast = softmax(axis = var_123, x = aw_chunk_93_cast)[name = tensor<string, []>("op_705_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_706_cast = softmax(axis = var_123, x = aw_chunk_95_cast)[name = tensor<string, []>("op_706_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_707_cast = softmax(axis = var_123, x = aw_chunk_97_cast)[name = tensor<string, []>("op_707_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_708_cast = softmax(axis = var_123, x = aw_chunk_99_cast)[name = tensor<string, []>("op_708_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_709_cast = softmax(axis = var_123, x = aw_chunk_101_cast)[name = tensor<string, []>("op_709_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_710_cast = softmax(axis = var_123, x = aw_chunk_103_cast)[name = tensor<string, []>("op_710_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_711_cast = softmax(axis = var_123, x = aw_chunk_105_cast)[name = tensor<string, []>("op_711_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_712_cast = softmax(axis = var_123, x = aw_chunk_107_cast)[name = tensor<string, []>("op_712_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_713_cast = softmax(axis = var_123, x = aw_chunk_109_cast)[name = tensor<string, []>("op_713_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_714_cast = softmax(axis = var_123, x = aw_chunk_111_cast)[name = tensor<string, []>("op_714_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_715_cast = softmax(axis = var_123, x = aw_chunk_113_cast)[name = tensor<string, []>("op_715_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_716_cast = softmax(axis = var_123, x = aw_chunk_115_cast)[name = tensor<string, []>("op_716_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_717_cast = softmax(axis = var_123, x = aw_chunk_117_cast)[name = tensor<string, []>("op_717_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_718_cast = softmax(axis = var_123, x = aw_chunk_119_cast)[name = tensor<string, []>("op_718_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_719_cast = softmax(axis = var_123, x = aw_chunk_121_cast)[name = tensor<string, []>("op_719_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_720_cast = softmax(axis = var_123, x = aw_chunk_123_cast)[name = tensor<string, []>("op_720_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_721_cast = softmax(axis = var_123, x = aw_chunk_125_cast)[name = tensor<string, []>("op_721_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_722_cast = softmax(axis = var_123, x = aw_chunk_127_cast)[name = tensor<string, []>("op_722_cast")];
            tensor<string, []> var_724_equation_0 = const()[name = tensor<string, []>("op_724_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_724_cast = einsum(equation = var_724_equation_0, values = (var_372_cast, var_659_cast))[name = tensor<string, []>("op_724_cast")];
            tensor<string, []> var_726_equation_0 = const()[name = tensor<string, []>("op_726_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_726_cast = einsum(equation = var_726_equation_0, values = (var_372_cast, var_660_cast))[name = tensor<string, []>("op_726_cast")];
            tensor<string, []> var_728_equation_0 = const()[name = tensor<string, []>("op_728_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_728_cast = einsum(equation = var_728_equation_0, values = (var_372_cast, var_661_cast))[name = tensor<string, []>("op_728_cast")];
            tensor<string, []> var_730_equation_0 = const()[name = tensor<string, []>("op_730_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_730_cast = einsum(equation = var_730_equation_0, values = (var_372_cast, var_662_cast))[name = tensor<string, []>("op_730_cast")];
            tensor<string, []> var_732_equation_0 = const()[name = tensor<string, []>("op_732_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_732_cast = einsum(equation = var_732_equation_0, values = (var_372_cast, var_663_cast))[name = tensor<string, []>("op_732_cast")];
            tensor<string, []> var_734_equation_0 = const()[name = tensor<string, []>("op_734_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_734_cast = einsum(equation = var_734_equation_0, values = (var_372_cast, var_664_cast))[name = tensor<string, []>("op_734_cast")];
            tensor<string, []> var_736_equation_0 = const()[name = tensor<string, []>("op_736_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_736_cast = einsum(equation = var_736_equation_0, values = (var_372_cast, var_665_cast))[name = tensor<string, []>("op_736_cast")];
            tensor<string, []> var_738_equation_0 = const()[name = tensor<string, []>("op_738_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_738_cast = einsum(equation = var_738_equation_0, values = (var_372_cast, var_666_cast))[name = tensor<string, []>("op_738_cast")];
            tensor<string, []> var_740_equation_0 = const()[name = tensor<string, []>("op_740_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_740_cast = einsum(equation = var_740_equation_0, values = (var_376_cast, var_667_cast))[name = tensor<string, []>("op_740_cast")];
            tensor<string, []> var_742_equation_0 = const()[name = tensor<string, []>("op_742_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_742_cast = einsum(equation = var_742_equation_0, values = (var_376_cast, var_668_cast))[name = tensor<string, []>("op_742_cast")];
            tensor<string, []> var_744_equation_0 = const()[name = tensor<string, []>("op_744_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_744_cast = einsum(equation = var_744_equation_0, values = (var_376_cast, var_669_cast))[name = tensor<string, []>("op_744_cast")];
            tensor<string, []> var_746_equation_0 = const()[name = tensor<string, []>("op_746_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_746_cast = einsum(equation = var_746_equation_0, values = (var_376_cast, var_670_cast))[name = tensor<string, []>("op_746_cast")];
            tensor<string, []> var_748_equation_0 = const()[name = tensor<string, []>("op_748_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_748_cast = einsum(equation = var_748_equation_0, values = (var_376_cast, var_671_cast))[name = tensor<string, []>("op_748_cast")];
            tensor<string, []> var_750_equation_0 = const()[name = tensor<string, []>("op_750_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_750_cast = einsum(equation = var_750_equation_0, values = (var_376_cast, var_672_cast))[name = tensor<string, []>("op_750_cast")];
            tensor<string, []> var_752_equation_0 = const()[name = tensor<string, []>("op_752_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_752_cast = einsum(equation = var_752_equation_0, values = (var_376_cast, var_673_cast))[name = tensor<string, []>("op_752_cast")];
            tensor<string, []> var_754_equation_0 = const()[name = tensor<string, []>("op_754_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_754_cast = einsum(equation = var_754_equation_0, values = (var_376_cast, var_674_cast))[name = tensor<string, []>("op_754_cast")];
            tensor<string, []> var_756_equation_0 = const()[name = tensor<string, []>("op_756_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_756_cast = einsum(equation = var_756_equation_0, values = (var_380_cast, var_675_cast))[name = tensor<string, []>("op_756_cast")];
            tensor<string, []> var_758_equation_0 = const()[name = tensor<string, []>("op_758_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_758_cast = einsum(equation = var_758_equation_0, values = (var_380_cast, var_676_cast))[name = tensor<string, []>("op_758_cast")];
            tensor<string, []> var_760_equation_0 = const()[name = tensor<string, []>("op_760_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_760_cast = einsum(equation = var_760_equation_0, values = (var_380_cast, var_677_cast))[name = tensor<string, []>("op_760_cast")];
            tensor<string, []> var_762_equation_0 = const()[name = tensor<string, []>("op_762_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_762_cast = einsum(equation = var_762_equation_0, values = (var_380_cast, var_678_cast))[name = tensor<string, []>("op_762_cast")];
            tensor<string, []> var_764_equation_0 = const()[name = tensor<string, []>("op_764_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_764_cast = einsum(equation = var_764_equation_0, values = (var_380_cast, var_679_cast))[name = tensor<string, []>("op_764_cast")];
            tensor<string, []> var_766_equation_0 = const()[name = tensor<string, []>("op_766_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_766_cast = einsum(equation = var_766_equation_0, values = (var_380_cast, var_680_cast))[name = tensor<string, []>("op_766_cast")];
            tensor<string, []> var_768_equation_0 = const()[name = tensor<string, []>("op_768_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_768_cast = einsum(equation = var_768_equation_0, values = (var_380_cast, var_681_cast))[name = tensor<string, []>("op_768_cast")];
            tensor<string, []> var_770_equation_0 = const()[name = tensor<string, []>("op_770_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_770_cast = einsum(equation = var_770_equation_0, values = (var_380_cast, var_682_cast))[name = tensor<string, []>("op_770_cast")];
            tensor<string, []> var_772_equation_0 = const()[name = tensor<string, []>("op_772_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_772_cast = einsum(equation = var_772_equation_0, values = (var_384_cast, var_683_cast))[name = tensor<string, []>("op_772_cast")];
            tensor<string, []> var_774_equation_0 = const()[name = tensor<string, []>("op_774_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_774_cast = einsum(equation = var_774_equation_0, values = (var_384_cast, var_684_cast))[name = tensor<string, []>("op_774_cast")];
            tensor<string, []> var_776_equation_0 = const()[name = tensor<string, []>("op_776_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_776_cast = einsum(equation = var_776_equation_0, values = (var_384_cast, var_685_cast))[name = tensor<string, []>("op_776_cast")];
            tensor<string, []> var_778_equation_0 = const()[name = tensor<string, []>("op_778_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_778_cast = einsum(equation = var_778_equation_0, values = (var_384_cast, var_686_cast))[name = tensor<string, []>("op_778_cast")];
            tensor<string, []> var_780_equation_0 = const()[name = tensor<string, []>("op_780_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_780_cast = einsum(equation = var_780_equation_0, values = (var_384_cast, var_687_cast))[name = tensor<string, []>("op_780_cast")];
            tensor<string, []> var_782_equation_0 = const()[name = tensor<string, []>("op_782_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_782_cast = einsum(equation = var_782_equation_0, values = (var_384_cast, var_688_cast))[name = tensor<string, []>("op_782_cast")];
            tensor<string, []> var_784_equation_0 = const()[name = tensor<string, []>("op_784_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_784_cast = einsum(equation = var_784_equation_0, values = (var_384_cast, var_689_cast))[name = tensor<string, []>("op_784_cast")];
            tensor<string, []> var_786_equation_0 = const()[name = tensor<string, []>("op_786_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_786_cast = einsum(equation = var_786_equation_0, values = (var_384_cast, var_690_cast))[name = tensor<string, []>("op_786_cast")];
            tensor<string, []> var_788_equation_0 = const()[name = tensor<string, []>("op_788_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_788_cast = einsum(equation = var_788_equation_0, values = (var_388_cast, var_691_cast))[name = tensor<string, []>("op_788_cast")];
            tensor<string, []> var_790_equation_0 = const()[name = tensor<string, []>("op_790_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_790_cast = einsum(equation = var_790_equation_0, values = (var_388_cast, var_692_cast))[name = tensor<string, []>("op_790_cast")];
            tensor<string, []> var_792_equation_0 = const()[name = tensor<string, []>("op_792_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_792_cast = einsum(equation = var_792_equation_0, values = (var_388_cast, var_693_cast))[name = tensor<string, []>("op_792_cast")];
            tensor<string, []> var_794_equation_0 = const()[name = tensor<string, []>("op_794_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_794_cast = einsum(equation = var_794_equation_0, values = (var_388_cast, var_694_cast))[name = tensor<string, []>("op_794_cast")];
            tensor<string, []> var_796_equation_0 = const()[name = tensor<string, []>("op_796_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_796_cast = einsum(equation = var_796_equation_0, values = (var_388_cast, var_695_cast))[name = tensor<string, []>("op_796_cast")];
            tensor<string, []> var_798_equation_0 = const()[name = tensor<string, []>("op_798_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_798_cast = einsum(equation = var_798_equation_0, values = (var_388_cast, var_696_cast))[name = tensor<string, []>("op_798_cast")];
            tensor<string, []> var_800_equation_0 = const()[name = tensor<string, []>("op_800_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_800_cast = einsum(equation = var_800_equation_0, values = (var_388_cast, var_697_cast))[name = tensor<string, []>("op_800_cast")];
            tensor<string, []> var_802_equation_0 = const()[name = tensor<string, []>("op_802_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_802_cast = einsum(equation = var_802_equation_0, values = (var_388_cast, var_698_cast))[name = tensor<string, []>("op_802_cast")];
            tensor<string, []> var_804_equation_0 = const()[name = tensor<string, []>("op_804_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_804_cast = einsum(equation = var_804_equation_0, values = (var_392_cast, var_699_cast))[name = tensor<string, []>("op_804_cast")];
            tensor<string, []> var_806_equation_0 = const()[name = tensor<string, []>("op_806_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_806_cast = einsum(equation = var_806_equation_0, values = (var_392_cast, var_700_cast))[name = tensor<string, []>("op_806_cast")];
            tensor<string, []> var_808_equation_0 = const()[name = tensor<string, []>("op_808_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_808_cast = einsum(equation = var_808_equation_0, values = (var_392_cast, var_701_cast))[name = tensor<string, []>("op_808_cast")];
            tensor<string, []> var_810_equation_0 = const()[name = tensor<string, []>("op_810_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_810_cast = einsum(equation = var_810_equation_0, values = (var_392_cast, var_702_cast))[name = tensor<string, []>("op_810_cast")];
            tensor<string, []> var_812_equation_0 = const()[name = tensor<string, []>("op_812_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_812_cast = einsum(equation = var_812_equation_0, values = (var_392_cast, var_703_cast))[name = tensor<string, []>("op_812_cast")];
            tensor<string, []> var_814_equation_0 = const()[name = tensor<string, []>("op_814_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_814_cast = einsum(equation = var_814_equation_0, values = (var_392_cast, var_704_cast))[name = tensor<string, []>("op_814_cast")];
            tensor<string, []> var_816_equation_0 = const()[name = tensor<string, []>("op_816_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_816_cast = einsum(equation = var_816_equation_0, values = (var_392_cast, var_705_cast))[name = tensor<string, []>("op_816_cast")];
            tensor<string, []> var_818_equation_0 = const()[name = tensor<string, []>("op_818_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_818_cast = einsum(equation = var_818_equation_0, values = (var_392_cast, var_706_cast))[name = tensor<string, []>("op_818_cast")];
            tensor<string, []> var_820_equation_0 = const()[name = tensor<string, []>("op_820_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_820_cast = einsum(equation = var_820_equation_0, values = (var_396_cast, var_707_cast))[name = tensor<string, []>("op_820_cast")];
            tensor<string, []> var_822_equation_0 = const()[name = tensor<string, []>("op_822_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_822_cast = einsum(equation = var_822_equation_0, values = (var_396_cast, var_708_cast))[name = tensor<string, []>("op_822_cast")];
            tensor<string, []> var_824_equation_0 = const()[name = tensor<string, []>("op_824_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_824_cast = einsum(equation = var_824_equation_0, values = (var_396_cast, var_709_cast))[name = tensor<string, []>("op_824_cast")];
            tensor<string, []> var_826_equation_0 = const()[name = tensor<string, []>("op_826_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_826_cast = einsum(equation = var_826_equation_0, values = (var_396_cast, var_710_cast))[name = tensor<string, []>("op_826_cast")];
            tensor<string, []> var_828_equation_0 = const()[name = tensor<string, []>("op_828_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_828_cast = einsum(equation = var_828_equation_0, values = (var_396_cast, var_711_cast))[name = tensor<string, []>("op_828_cast")];
            tensor<string, []> var_830_equation_0 = const()[name = tensor<string, []>("op_830_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_830_cast = einsum(equation = var_830_equation_0, values = (var_396_cast, var_712_cast))[name = tensor<string, []>("op_830_cast")];
            tensor<string, []> var_832_equation_0 = const()[name = tensor<string, []>("op_832_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_832_cast = einsum(equation = var_832_equation_0, values = (var_396_cast, var_713_cast))[name = tensor<string, []>("op_832_cast")];
            tensor<string, []> var_834_equation_0 = const()[name = tensor<string, []>("op_834_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_834_cast = einsum(equation = var_834_equation_0, values = (var_396_cast, var_714_cast))[name = tensor<string, []>("op_834_cast")];
            tensor<string, []> var_836_equation_0 = const()[name = tensor<string, []>("op_836_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_836_cast = einsum(equation = var_836_equation_0, values = (var_400_cast, var_715_cast))[name = tensor<string, []>("op_836_cast")];
            tensor<string, []> var_838_equation_0 = const()[name = tensor<string, []>("op_838_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_838_cast = einsum(equation = var_838_equation_0, values = (var_400_cast, var_716_cast))[name = tensor<string, []>("op_838_cast")];
            tensor<string, []> var_840_equation_0 = const()[name = tensor<string, []>("op_840_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_840_cast = einsum(equation = var_840_equation_0, values = (var_400_cast, var_717_cast))[name = tensor<string, []>("op_840_cast")];
            tensor<string, []> var_842_equation_0 = const()[name = tensor<string, []>("op_842_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_842_cast = einsum(equation = var_842_equation_0, values = (var_400_cast, var_718_cast))[name = tensor<string, []>("op_842_cast")];
            tensor<string, []> var_844_equation_0 = const()[name = tensor<string, []>("op_844_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_844_cast = einsum(equation = var_844_equation_0, values = (var_400_cast, var_719_cast))[name = tensor<string, []>("op_844_cast")];
            tensor<string, []> var_846_equation_0 = const()[name = tensor<string, []>("op_846_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_846_cast = einsum(equation = var_846_equation_0, values = (var_400_cast, var_720_cast))[name = tensor<string, []>("op_846_cast")];
            tensor<string, []> var_848_equation_0 = const()[name = tensor<string, []>("op_848_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_848_cast = einsum(equation = var_848_equation_0, values = (var_400_cast, var_721_cast))[name = tensor<string, []>("op_848_cast")];
            tensor<string, []> var_850_equation_0 = const()[name = tensor<string, []>("op_850_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_850_cast = einsum(equation = var_850_equation_0, values = (var_400_cast, var_722_cast))[name = tensor<string, []>("op_850_cast")];
            tensor<bool, []> var_852_interleave_0 = const()[name = tensor<string, []>("op_852_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_852_cast = concat(axis = var_95, interleave = var_852_interleave_0, values = (var_724_cast, var_726_cast, var_728_cast, var_730_cast, var_732_cast, var_734_cast, var_736_cast, var_738_cast))[name = tensor<string, []>("op_852_cast")];
            tensor<bool, []> var_854_interleave_0 = const()[name = tensor<string, []>("op_854_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_854_cast = concat(axis = var_95, interleave = var_854_interleave_0, values = (var_740_cast, var_742_cast, var_744_cast, var_746_cast, var_748_cast, var_750_cast, var_752_cast, var_754_cast))[name = tensor<string, []>("op_854_cast")];
            tensor<bool, []> var_856_interleave_0 = const()[name = tensor<string, []>("op_856_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_856_cast = concat(axis = var_95, interleave = var_856_interleave_0, values = (var_756_cast, var_758_cast, var_760_cast, var_762_cast, var_764_cast, var_766_cast, var_768_cast, var_770_cast))[name = tensor<string, []>("op_856_cast")];
            tensor<bool, []> var_858_interleave_0 = const()[name = tensor<string, []>("op_858_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_858_cast = concat(axis = var_95, interleave = var_858_interleave_0, values = (var_772_cast, var_774_cast, var_776_cast, var_778_cast, var_780_cast, var_782_cast, var_784_cast, var_786_cast))[name = tensor<string, []>("op_858_cast")];
            tensor<bool, []> var_860_interleave_0 = const()[name = tensor<string, []>("op_860_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_860_cast = concat(axis = var_95, interleave = var_860_interleave_0, values = (var_788_cast, var_790_cast, var_792_cast, var_794_cast, var_796_cast, var_798_cast, var_800_cast, var_802_cast))[name = tensor<string, []>("op_860_cast")];
            tensor<bool, []> var_862_interleave_0 = const()[name = tensor<string, []>("op_862_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_862_cast = concat(axis = var_95, interleave = var_862_interleave_0, values = (var_804_cast, var_806_cast, var_808_cast, var_810_cast, var_812_cast, var_814_cast, var_816_cast, var_818_cast))[name = tensor<string, []>("op_862_cast")];
            tensor<bool, []> var_864_interleave_0 = const()[name = tensor<string, []>("op_864_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_864_cast = concat(axis = var_95, interleave = var_864_interleave_0, values = (var_820_cast, var_822_cast, var_824_cast, var_826_cast, var_828_cast, var_830_cast, var_832_cast, var_834_cast))[name = tensor<string, []>("op_864_cast")];
            tensor<bool, []> var_866_interleave_0 = const()[name = tensor<string, []>("op_866_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_866_cast = concat(axis = var_95, interleave = var_866_interleave_0, values = (var_836_cast, var_838_cast, var_840_cast, var_842_cast, var_844_cast, var_846_cast, var_848_cast, var_850_cast))[name = tensor<string, []>("op_866_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_123, interleave = input_25_interleave_0, values = (var_852_cast, var_854_cast, var_856_cast, var_858_cast, var_860_cast, var_862_cast, var_864_cast, var_866_cast))[name = tensor<string, []>("input_25_cast")];
            tensor<int32, [2]> var_872 = const()[name = tensor<string, []>("op_872"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_874 = const()[name = tensor<string, []>("op_874"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_876_pad_type_0 = const()[name = tensor<string, []>("op_876_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_876_pad_0 = const()[name = tensor<string, []>("op_876_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3560256))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3637120))), name = tensor<string, []>("down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            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, []>(3637312)))];
            tensor<fp16, [2, 320, 1, 4096]> var_876_cast = conv(bias = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_874, groups = var_123, pad = var_876_pad_0, pad_type = var_876_pad_type_0, strides = var_872, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_25_cast)[name = tensor<string, []>("op_876_cast")];
            tensor<fp16, [2, 320, 1, 4096]> inputs_3_cast = add(x = var_876_cast, y = inputs_1_cast)[name = tensor<string, []>("inputs_3_cast")];
            tensor<int32, [1]> var_880 = const()[name = tensor<string, []>("op_880"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_3_cast = reduce_mean(axes = var_880, keep_dims = var_118, x = inputs_3_cast)[name = tensor<string, []>("channels_mean_3_cast")];
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_3_cast = sub(x = inputs_3_cast, y = channels_mean_3_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_3_cast")];
            tensor<int32, [1]> var_884 = const()[name = tensor<string, []>("op_884"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_885_cast = reduce_mean(axes = var_884, keep_dims = var_118, x = zero_mean_sq_3_cast)[name = tensor<string, []>("op_885_cast")];
            tensor<fp16, []> var_886_to_fp16 = const()[name = tensor<string, []>("op_886_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_887_cast = add(x = var_885_cast, y = var_886_to_fp16)[name = tensor<string, []>("op_887_cast")];
            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_887_cast)[name = tensor<string, []>("denom_3_cast")];
            tensor<fp16, [2, 320, 1, 4096]> out_3_cast = mul(x = zero_mean_3_cast, y = denom_3_cast)[name = tensor<string, []>("out_3_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_891_to_fp16 = const()[name = tensor<string, []>("op_891_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3638016)))];
            tensor<fp16, [2, 320, 1, 4096]> var_892_cast = add(x = out_3_cast, y = var_891_to_fp16)[name = tensor<string, []>("op_892_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_894_to_fp16 = const()[name = tensor<string, []>("op_894_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3638720)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_11_cast = mul(x = var_892_cast, y = var_894_to_fp16)[name = tensor<string, []>("hidden_states_11_cast")];
            tensor<int32, [2]> var_901 = const()[name = tensor<string, []>("op_901"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_903 = const()[name = tensor<string, []>("op_903"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3639424))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3716288))), name = tensor<string, []>("down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            tensor<fp16, [2, 320, 1, 4096]> q_3_cast = conv(dilations = var_903, groups = var_123, pad = q_3_pad_0, pad_type = q_3_pad_type_0, strides = var_901, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_11_cast)[name = tensor<string, []>("q_3_cast")];
            tensor<int32, [2]> var_907 = const()[name = tensor<string, []>("op_907"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_909 = const()[name = tensor<string, []>("op_909"), 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, 768, 1, 1]> down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [184320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3716480))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3900864))), name = tensor<string, []>("down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 768, 1, 1])];
            tensor<fp16, [2, 320, 1, 77]> k_5_cast = conv(dilations = var_909, groups = var_123, pad = k_5_pad_0, pad_type = k_5_pad_type_0, strides = var_907, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_5_cast")];
            tensor<int32, [2]> var_913 = const()[name = tensor<string, []>("op_913"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_915 = const()[name = tensor<string, []>("op_915"), 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, 768, 1, 1]> down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [184320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3901056))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4085440))), name = tensor<string, []>("down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 768, 1, 1])];
            tensor<fp16, [2, 320, 1, 77]> v_3_cast = conv(dilations = var_915, groups = var_123, pad = v_3_pad_0, pad_type = v_3_pad_type_0, strides = var_913, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_3_cast")];
            tensor<int32, [4]> var_919_begin_0 = const()[name = tensor<string, []>("op_919_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_919_end_0 = const()[name = tensor<string, []>("op_919_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_919_end_mask_0 = const()[name = tensor<string, []>("op_919_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_919_cast = slice_by_index(begin = var_919_begin_0, end = var_919_end_0, end_mask = var_919_end_mask_0, x = q_3_cast)[name = tensor<string, []>("op_919_cast")];
            tensor<int32, [4]> var_923_begin_0 = const()[name = tensor<string, []>("op_923_begin_0"), val = tensor<int32, [4]>([0, 40, 0, 0])];
            tensor<int32, [4]> var_923_end_0 = const()[name = tensor<string, []>("op_923_end_0"), val = tensor<int32, [4]>([2, 80, 1, 4096])];
            tensor<bool, [4]> var_923_end_mask_0 = const()[name = tensor<string, []>("op_923_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_923_cast = slice_by_index(begin = var_923_begin_0, end = var_923_end_0, end_mask = var_923_end_mask_0, x = q_3_cast)[name = tensor<string, []>("op_923_cast")];
            tensor<int32, [4]> var_927_begin_0 = const()[name = tensor<string, []>("op_927_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_927_end_0 = const()[name = tensor<string, []>("op_927_end_0"), val = tensor<int32, [4]>([2, 120, 1, 4096])];
            tensor<bool, [4]> var_927_end_mask_0 = const()[name = tensor<string, []>("op_927_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_927_cast = slice_by_index(begin = var_927_begin_0, end = var_927_end_0, end_mask = var_927_end_mask_0, x = q_3_cast)[name = tensor<string, []>("op_927_cast")];
            tensor<int32, [4]> var_931_begin_0 = const()[name = tensor<string, []>("op_931_begin_0"), val = tensor<int32, [4]>([0, 120, 0, 0])];
            tensor<int32, [4]> var_931_end_0 = const()[name = tensor<string, []>("op_931_end_0"), val = tensor<int32, [4]>([2, 160, 1, 4096])];
            tensor<bool, [4]> var_931_end_mask_0 = const()[name = tensor<string, []>("op_931_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_931_cast = slice_by_index(begin = var_931_begin_0, end = var_931_end_0, end_mask = var_931_end_mask_0, x = q_3_cast)[name = tensor<string, []>("op_931_cast")];
            tensor<int32, [4]> var_935_begin_0 = const()[name = tensor<string, []>("op_935_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_935_end_0 = const()[name = tensor<string, []>("op_935_end_0"), val = tensor<int32, [4]>([2, 200, 1, 4096])];
            tensor<bool, [4]> var_935_end_mask_0 = const()[name = tensor<string, []>("op_935_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_935_cast = slice_by_index(begin = var_935_begin_0, end = var_935_end_0, end_mask = var_935_end_mask_0, x = q_3_cast)[name = tensor<string, []>("op_935_cast")];
            tensor<int32, [4]> var_939_begin_0 = const()[name = tensor<string, []>("op_939_begin_0"), val = tensor<int32, [4]>([0, 200, 0, 0])];
            tensor<int32, [4]> var_939_end_0 = const()[name = tensor<string, []>("op_939_end_0"), val = tensor<int32, [4]>([2, 240, 1, 4096])];
            tensor<bool, [4]> var_939_end_mask_0 = const()[name = tensor<string, []>("op_939_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_939_cast = slice_by_index(begin = var_939_begin_0, end = var_939_end_0, end_mask = var_939_end_mask_0, x = q_3_cast)[name = tensor<string, []>("op_939_cast")];
            tensor<int32, [4]> var_943_begin_0 = const()[name = tensor<string, []>("op_943_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_943_end_0 = const()[name = tensor<string, []>("op_943_end_0"), val = tensor<int32, [4]>([2, 280, 1, 4096])];
            tensor<bool, [4]> var_943_end_mask_0 = const()[name = tensor<string, []>("op_943_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_943_cast = slice_by_index(begin = var_943_begin_0, end = var_943_end_0, end_mask = var_943_end_mask_0, x = q_3_cast)[name = tensor<string, []>("op_943_cast")];
            tensor<int32, [4]> var_947_begin_0 = const()[name = tensor<string, []>("op_947_begin_0"), val = tensor<int32, [4]>([0, 280, 0, 0])];
            tensor<int32, [4]> var_947_end_0 = const()[name = tensor<string, []>("op_947_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_947_end_mask_0 = const()[name = tensor<string, []>("op_947_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_947_cast = slice_by_index(begin = var_947_begin_0, end = var_947_end_0, end_mask = var_947_end_mask_0, x = q_3_cast)[name = tensor<string, []>("op_947_cast")];
            tensor<int32, [4]> var_950_begin_0 = const()[name = tensor<string, []>("op_950_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_950_end_0 = const()[name = tensor<string, []>("op_950_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_950_end_mask_0 = const()[name = tensor<string, []>("op_950_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_950_cast = slice_by_index(begin = var_950_begin_0, end = var_950_end_0, end_mask = var_950_end_mask_0, x = var_919_cast)[name = tensor<string, []>("op_950_cast")];
            tensor<int32, [4]> var_951_begin_0 = const()[name = tensor<string, []>("op_951_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_951_end_0 = const()[name = tensor<string, []>("op_951_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_951_end_mask_0 = const()[name = tensor<string, []>("op_951_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_951_cast = slice_by_index(begin = var_951_begin_0, end = var_951_end_0, end_mask = var_951_end_mask_0, x = var_919_cast)[name = tensor<string, []>("op_951_cast")];
            tensor<int32, [4]> var_952_begin_0 = const()[name = tensor<string, []>("op_952_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_952_end_0 = const()[name = tensor<string, []>("op_952_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_952_end_mask_0 = const()[name = tensor<string, []>("op_952_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_952_cast = slice_by_index(begin = var_952_begin_0, end = var_952_end_0, end_mask = var_952_end_mask_0, x = var_919_cast)[name = tensor<string, []>("op_952_cast")];
            tensor<int32, [4]> var_953_begin_0 = const()[name = tensor<string, []>("op_953_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_953_end_0 = const()[name = tensor<string, []>("op_953_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_953_end_mask_0 = const()[name = tensor<string, []>("op_953_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_953_cast = slice_by_index(begin = var_953_begin_0, end = var_953_end_0, end_mask = var_953_end_mask_0, x = var_919_cast)[name = tensor<string, []>("op_953_cast")];
            tensor<int32, [4]> var_954_begin_0 = const()[name = tensor<string, []>("op_954_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_954_end_0 = const()[name = tensor<string, []>("op_954_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_954_end_mask_0 = const()[name = tensor<string, []>("op_954_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_954_cast = slice_by_index(begin = var_954_begin_0, end = var_954_end_0, end_mask = var_954_end_mask_0, x = var_919_cast)[name = tensor<string, []>("op_954_cast")];
            tensor<int32, [4]> var_955_begin_0 = const()[name = tensor<string, []>("op_955_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_955_end_0 = const()[name = tensor<string, []>("op_955_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_955_end_mask_0 = const()[name = tensor<string, []>("op_955_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_955_cast = slice_by_index(begin = var_955_begin_0, end = var_955_end_0, end_mask = var_955_end_mask_0, x = var_919_cast)[name = tensor<string, []>("op_955_cast")];
            tensor<int32, [4]> var_956_begin_0 = const()[name = tensor<string, []>("op_956_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_956_end_0 = const()[name = tensor<string, []>("op_956_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_956_end_mask_0 = const()[name = tensor<string, []>("op_956_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_956_cast = slice_by_index(begin = var_956_begin_0, end = var_956_end_0, end_mask = var_956_end_mask_0, x = var_919_cast)[name = tensor<string, []>("op_956_cast")];
            tensor<int32, [4]> var_957_begin_0 = const()[name = tensor<string, []>("op_957_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_957_end_0 = const()[name = tensor<string, []>("op_957_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_957_end_mask_0 = const()[name = tensor<string, []>("op_957_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_957_cast = slice_by_index(begin = var_957_begin_0, end = var_957_end_0, end_mask = var_957_end_mask_0, x = var_919_cast)[name = tensor<string, []>("op_957_cast")];
            tensor<int32, [4]> var_958_begin_0 = const()[name = tensor<string, []>("op_958_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_958_end_0 = const()[name = tensor<string, []>("op_958_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_958_end_mask_0 = const()[name = tensor<string, []>("op_958_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_958_cast = slice_by_index(begin = var_958_begin_0, end = var_958_end_0, end_mask = var_958_end_mask_0, x = var_923_cast)[name = tensor<string, []>("op_958_cast")];
            tensor<int32, [4]> var_959_begin_0 = const()[name = tensor<string, []>("op_959_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_959_end_0 = const()[name = tensor<string, []>("op_959_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_959_end_mask_0 = const()[name = tensor<string, []>("op_959_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_959_cast = slice_by_index(begin = var_959_begin_0, end = var_959_end_0, end_mask = var_959_end_mask_0, x = var_923_cast)[name = tensor<string, []>("op_959_cast")];
            tensor<int32, [4]> var_960_begin_0 = const()[name = tensor<string, []>("op_960_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_960_end_0 = const()[name = tensor<string, []>("op_960_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_960_end_mask_0 = const()[name = tensor<string, []>("op_960_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_960_cast = slice_by_index(begin = var_960_begin_0, end = var_960_end_0, end_mask = var_960_end_mask_0, x = var_923_cast)[name = tensor<string, []>("op_960_cast")];
            tensor<int32, [4]> var_961_begin_0 = const()[name = tensor<string, []>("op_961_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_961_end_0 = const()[name = tensor<string, []>("op_961_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_961_end_mask_0 = const()[name = tensor<string, []>("op_961_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_961_cast = slice_by_index(begin = var_961_begin_0, end = var_961_end_0, end_mask = var_961_end_mask_0, x = var_923_cast)[name = tensor<string, []>("op_961_cast")];
            tensor<int32, [4]> var_962_begin_0 = const()[name = tensor<string, []>("op_962_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_962_end_0 = const()[name = tensor<string, []>("op_962_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_962_end_mask_0 = const()[name = tensor<string, []>("op_962_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_962_cast = slice_by_index(begin = var_962_begin_0, end = var_962_end_0, end_mask = var_962_end_mask_0, x = var_923_cast)[name = tensor<string, []>("op_962_cast")];
            tensor<int32, [4]> var_963_begin_0 = const()[name = tensor<string, []>("op_963_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_963_end_0 = const()[name = tensor<string, []>("op_963_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_963_end_mask_0 = const()[name = tensor<string, []>("op_963_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_963_cast = slice_by_index(begin = var_963_begin_0, end = var_963_end_0, end_mask = var_963_end_mask_0, x = var_923_cast)[name = tensor<string, []>("op_963_cast")];
            tensor<int32, [4]> var_964_begin_0 = const()[name = tensor<string, []>("op_964_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_964_end_0 = const()[name = tensor<string, []>("op_964_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_964_end_mask_0 = const()[name = tensor<string, []>("op_964_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_964_cast = slice_by_index(begin = var_964_begin_0, end = var_964_end_0, end_mask = var_964_end_mask_0, x = var_923_cast)[name = tensor<string, []>("op_964_cast")];
            tensor<int32, [4]> var_965_begin_0 = const()[name = tensor<string, []>("op_965_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_965_end_0 = const()[name = tensor<string, []>("op_965_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_965_end_mask_0 = const()[name = tensor<string, []>("op_965_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_965_cast = slice_by_index(begin = var_965_begin_0, end = var_965_end_0, end_mask = var_965_end_mask_0, x = var_923_cast)[name = tensor<string, []>("op_965_cast")];
            tensor<int32, [4]> var_966_begin_0 = const()[name = tensor<string, []>("op_966_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_966_end_0 = const()[name = tensor<string, []>("op_966_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_966_end_mask_0 = const()[name = tensor<string, []>("op_966_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_966_cast = slice_by_index(begin = var_966_begin_0, end = var_966_end_0, end_mask = var_966_end_mask_0, x = var_927_cast)[name = tensor<string, []>("op_966_cast")];
            tensor<int32, [4]> var_967_begin_0 = const()[name = tensor<string, []>("op_967_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_967_end_0 = const()[name = tensor<string, []>("op_967_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_967_end_mask_0 = const()[name = tensor<string, []>("op_967_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_967_cast = slice_by_index(begin = var_967_begin_0, end = var_967_end_0, end_mask = var_967_end_mask_0, x = var_927_cast)[name = tensor<string, []>("op_967_cast")];
            tensor<int32, [4]> var_968_begin_0 = const()[name = tensor<string, []>("op_968_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_968_end_0 = const()[name = tensor<string, []>("op_968_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_968_end_mask_0 = const()[name = tensor<string, []>("op_968_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_968_cast = slice_by_index(begin = var_968_begin_0, end = var_968_end_0, end_mask = var_968_end_mask_0, x = var_927_cast)[name = tensor<string, []>("op_968_cast")];
            tensor<int32, [4]> var_969_begin_0 = const()[name = tensor<string, []>("op_969_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_969_end_0 = const()[name = tensor<string, []>("op_969_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_969_end_mask_0 = const()[name = tensor<string, []>("op_969_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_969_cast = slice_by_index(begin = var_969_begin_0, end = var_969_end_0, end_mask = var_969_end_mask_0, x = var_927_cast)[name = tensor<string, []>("op_969_cast")];
            tensor<int32, [4]> var_970_begin_0 = const()[name = tensor<string, []>("op_970_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_970_end_0 = const()[name = tensor<string, []>("op_970_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_970_end_mask_0 = const()[name = tensor<string, []>("op_970_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_970_cast = slice_by_index(begin = var_970_begin_0, end = var_970_end_0, end_mask = var_970_end_mask_0, x = var_927_cast)[name = tensor<string, []>("op_970_cast")];
            tensor<int32, [4]> var_971_begin_0 = const()[name = tensor<string, []>("op_971_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_971_end_0 = const()[name = tensor<string, []>("op_971_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_971_end_mask_0 = const()[name = tensor<string, []>("op_971_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_971_cast = slice_by_index(begin = var_971_begin_0, end = var_971_end_0, end_mask = var_971_end_mask_0, x = var_927_cast)[name = tensor<string, []>("op_971_cast")];
            tensor<int32, [4]> var_972_begin_0 = const()[name = tensor<string, []>("op_972_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_972_end_0 = const()[name = tensor<string, []>("op_972_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_972_end_mask_0 = const()[name = tensor<string, []>("op_972_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_972_cast = slice_by_index(begin = var_972_begin_0, end = var_972_end_0, end_mask = var_972_end_mask_0, x = var_927_cast)[name = tensor<string, []>("op_972_cast")];
            tensor<int32, [4]> var_973_begin_0 = const()[name = tensor<string, []>("op_973_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_973_end_0 = const()[name = tensor<string, []>("op_973_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_973_end_mask_0 = const()[name = tensor<string, []>("op_973_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_973_cast = slice_by_index(begin = var_973_begin_0, end = var_973_end_0, end_mask = var_973_end_mask_0, x = var_927_cast)[name = tensor<string, []>("op_973_cast")];
            tensor<int32, [4]> var_974_begin_0 = const()[name = tensor<string, []>("op_974_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_974_end_0 = const()[name = tensor<string, []>("op_974_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_974_end_mask_0 = const()[name = tensor<string, []>("op_974_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_974_cast = slice_by_index(begin = var_974_begin_0, end = var_974_end_0, end_mask = var_974_end_mask_0, x = var_931_cast)[name = tensor<string, []>("op_974_cast")];
            tensor<int32, [4]> var_975_begin_0 = const()[name = tensor<string, []>("op_975_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_975_end_0 = const()[name = tensor<string, []>("op_975_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_975_end_mask_0 = const()[name = tensor<string, []>("op_975_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_975_cast = slice_by_index(begin = var_975_begin_0, end = var_975_end_0, end_mask = var_975_end_mask_0, x = var_931_cast)[name = tensor<string, []>("op_975_cast")];
            tensor<int32, [4]> var_976_begin_0 = const()[name = tensor<string, []>("op_976_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_976_end_0 = const()[name = tensor<string, []>("op_976_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_976_end_mask_0 = const()[name = tensor<string, []>("op_976_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_976_cast = slice_by_index(begin = var_976_begin_0, end = var_976_end_0, end_mask = var_976_end_mask_0, x = var_931_cast)[name = tensor<string, []>("op_976_cast")];
            tensor<int32, [4]> var_977_begin_0 = const()[name = tensor<string, []>("op_977_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_977_end_0 = const()[name = tensor<string, []>("op_977_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_977_end_mask_0 = const()[name = tensor<string, []>("op_977_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_977_cast = slice_by_index(begin = var_977_begin_0, end = var_977_end_0, end_mask = var_977_end_mask_0, x = var_931_cast)[name = tensor<string, []>("op_977_cast")];
            tensor<int32, [4]> var_978_begin_0 = const()[name = tensor<string, []>("op_978_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_978_end_0 = const()[name = tensor<string, []>("op_978_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_978_end_mask_0 = const()[name = tensor<string, []>("op_978_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_978_cast = slice_by_index(begin = var_978_begin_0, end = var_978_end_0, end_mask = var_978_end_mask_0, x = var_931_cast)[name = tensor<string, []>("op_978_cast")];
            tensor<int32, [4]> var_979_begin_0 = const()[name = tensor<string, []>("op_979_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_979_end_0 = const()[name = tensor<string, []>("op_979_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_979_end_mask_0 = const()[name = tensor<string, []>("op_979_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_979_cast = slice_by_index(begin = var_979_begin_0, end = var_979_end_0, end_mask = var_979_end_mask_0, x = var_931_cast)[name = tensor<string, []>("op_979_cast")];
            tensor<int32, [4]> var_980_begin_0 = const()[name = tensor<string, []>("op_980_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_980_end_0 = const()[name = tensor<string, []>("op_980_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_980_end_mask_0 = const()[name = tensor<string, []>("op_980_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_980_cast = slice_by_index(begin = var_980_begin_0, end = var_980_end_0, end_mask = var_980_end_mask_0, x = var_931_cast)[name = tensor<string, []>("op_980_cast")];
            tensor<int32, [4]> var_981_begin_0 = const()[name = tensor<string, []>("op_981_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_981_end_0 = const()[name = tensor<string, []>("op_981_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_981_end_mask_0 = const()[name = tensor<string, []>("op_981_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_981_cast = slice_by_index(begin = var_981_begin_0, end = var_981_end_0, end_mask = var_981_end_mask_0, x = var_931_cast)[name = tensor<string, []>("op_981_cast")];
            tensor<int32, [4]> var_982_begin_0 = const()[name = tensor<string, []>("op_982_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_982_end_0 = const()[name = tensor<string, []>("op_982_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_982_end_mask_0 = const()[name = tensor<string, []>("op_982_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_982_cast = slice_by_index(begin = var_982_begin_0, end = var_982_end_0, end_mask = var_982_end_mask_0, x = var_935_cast)[name = tensor<string, []>("op_982_cast")];
            tensor<int32, [4]> var_983_begin_0 = const()[name = tensor<string, []>("op_983_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_983_end_0 = const()[name = tensor<string, []>("op_983_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_983_end_mask_0 = const()[name = tensor<string, []>("op_983_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_983_cast = slice_by_index(begin = var_983_begin_0, end = var_983_end_0, end_mask = var_983_end_mask_0, x = var_935_cast)[name = tensor<string, []>("op_983_cast")];
            tensor<int32, [4]> var_984_begin_0 = const()[name = tensor<string, []>("op_984_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_984_end_0 = const()[name = tensor<string, []>("op_984_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_984_end_mask_0 = const()[name = tensor<string, []>("op_984_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_984_cast = slice_by_index(begin = var_984_begin_0, end = var_984_end_0, end_mask = var_984_end_mask_0, x = var_935_cast)[name = tensor<string, []>("op_984_cast")];
            tensor<int32, [4]> var_985_begin_0 = const()[name = tensor<string, []>("op_985_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_985_end_0 = const()[name = tensor<string, []>("op_985_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_985_end_mask_0 = const()[name = tensor<string, []>("op_985_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_985_cast = slice_by_index(begin = var_985_begin_0, end = var_985_end_0, end_mask = var_985_end_mask_0, x = var_935_cast)[name = tensor<string, []>("op_985_cast")];
            tensor<int32, [4]> var_986_begin_0 = const()[name = tensor<string, []>("op_986_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_986_end_0 = const()[name = tensor<string, []>("op_986_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_986_end_mask_0 = const()[name = tensor<string, []>("op_986_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_986_cast = slice_by_index(begin = var_986_begin_0, end = var_986_end_0, end_mask = var_986_end_mask_0, x = var_935_cast)[name = tensor<string, []>("op_986_cast")];
            tensor<int32, [4]> var_987_begin_0 = const()[name = tensor<string, []>("op_987_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_987_end_0 = const()[name = tensor<string, []>("op_987_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_987_end_mask_0 = const()[name = tensor<string, []>("op_987_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_987_cast = slice_by_index(begin = var_987_begin_0, end = var_987_end_0, end_mask = var_987_end_mask_0, x = var_935_cast)[name = tensor<string, []>("op_987_cast")];
            tensor<int32, [4]> var_988_begin_0 = const()[name = tensor<string, []>("op_988_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_988_end_0 = const()[name = tensor<string, []>("op_988_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_988_end_mask_0 = const()[name = tensor<string, []>("op_988_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_988_cast = slice_by_index(begin = var_988_begin_0, end = var_988_end_0, end_mask = var_988_end_mask_0, x = var_935_cast)[name = tensor<string, []>("op_988_cast")];
            tensor<int32, [4]> var_989_begin_0 = const()[name = tensor<string, []>("op_989_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_989_end_0 = const()[name = tensor<string, []>("op_989_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_989_end_mask_0 = const()[name = tensor<string, []>("op_989_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_989_cast = slice_by_index(begin = var_989_begin_0, end = var_989_end_0, end_mask = var_989_end_mask_0, x = var_935_cast)[name = tensor<string, []>("op_989_cast")];
            tensor<int32, [4]> var_990_begin_0 = const()[name = tensor<string, []>("op_990_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_990_end_0 = const()[name = tensor<string, []>("op_990_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_990_end_mask_0 = const()[name = tensor<string, []>("op_990_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_990_cast = slice_by_index(begin = var_990_begin_0, end = var_990_end_0, end_mask = var_990_end_mask_0, x = var_939_cast)[name = tensor<string, []>("op_990_cast")];
            tensor<int32, [4]> var_991_begin_0 = const()[name = tensor<string, []>("op_991_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_991_end_0 = const()[name = tensor<string, []>("op_991_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_991_end_mask_0 = const()[name = tensor<string, []>("op_991_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_991_cast = slice_by_index(begin = var_991_begin_0, end = var_991_end_0, end_mask = var_991_end_mask_0, x = var_939_cast)[name = tensor<string, []>("op_991_cast")];
            tensor<int32, [4]> var_992_begin_0 = const()[name = tensor<string, []>("op_992_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_992_end_0 = const()[name = tensor<string, []>("op_992_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_992_end_mask_0 = const()[name = tensor<string, []>("op_992_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_992_cast = slice_by_index(begin = var_992_begin_0, end = var_992_end_0, end_mask = var_992_end_mask_0, x = var_939_cast)[name = tensor<string, []>("op_992_cast")];
            tensor<int32, [4]> var_993_begin_0 = const()[name = tensor<string, []>("op_993_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_993_end_0 = const()[name = tensor<string, []>("op_993_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_993_end_mask_0 = const()[name = tensor<string, []>("op_993_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_993_cast = slice_by_index(begin = var_993_begin_0, end = var_993_end_0, end_mask = var_993_end_mask_0, x = var_939_cast)[name = tensor<string, []>("op_993_cast")];
            tensor<int32, [4]> var_994_begin_0 = const()[name = tensor<string, []>("op_994_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_994_end_0 = const()[name = tensor<string, []>("op_994_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_994_end_mask_0 = const()[name = tensor<string, []>("op_994_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_994_cast = slice_by_index(begin = var_994_begin_0, end = var_994_end_0, end_mask = var_994_end_mask_0, x = var_939_cast)[name = tensor<string, []>("op_994_cast")];
            tensor<int32, [4]> var_995_begin_0 = const()[name = tensor<string, []>("op_995_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_995_end_0 = const()[name = tensor<string, []>("op_995_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_995_end_mask_0 = const()[name = tensor<string, []>("op_995_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_995_cast = slice_by_index(begin = var_995_begin_0, end = var_995_end_0, end_mask = var_995_end_mask_0, x = var_939_cast)[name = tensor<string, []>("op_995_cast")];
            tensor<int32, [4]> var_996_begin_0 = const()[name = tensor<string, []>("op_996_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_996_end_0 = const()[name = tensor<string, []>("op_996_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_996_end_mask_0 = const()[name = tensor<string, []>("op_996_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_996_cast = slice_by_index(begin = var_996_begin_0, end = var_996_end_0, end_mask = var_996_end_mask_0, x = var_939_cast)[name = tensor<string, []>("op_996_cast")];
            tensor<int32, [4]> var_997_begin_0 = const()[name = tensor<string, []>("op_997_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_997_end_0 = const()[name = tensor<string, []>("op_997_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_997_end_mask_0 = const()[name = tensor<string, []>("op_997_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_997_cast = slice_by_index(begin = var_997_begin_0, end = var_997_end_0, end_mask = var_997_end_mask_0, x = var_939_cast)[name = tensor<string, []>("op_997_cast")];
            tensor<int32, [4]> var_998_begin_0 = const()[name = tensor<string, []>("op_998_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_998_end_0 = const()[name = tensor<string, []>("op_998_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_998_end_mask_0 = const()[name = tensor<string, []>("op_998_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_998_cast = slice_by_index(begin = var_998_begin_0, end = var_998_end_0, end_mask = var_998_end_mask_0, x = var_943_cast)[name = tensor<string, []>("op_998_cast")];
            tensor<int32, [4]> var_999_begin_0 = const()[name = tensor<string, []>("op_999_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_999_end_0 = const()[name = tensor<string, []>("op_999_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_999_end_mask_0 = const()[name = tensor<string, []>("op_999_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_999_cast = slice_by_index(begin = var_999_begin_0, end = var_999_end_0, end_mask = var_999_end_mask_0, x = var_943_cast)[name = tensor<string, []>("op_999_cast")];
            tensor<int32, [4]> var_1000_begin_0 = const()[name = tensor<string, []>("op_1000_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_1000_end_0 = const()[name = tensor<string, []>("op_1000_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_1000_end_mask_0 = const()[name = tensor<string, []>("op_1000_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1000_cast = slice_by_index(begin = var_1000_begin_0, end = var_1000_end_0, end_mask = var_1000_end_mask_0, x = var_943_cast)[name = tensor<string, []>("op_1000_cast")];
            tensor<int32, [4]> var_1001_begin_0 = const()[name = tensor<string, []>("op_1001_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_1001_end_0 = const()[name = tensor<string, []>("op_1001_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_1001_end_mask_0 = const()[name = tensor<string, []>("op_1001_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1001_cast = slice_by_index(begin = var_1001_begin_0, end = var_1001_end_0, end_mask = var_1001_end_mask_0, x = var_943_cast)[name = tensor<string, []>("op_1001_cast")];
            tensor<int32, [4]> var_1002_begin_0 = const()[name = tensor<string, []>("op_1002_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_1002_end_0 = const()[name = tensor<string, []>("op_1002_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_1002_end_mask_0 = const()[name = tensor<string, []>("op_1002_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1002_cast = slice_by_index(begin = var_1002_begin_0, end = var_1002_end_0, end_mask = var_1002_end_mask_0, x = var_943_cast)[name = tensor<string, []>("op_1002_cast")];
            tensor<int32, [4]> var_1003_begin_0 = const()[name = tensor<string, []>("op_1003_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_1003_end_0 = const()[name = tensor<string, []>("op_1003_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_1003_end_mask_0 = const()[name = tensor<string, []>("op_1003_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1003_cast = slice_by_index(begin = var_1003_begin_0, end = var_1003_end_0, end_mask = var_1003_end_mask_0, x = var_943_cast)[name = tensor<string, []>("op_1003_cast")];
            tensor<int32, [4]> var_1004_begin_0 = const()[name = tensor<string, []>("op_1004_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_1004_end_0 = const()[name = tensor<string, []>("op_1004_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_1004_end_mask_0 = const()[name = tensor<string, []>("op_1004_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1004_cast = slice_by_index(begin = var_1004_begin_0, end = var_1004_end_0, end_mask = var_1004_end_mask_0, x = var_943_cast)[name = tensor<string, []>("op_1004_cast")];
            tensor<int32, [4]> var_1005_begin_0 = const()[name = tensor<string, []>("op_1005_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_1005_end_0 = const()[name = tensor<string, []>("op_1005_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_1005_end_mask_0 = const()[name = tensor<string, []>("op_1005_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1005_cast = slice_by_index(begin = var_1005_begin_0, end = var_1005_end_0, end_mask = var_1005_end_mask_0, x = var_943_cast)[name = tensor<string, []>("op_1005_cast")];
            tensor<int32, [4]> var_1006_begin_0 = const()[name = tensor<string, []>("op_1006_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1006_end_0 = const()[name = tensor<string, []>("op_1006_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_1006_end_mask_0 = const()[name = tensor<string, []>("op_1006_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1006_cast = slice_by_index(begin = var_1006_begin_0, end = var_1006_end_0, end_mask = var_1006_end_mask_0, x = var_947_cast)[name = tensor<string, []>("op_1006_cast")];
            tensor<int32, [4]> var_1007_begin_0 = const()[name = tensor<string, []>("op_1007_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_1007_end_0 = const()[name = tensor<string, []>("op_1007_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_1007_end_mask_0 = const()[name = tensor<string, []>("op_1007_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1007_cast = slice_by_index(begin = var_1007_begin_0, end = var_1007_end_0, end_mask = var_1007_end_mask_0, x = var_947_cast)[name = tensor<string, []>("op_1007_cast")];
            tensor<int32, [4]> var_1008_begin_0 = const()[name = tensor<string, []>("op_1008_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_1008_end_0 = const()[name = tensor<string, []>("op_1008_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_1008_end_mask_0 = const()[name = tensor<string, []>("op_1008_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1008_cast = slice_by_index(begin = var_1008_begin_0, end = var_1008_end_0, end_mask = var_1008_end_mask_0, x = var_947_cast)[name = tensor<string, []>("op_1008_cast")];
            tensor<int32, [4]> var_1009_begin_0 = const()[name = tensor<string, []>("op_1009_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_1009_end_0 = const()[name = tensor<string, []>("op_1009_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_1009_end_mask_0 = const()[name = tensor<string, []>("op_1009_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1009_cast = slice_by_index(begin = var_1009_begin_0, end = var_1009_end_0, end_mask = var_1009_end_mask_0, x = var_947_cast)[name = tensor<string, []>("op_1009_cast")];
            tensor<int32, [4]> var_1010_begin_0 = const()[name = tensor<string, []>("op_1010_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_1010_end_0 = const()[name = tensor<string, []>("op_1010_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_1010_end_mask_0 = const()[name = tensor<string, []>("op_1010_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1010_cast = slice_by_index(begin = var_1010_begin_0, end = var_1010_end_0, end_mask = var_1010_end_mask_0, x = var_947_cast)[name = tensor<string, []>("op_1010_cast")];
            tensor<int32, [4]> var_1011_begin_0 = const()[name = tensor<string, []>("op_1011_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_1011_end_0 = const()[name = tensor<string, []>("op_1011_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_1011_end_mask_0 = const()[name = tensor<string, []>("op_1011_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1011_cast = slice_by_index(begin = var_1011_begin_0, end = var_1011_end_0, end_mask = var_1011_end_mask_0, x = var_947_cast)[name = tensor<string, []>("op_1011_cast")];
            tensor<int32, [4]> var_1012_begin_0 = const()[name = tensor<string, []>("op_1012_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_1012_end_0 = const()[name = tensor<string, []>("op_1012_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_1012_end_mask_0 = const()[name = tensor<string, []>("op_1012_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1012_cast = slice_by_index(begin = var_1012_begin_0, end = var_1012_end_0, end_mask = var_1012_end_mask_0, x = var_947_cast)[name = tensor<string, []>("op_1012_cast")];
            tensor<int32, [4]> var_1013_begin_0 = const()[name = tensor<string, []>("op_1013_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_1013_end_0 = const()[name = tensor<string, []>("op_1013_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_1013_end_mask_0 = const()[name = tensor<string, []>("op_1013_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1013_cast = slice_by_index(begin = var_1013_begin_0, end = var_1013_end_0, end_mask = var_1013_end_mask_0, x = var_947_cast)[name = tensor<string, []>("op_1013_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_1018_begin_0 = const()[name = tensor<string, []>("op_1018_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1018_end_0 = const()[name = tensor<string, []>("op_1018_end_0"), val = tensor<int32, [4]>([2, 77, 1, 40])];
            tensor<bool, [4]> var_1018_end_mask_0 = const()[name = tensor<string, []>("op_1018_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)[name = tensor<string, []>("transpose_30")];
            tensor<fp16, [2, 77, 1, 40]> var_1018_cast = slice_by_index(begin = var_1018_begin_0, end = var_1018_end_0, end_mask = var_1018_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_1018_cast")];
            tensor<int32, [4]> var_1022_begin_0 = const()[name = tensor<string, []>("op_1022_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 40])];
            tensor<int32, [4]> var_1022_end_0 = const()[name = tensor<string, []>("op_1022_end_0"), val = tensor<int32, [4]>([2, 77, 1, 80])];
            tensor<bool, [4]> var_1022_end_mask_0 = const()[name = tensor<string, []>("op_1022_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 40]> var_1022_cast = slice_by_index(begin = var_1022_begin_0, end = var_1022_end_0, end_mask = var_1022_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_1022_cast")];
            tensor<int32, [4]> var_1026_begin_0 = const()[name = tensor<string, []>("op_1026_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 80])];
            tensor<int32, [4]> var_1026_end_0 = const()[name = tensor<string, []>("op_1026_end_0"), val = tensor<int32, [4]>([2, 77, 1, 120])];
            tensor<bool, [4]> var_1026_end_mask_0 = const()[name = tensor<string, []>("op_1026_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 40]> var_1026_cast = slice_by_index(begin = var_1026_begin_0, end = var_1026_end_0, end_mask = var_1026_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_1026_cast")];
            tensor<int32, [4]> var_1030_begin_0 = const()[name = tensor<string, []>("op_1030_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 120])];
            tensor<int32, [4]> var_1030_end_0 = const()[name = tensor<string, []>("op_1030_end_0"), val = tensor<int32, [4]>([2, 77, 1, 160])];
            tensor<bool, [4]> var_1030_end_mask_0 = const()[name = tensor<string, []>("op_1030_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 40]> var_1030_cast = slice_by_index(begin = var_1030_begin_0, end = var_1030_end_0, end_mask = var_1030_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_1030_cast")];
            tensor<int32, [4]> var_1034_begin_0 = const()[name = tensor<string, []>("op_1034_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_1034_end_0 = const()[name = tensor<string, []>("op_1034_end_0"), val = tensor<int32, [4]>([2, 77, 1, 200])];
            tensor<bool, [4]> var_1034_end_mask_0 = const()[name = tensor<string, []>("op_1034_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 40]> var_1034_cast = slice_by_index(begin = var_1034_begin_0, end = var_1034_end_0, end_mask = var_1034_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_1034_cast")];
            tensor<int32, [4]> var_1038_begin_0 = const()[name = tensor<string, []>("op_1038_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 200])];
            tensor<int32, [4]> var_1038_end_0 = const()[name = tensor<string, []>("op_1038_end_0"), val = tensor<int32, [4]>([2, 77, 1, 240])];
            tensor<bool, [4]> var_1038_end_mask_0 = const()[name = tensor<string, []>("op_1038_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 40]> var_1038_cast = slice_by_index(begin = var_1038_begin_0, end = var_1038_end_0, end_mask = var_1038_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_1038_cast")];
            tensor<int32, [4]> var_1042_begin_0 = const()[name = tensor<string, []>("op_1042_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 240])];
            tensor<int32, [4]> var_1042_end_0 = const()[name = tensor<string, []>("op_1042_end_0"), val = tensor<int32, [4]>([2, 77, 1, 280])];
            tensor<bool, [4]> var_1042_end_mask_0 = const()[name = tensor<string, []>("op_1042_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 40]> var_1042_cast = slice_by_index(begin = var_1042_begin_0, end = var_1042_end_0, end_mask = var_1042_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_1042_cast")];
            tensor<int32, [4]> var_1046_begin_0 = const()[name = tensor<string, []>("op_1046_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 280])];
            tensor<int32, [4]> var_1046_end_0 = const()[name = tensor<string, []>("op_1046_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_1046_end_mask_0 = const()[name = tensor<string, []>("op_1046_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 40]> var_1046_cast = slice_by_index(begin = var_1046_begin_0, end = var_1046_end_0, end_mask = var_1046_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_1046_cast")];
            tensor<int32, [4]> var_1048_begin_0 = const()[name = tensor<string, []>("op_1048_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1048_end_0 = const()[name = tensor<string, []>("op_1048_end_0"), val = tensor<int32, [4]>([2, 40, 1, 77])];
            tensor<bool, [4]> var_1048_end_mask_0 = const()[name = tensor<string, []>("op_1048_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_1048_cast = slice_by_index(begin = var_1048_begin_0, end = var_1048_end_0, end_mask = var_1048_end_mask_0, x = v_3_cast)[name = tensor<string, []>("op_1048_cast")];
            tensor<int32, [4]> var_1052_begin_0 = const()[name = tensor<string, []>("op_1052_begin_0"), val = tensor<int32, [4]>([0, 40, 0, 0])];
            tensor<int32, [4]> var_1052_end_0 = const()[name = tensor<string, []>("op_1052_end_0"), val = tensor<int32, [4]>([2, 80, 1, 77])];
            tensor<bool, [4]> var_1052_end_mask_0 = const()[name = tensor<string, []>("op_1052_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_1052_cast = slice_by_index(begin = var_1052_begin_0, end = var_1052_end_0, end_mask = var_1052_end_mask_0, x = v_3_cast)[name = tensor<string, []>("op_1052_cast")];
            tensor<int32, [4]> var_1056_begin_0 = const()[name = tensor<string, []>("op_1056_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_1056_end_0 = const()[name = tensor<string, []>("op_1056_end_0"), val = tensor<int32, [4]>([2, 120, 1, 77])];
            tensor<bool, [4]> var_1056_end_mask_0 = const()[name = tensor<string, []>("op_1056_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_1056_cast = slice_by_index(begin = var_1056_begin_0, end = var_1056_end_0, end_mask = var_1056_end_mask_0, x = v_3_cast)[name = tensor<string, []>("op_1056_cast")];
            tensor<int32, [4]> var_1060_begin_0 = const()[name = tensor<string, []>("op_1060_begin_0"), val = tensor<int32, [4]>([0, 120, 0, 0])];
            tensor<int32, [4]> var_1060_end_0 = const()[name = tensor<string, []>("op_1060_end_0"), val = tensor<int32, [4]>([2, 160, 1, 77])];
            tensor<bool, [4]> var_1060_end_mask_0 = const()[name = tensor<string, []>("op_1060_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_1060_cast = slice_by_index(begin = var_1060_begin_0, end = var_1060_end_0, end_mask = var_1060_end_mask_0, x = v_3_cast)[name = tensor<string, []>("op_1060_cast")];
            tensor<int32, [4]> var_1064_begin_0 = const()[name = tensor<string, []>("op_1064_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_1064_end_0 = const()[name = tensor<string, []>("op_1064_end_0"), val = tensor<int32, [4]>([2, 200, 1, 77])];
            tensor<bool, [4]> var_1064_end_mask_0 = const()[name = tensor<string, []>("op_1064_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_1064_cast = slice_by_index(begin = var_1064_begin_0, end = var_1064_end_0, end_mask = var_1064_end_mask_0, x = v_3_cast)[name = tensor<string, []>("op_1064_cast")];
            tensor<int32, [4]> var_1068_begin_0 = const()[name = tensor<string, []>("op_1068_begin_0"), val = tensor<int32, [4]>([0, 200, 0, 0])];
            tensor<int32, [4]> var_1068_end_0 = const()[name = tensor<string, []>("op_1068_end_0"), val = tensor<int32, [4]>([2, 240, 1, 77])];
            tensor<bool, [4]> var_1068_end_mask_0 = const()[name = tensor<string, []>("op_1068_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_1068_cast = slice_by_index(begin = var_1068_begin_0, end = var_1068_end_0, end_mask = var_1068_end_mask_0, x = v_3_cast)[name = tensor<string, []>("op_1068_cast")];
            tensor<int32, [4]> var_1072_begin_0 = const()[name = tensor<string, []>("op_1072_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_1072_end_0 = const()[name = tensor<string, []>("op_1072_end_0"), val = tensor<int32, [4]>([2, 280, 1, 77])];
            tensor<bool, [4]> var_1072_end_mask_0 = const()[name = tensor<string, []>("op_1072_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_1072_cast = slice_by_index(begin = var_1072_begin_0, end = var_1072_end_0, end_mask = var_1072_end_mask_0, x = v_3_cast)[name = tensor<string, []>("op_1072_cast")];
            tensor<int32, [4]> var_1076_begin_0 = const()[name = tensor<string, []>("op_1076_begin_0"), val = tensor<int32, [4]>([0, 280, 0, 0])];
            tensor<int32, [4]> var_1076_end_0 = const()[name = tensor<string, []>("op_1076_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_1076_end_mask_0 = const()[name = tensor<string, []>("op_1076_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_1076_cast = slice_by_index(begin = var_1076_begin_0, end = var_1076_end_0, end_mask = var_1076_end_mask_0, x = v_3_cast)[name = tensor<string, []>("op_1076_cast")];
            tensor<string, []> var_1080_equation_0 = const()[name = tensor<string, []>("op_1080_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1080_cast = einsum(equation = var_1080_equation_0, values = (var_1018_cast, var_950_cast))[name = tensor<string, []>("op_1080_cast")];
            tensor<fp16, []> var_1081_to_fp16 = const()[name = tensor<string, []>("op_1081_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_129_cast = mul(x = var_1080_cast, y = var_1081_to_fp16)[name = tensor<string, []>("aw_chunk_129_cast")];
            tensor<string, []> var_1084_equation_0 = const()[name = tensor<string, []>("op_1084_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1084_cast = einsum(equation = var_1084_equation_0, values = (var_1018_cast, var_951_cast))[name = tensor<string, []>("op_1084_cast")];
            tensor<fp16, []> var_1085_to_fp16 = const()[name = tensor<string, []>("op_1085_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_131_cast = mul(x = var_1084_cast, y = var_1085_to_fp16)[name = tensor<string, []>("aw_chunk_131_cast")];
            tensor<string, []> var_1088_equation_0 = const()[name = tensor<string, []>("op_1088_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1088_cast = einsum(equation = var_1088_equation_0, values = (var_1018_cast, var_952_cast))[name = tensor<string, []>("op_1088_cast")];
            tensor<fp16, []> var_1089_to_fp16 = const()[name = tensor<string, []>("op_1089_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_133_cast = mul(x = var_1088_cast, y = var_1089_to_fp16)[name = tensor<string, []>("aw_chunk_133_cast")];
            tensor<string, []> var_1092_equation_0 = const()[name = tensor<string, []>("op_1092_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1092_cast = einsum(equation = var_1092_equation_0, values = (var_1018_cast, var_953_cast))[name = tensor<string, []>("op_1092_cast")];
            tensor<fp16, []> var_1093_to_fp16 = const()[name = tensor<string, []>("op_1093_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_135_cast = mul(x = var_1092_cast, y = var_1093_to_fp16)[name = tensor<string, []>("aw_chunk_135_cast")];
            tensor<string, []> var_1096_equation_0 = const()[name = tensor<string, []>("op_1096_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1096_cast = einsum(equation = var_1096_equation_0, values = (var_1018_cast, var_954_cast))[name = tensor<string, []>("op_1096_cast")];
            tensor<fp16, []> var_1097_to_fp16 = const()[name = tensor<string, []>("op_1097_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_137_cast = mul(x = var_1096_cast, y = var_1097_to_fp16)[name = tensor<string, []>("aw_chunk_137_cast")];
            tensor<string, []> var_1100_equation_0 = const()[name = tensor<string, []>("op_1100_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1100_cast = einsum(equation = var_1100_equation_0, values = (var_1018_cast, var_955_cast))[name = tensor<string, []>("op_1100_cast")];
            tensor<fp16, []> var_1101_to_fp16 = const()[name = tensor<string, []>("op_1101_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_139_cast = mul(x = var_1100_cast, y = var_1101_to_fp16)[name = tensor<string, []>("aw_chunk_139_cast")];
            tensor<string, []> var_1104_equation_0 = const()[name = tensor<string, []>("op_1104_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1104_cast = einsum(equation = var_1104_equation_0, values = (var_1018_cast, var_956_cast))[name = tensor<string, []>("op_1104_cast")];
            tensor<fp16, []> var_1105_to_fp16 = const()[name = tensor<string, []>("op_1105_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_141_cast = mul(x = var_1104_cast, y = var_1105_to_fp16)[name = tensor<string, []>("aw_chunk_141_cast")];
            tensor<string, []> var_1108_equation_0 = const()[name = tensor<string, []>("op_1108_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1108_cast = einsum(equation = var_1108_equation_0, values = (var_1018_cast, var_957_cast))[name = tensor<string, []>("op_1108_cast")];
            tensor<fp16, []> var_1109_to_fp16 = const()[name = tensor<string, []>("op_1109_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_143_cast = mul(x = var_1108_cast, y = var_1109_to_fp16)[name = tensor<string, []>("aw_chunk_143_cast")];
            tensor<string, []> var_1112_equation_0 = const()[name = tensor<string, []>("op_1112_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1112_cast = einsum(equation = var_1112_equation_0, values = (var_1022_cast, var_958_cast))[name = tensor<string, []>("op_1112_cast")];
            tensor<fp16, []> var_1113_to_fp16 = const()[name = tensor<string, []>("op_1113_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_145_cast = mul(x = var_1112_cast, y = var_1113_to_fp16)[name = tensor<string, []>("aw_chunk_145_cast")];
            tensor<string, []> var_1116_equation_0 = const()[name = tensor<string, []>("op_1116_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1116_cast = einsum(equation = var_1116_equation_0, values = (var_1022_cast, var_959_cast))[name = tensor<string, []>("op_1116_cast")];
            tensor<fp16, []> var_1117_to_fp16 = const()[name = tensor<string, []>("op_1117_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_147_cast = mul(x = var_1116_cast, y = var_1117_to_fp16)[name = tensor<string, []>("aw_chunk_147_cast")];
            tensor<string, []> var_1120_equation_0 = const()[name = tensor<string, []>("op_1120_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1120_cast = einsum(equation = var_1120_equation_0, values = (var_1022_cast, var_960_cast))[name = tensor<string, []>("op_1120_cast")];
            tensor<fp16, []> var_1121_to_fp16 = const()[name = tensor<string, []>("op_1121_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_149_cast = mul(x = var_1120_cast, y = var_1121_to_fp16)[name = tensor<string, []>("aw_chunk_149_cast")];
            tensor<string, []> var_1124_equation_0 = const()[name = tensor<string, []>("op_1124_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1124_cast = einsum(equation = var_1124_equation_0, values = (var_1022_cast, var_961_cast))[name = tensor<string, []>("op_1124_cast")];
            tensor<fp16, []> var_1125_to_fp16 = const()[name = tensor<string, []>("op_1125_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_151_cast = mul(x = var_1124_cast, y = var_1125_to_fp16)[name = tensor<string, []>("aw_chunk_151_cast")];
            tensor<string, []> var_1128_equation_0 = const()[name = tensor<string, []>("op_1128_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1128_cast = einsum(equation = var_1128_equation_0, values = (var_1022_cast, var_962_cast))[name = tensor<string, []>("op_1128_cast")];
            tensor<fp16, []> var_1129_to_fp16 = const()[name = tensor<string, []>("op_1129_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_153_cast = mul(x = var_1128_cast, y = var_1129_to_fp16)[name = tensor<string, []>("aw_chunk_153_cast")];
            tensor<string, []> var_1132_equation_0 = const()[name = tensor<string, []>("op_1132_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1132_cast = einsum(equation = var_1132_equation_0, values = (var_1022_cast, var_963_cast))[name = tensor<string, []>("op_1132_cast")];
            tensor<fp16, []> var_1133_to_fp16 = const()[name = tensor<string, []>("op_1133_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_155_cast = mul(x = var_1132_cast, y = var_1133_to_fp16)[name = tensor<string, []>("aw_chunk_155_cast")];
            tensor<string, []> var_1136_equation_0 = const()[name = tensor<string, []>("op_1136_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1136_cast = einsum(equation = var_1136_equation_0, values = (var_1022_cast, var_964_cast))[name = tensor<string, []>("op_1136_cast")];
            tensor<fp16, []> var_1137_to_fp16 = const()[name = tensor<string, []>("op_1137_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_157_cast = mul(x = var_1136_cast, y = var_1137_to_fp16)[name = tensor<string, []>("aw_chunk_157_cast")];
            tensor<string, []> var_1140_equation_0 = const()[name = tensor<string, []>("op_1140_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1140_cast = einsum(equation = var_1140_equation_0, values = (var_1022_cast, var_965_cast))[name = tensor<string, []>("op_1140_cast")];
            tensor<fp16, []> var_1141_to_fp16 = const()[name = tensor<string, []>("op_1141_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_159_cast = mul(x = var_1140_cast, y = var_1141_to_fp16)[name = tensor<string, []>("aw_chunk_159_cast")];
            tensor<string, []> var_1144_equation_0 = const()[name = tensor<string, []>("op_1144_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1144_cast = einsum(equation = var_1144_equation_0, values = (var_1026_cast, var_966_cast))[name = tensor<string, []>("op_1144_cast")];
            tensor<fp16, []> var_1145_to_fp16 = const()[name = tensor<string, []>("op_1145_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_161_cast = mul(x = var_1144_cast, y = var_1145_to_fp16)[name = tensor<string, []>("aw_chunk_161_cast")];
            tensor<string, []> var_1148_equation_0 = const()[name = tensor<string, []>("op_1148_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1148_cast = einsum(equation = var_1148_equation_0, values = (var_1026_cast, var_967_cast))[name = tensor<string, []>("op_1148_cast")];
            tensor<fp16, []> var_1149_to_fp16 = const()[name = tensor<string, []>("op_1149_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_163_cast = mul(x = var_1148_cast, y = var_1149_to_fp16)[name = tensor<string, []>("aw_chunk_163_cast")];
            tensor<string, []> var_1152_equation_0 = const()[name = tensor<string, []>("op_1152_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1152_cast = einsum(equation = var_1152_equation_0, values = (var_1026_cast, var_968_cast))[name = tensor<string, []>("op_1152_cast")];
            tensor<fp16, []> var_1153_to_fp16 = const()[name = tensor<string, []>("op_1153_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_165_cast = mul(x = var_1152_cast, y = var_1153_to_fp16)[name = tensor<string, []>("aw_chunk_165_cast")];
            tensor<string, []> var_1156_equation_0 = const()[name = tensor<string, []>("op_1156_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1156_cast = einsum(equation = var_1156_equation_0, values = (var_1026_cast, var_969_cast))[name = tensor<string, []>("op_1156_cast")];
            tensor<fp16, []> var_1157_to_fp16 = const()[name = tensor<string, []>("op_1157_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_167_cast = mul(x = var_1156_cast, y = var_1157_to_fp16)[name = tensor<string, []>("aw_chunk_167_cast")];
            tensor<string, []> var_1160_equation_0 = const()[name = tensor<string, []>("op_1160_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1160_cast = einsum(equation = var_1160_equation_0, values = (var_1026_cast, var_970_cast))[name = tensor<string, []>("op_1160_cast")];
            tensor<fp16, []> var_1161_to_fp16 = const()[name = tensor<string, []>("op_1161_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_169_cast = mul(x = var_1160_cast, y = var_1161_to_fp16)[name = tensor<string, []>("aw_chunk_169_cast")];
            tensor<string, []> var_1164_equation_0 = const()[name = tensor<string, []>("op_1164_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1164_cast = einsum(equation = var_1164_equation_0, values = (var_1026_cast, var_971_cast))[name = tensor<string, []>("op_1164_cast")];
            tensor<fp16, []> var_1165_to_fp16 = const()[name = tensor<string, []>("op_1165_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_171_cast = mul(x = var_1164_cast, y = var_1165_to_fp16)[name = tensor<string, []>("aw_chunk_171_cast")];
            tensor<string, []> var_1168_equation_0 = const()[name = tensor<string, []>("op_1168_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1168_cast = einsum(equation = var_1168_equation_0, values = (var_1026_cast, var_972_cast))[name = tensor<string, []>("op_1168_cast")];
            tensor<fp16, []> var_1169_to_fp16 = const()[name = tensor<string, []>("op_1169_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_173_cast = mul(x = var_1168_cast, y = var_1169_to_fp16)[name = tensor<string, []>("aw_chunk_173_cast")];
            tensor<string, []> var_1172_equation_0 = const()[name = tensor<string, []>("op_1172_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1172_cast = einsum(equation = var_1172_equation_0, values = (var_1026_cast, var_973_cast))[name = tensor<string, []>("op_1172_cast")];
            tensor<fp16, []> var_1173_to_fp16 = const()[name = tensor<string, []>("op_1173_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_175_cast = mul(x = var_1172_cast, y = var_1173_to_fp16)[name = tensor<string, []>("aw_chunk_175_cast")];
            tensor<string, []> var_1176_equation_0 = const()[name = tensor<string, []>("op_1176_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1176_cast = einsum(equation = var_1176_equation_0, values = (var_1030_cast, var_974_cast))[name = tensor<string, []>("op_1176_cast")];
            tensor<fp16, []> var_1177_to_fp16 = const()[name = tensor<string, []>("op_1177_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_177_cast = mul(x = var_1176_cast, y = var_1177_to_fp16)[name = tensor<string, []>("aw_chunk_177_cast")];
            tensor<string, []> var_1180_equation_0 = const()[name = tensor<string, []>("op_1180_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1180_cast = einsum(equation = var_1180_equation_0, values = (var_1030_cast, var_975_cast))[name = tensor<string, []>("op_1180_cast")];
            tensor<fp16, []> var_1181_to_fp16 = const()[name = tensor<string, []>("op_1181_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_179_cast = mul(x = var_1180_cast, y = var_1181_to_fp16)[name = tensor<string, []>("aw_chunk_179_cast")];
            tensor<string, []> var_1184_equation_0 = const()[name = tensor<string, []>("op_1184_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1184_cast = einsum(equation = var_1184_equation_0, values = (var_1030_cast, var_976_cast))[name = tensor<string, []>("op_1184_cast")];
            tensor<fp16, []> var_1185_to_fp16 = const()[name = tensor<string, []>("op_1185_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_181_cast = mul(x = var_1184_cast, y = var_1185_to_fp16)[name = tensor<string, []>("aw_chunk_181_cast")];
            tensor<string, []> var_1188_equation_0 = const()[name = tensor<string, []>("op_1188_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1188_cast = einsum(equation = var_1188_equation_0, values = (var_1030_cast, var_977_cast))[name = tensor<string, []>("op_1188_cast")];
            tensor<fp16, []> var_1189_to_fp16 = const()[name = tensor<string, []>("op_1189_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_183_cast = mul(x = var_1188_cast, y = var_1189_to_fp16)[name = tensor<string, []>("aw_chunk_183_cast")];
            tensor<string, []> var_1192_equation_0 = const()[name = tensor<string, []>("op_1192_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1192_cast = einsum(equation = var_1192_equation_0, values = (var_1030_cast, var_978_cast))[name = tensor<string, []>("op_1192_cast")];
            tensor<fp16, []> var_1193_to_fp16 = const()[name = tensor<string, []>("op_1193_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_185_cast = mul(x = var_1192_cast, y = var_1193_to_fp16)[name = tensor<string, []>("aw_chunk_185_cast")];
            tensor<string, []> var_1196_equation_0 = const()[name = tensor<string, []>("op_1196_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1196_cast = einsum(equation = var_1196_equation_0, values = (var_1030_cast, var_979_cast))[name = tensor<string, []>("op_1196_cast")];
            tensor<fp16, []> var_1197_to_fp16 = const()[name = tensor<string, []>("op_1197_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_187_cast = mul(x = var_1196_cast, y = var_1197_to_fp16)[name = tensor<string, []>("aw_chunk_187_cast")];
            tensor<string, []> var_1200_equation_0 = const()[name = tensor<string, []>("op_1200_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1200_cast = einsum(equation = var_1200_equation_0, values = (var_1030_cast, var_980_cast))[name = tensor<string, []>("op_1200_cast")];
            tensor<fp16, []> var_1201_to_fp16 = const()[name = tensor<string, []>("op_1201_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_189_cast = mul(x = var_1200_cast, y = var_1201_to_fp16)[name = tensor<string, []>("aw_chunk_189_cast")];
            tensor<string, []> var_1204_equation_0 = const()[name = tensor<string, []>("op_1204_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1204_cast = einsum(equation = var_1204_equation_0, values = (var_1030_cast, var_981_cast))[name = tensor<string, []>("op_1204_cast")];
            tensor<fp16, []> var_1205_to_fp16 = const()[name = tensor<string, []>("op_1205_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_191_cast = mul(x = var_1204_cast, y = var_1205_to_fp16)[name = tensor<string, []>("aw_chunk_191_cast")];
            tensor<string, []> var_1208_equation_0 = const()[name = tensor<string, []>("op_1208_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1208_cast = einsum(equation = var_1208_equation_0, values = (var_1034_cast, var_982_cast))[name = tensor<string, []>("op_1208_cast")];
            tensor<fp16, []> var_1209_to_fp16 = const()[name = tensor<string, []>("op_1209_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_193_cast = mul(x = var_1208_cast, y = var_1209_to_fp16)[name = tensor<string, []>("aw_chunk_193_cast")];
            tensor<string, []> var_1212_equation_0 = const()[name = tensor<string, []>("op_1212_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1212_cast = einsum(equation = var_1212_equation_0, values = (var_1034_cast, var_983_cast))[name = tensor<string, []>("op_1212_cast")];
            tensor<fp16, []> var_1213_to_fp16 = const()[name = tensor<string, []>("op_1213_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_195_cast = mul(x = var_1212_cast, y = var_1213_to_fp16)[name = tensor<string, []>("aw_chunk_195_cast")];
            tensor<string, []> var_1216_equation_0 = const()[name = tensor<string, []>("op_1216_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1216_cast = einsum(equation = var_1216_equation_0, values = (var_1034_cast, var_984_cast))[name = tensor<string, []>("op_1216_cast")];
            tensor<fp16, []> var_1217_to_fp16 = const()[name = tensor<string, []>("op_1217_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_197_cast = mul(x = var_1216_cast, y = var_1217_to_fp16)[name = tensor<string, []>("aw_chunk_197_cast")];
            tensor<string, []> var_1220_equation_0 = const()[name = tensor<string, []>("op_1220_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1220_cast = einsum(equation = var_1220_equation_0, values = (var_1034_cast, var_985_cast))[name = tensor<string, []>("op_1220_cast")];
            tensor<fp16, []> var_1221_to_fp16 = const()[name = tensor<string, []>("op_1221_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_199_cast = mul(x = var_1220_cast, y = var_1221_to_fp16)[name = tensor<string, []>("aw_chunk_199_cast")];
            tensor<string, []> var_1224_equation_0 = const()[name = tensor<string, []>("op_1224_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1224_cast = einsum(equation = var_1224_equation_0, values = (var_1034_cast, var_986_cast))[name = tensor<string, []>("op_1224_cast")];
            tensor<fp16, []> var_1225_to_fp16 = const()[name = tensor<string, []>("op_1225_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_201_cast = mul(x = var_1224_cast, y = var_1225_to_fp16)[name = tensor<string, []>("aw_chunk_201_cast")];
            tensor<string, []> var_1228_equation_0 = const()[name = tensor<string, []>("op_1228_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1228_cast = einsum(equation = var_1228_equation_0, values = (var_1034_cast, var_987_cast))[name = tensor<string, []>("op_1228_cast")];
            tensor<fp16, []> var_1229_to_fp16 = const()[name = tensor<string, []>("op_1229_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_203_cast = mul(x = var_1228_cast, y = var_1229_to_fp16)[name = tensor<string, []>("aw_chunk_203_cast")];
            tensor<string, []> var_1232_equation_0 = const()[name = tensor<string, []>("op_1232_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1232_cast = einsum(equation = var_1232_equation_0, values = (var_1034_cast, var_988_cast))[name = tensor<string, []>("op_1232_cast")];
            tensor<fp16, []> var_1233_to_fp16 = const()[name = tensor<string, []>("op_1233_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_205_cast = mul(x = var_1232_cast, y = var_1233_to_fp16)[name = tensor<string, []>("aw_chunk_205_cast")];
            tensor<string, []> var_1236_equation_0 = const()[name = tensor<string, []>("op_1236_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1236_cast = einsum(equation = var_1236_equation_0, values = (var_1034_cast, var_989_cast))[name = tensor<string, []>("op_1236_cast")];
            tensor<fp16, []> var_1237_to_fp16 = const()[name = tensor<string, []>("op_1237_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_207_cast = mul(x = var_1236_cast, y = var_1237_to_fp16)[name = tensor<string, []>("aw_chunk_207_cast")];
            tensor<string, []> var_1240_equation_0 = const()[name = tensor<string, []>("op_1240_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1240_cast = einsum(equation = var_1240_equation_0, values = (var_1038_cast, var_990_cast))[name = tensor<string, []>("op_1240_cast")];
            tensor<fp16, []> var_1241_to_fp16 = const()[name = tensor<string, []>("op_1241_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_209_cast = mul(x = var_1240_cast, y = var_1241_to_fp16)[name = tensor<string, []>("aw_chunk_209_cast")];
            tensor<string, []> var_1244_equation_0 = const()[name = tensor<string, []>("op_1244_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1244_cast = einsum(equation = var_1244_equation_0, values = (var_1038_cast, var_991_cast))[name = tensor<string, []>("op_1244_cast")];
            tensor<fp16, []> var_1245_to_fp16 = const()[name = tensor<string, []>("op_1245_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_211_cast = mul(x = var_1244_cast, y = var_1245_to_fp16)[name = tensor<string, []>("aw_chunk_211_cast")];
            tensor<string, []> var_1248_equation_0 = const()[name = tensor<string, []>("op_1248_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1248_cast = einsum(equation = var_1248_equation_0, values = (var_1038_cast, var_992_cast))[name = tensor<string, []>("op_1248_cast")];
            tensor<fp16, []> var_1249_to_fp16 = const()[name = tensor<string, []>("op_1249_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_213_cast = mul(x = var_1248_cast, y = var_1249_to_fp16)[name = tensor<string, []>("aw_chunk_213_cast")];
            tensor<string, []> var_1252_equation_0 = const()[name = tensor<string, []>("op_1252_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1252_cast = einsum(equation = var_1252_equation_0, values = (var_1038_cast, var_993_cast))[name = tensor<string, []>("op_1252_cast")];
            tensor<fp16, []> var_1253_to_fp16 = const()[name = tensor<string, []>("op_1253_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_215_cast = mul(x = var_1252_cast, y = var_1253_to_fp16)[name = tensor<string, []>("aw_chunk_215_cast")];
            tensor<string, []> var_1256_equation_0 = const()[name = tensor<string, []>("op_1256_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1256_cast = einsum(equation = var_1256_equation_0, values = (var_1038_cast, var_994_cast))[name = tensor<string, []>("op_1256_cast")];
            tensor<fp16, []> var_1257_to_fp16 = const()[name = tensor<string, []>("op_1257_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_217_cast = mul(x = var_1256_cast, y = var_1257_to_fp16)[name = tensor<string, []>("aw_chunk_217_cast")];
            tensor<string, []> var_1260_equation_0 = const()[name = tensor<string, []>("op_1260_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1260_cast = einsum(equation = var_1260_equation_0, values = (var_1038_cast, var_995_cast))[name = tensor<string, []>("op_1260_cast")];
            tensor<fp16, []> var_1261_to_fp16 = const()[name = tensor<string, []>("op_1261_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_219_cast = mul(x = var_1260_cast, y = var_1261_to_fp16)[name = tensor<string, []>("aw_chunk_219_cast")];
            tensor<string, []> var_1264_equation_0 = const()[name = tensor<string, []>("op_1264_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1264_cast = einsum(equation = var_1264_equation_0, values = (var_1038_cast, var_996_cast))[name = tensor<string, []>("op_1264_cast")];
            tensor<fp16, []> var_1265_to_fp16 = const()[name = tensor<string, []>("op_1265_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_221_cast = mul(x = var_1264_cast, y = var_1265_to_fp16)[name = tensor<string, []>("aw_chunk_221_cast")];
            tensor<string, []> var_1268_equation_0 = const()[name = tensor<string, []>("op_1268_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1268_cast = einsum(equation = var_1268_equation_0, values = (var_1038_cast, var_997_cast))[name = tensor<string, []>("op_1268_cast")];
            tensor<fp16, []> var_1269_to_fp16 = const()[name = tensor<string, []>("op_1269_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_223_cast = mul(x = var_1268_cast, y = var_1269_to_fp16)[name = tensor<string, []>("aw_chunk_223_cast")];
            tensor<string, []> var_1272_equation_0 = const()[name = tensor<string, []>("op_1272_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1272_cast = einsum(equation = var_1272_equation_0, values = (var_1042_cast, var_998_cast))[name = tensor<string, []>("op_1272_cast")];
            tensor<fp16, []> var_1273_to_fp16 = const()[name = tensor<string, []>("op_1273_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_225_cast = mul(x = var_1272_cast, y = var_1273_to_fp16)[name = tensor<string, []>("aw_chunk_225_cast")];
            tensor<string, []> var_1276_equation_0 = const()[name = tensor<string, []>("op_1276_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1276_cast = einsum(equation = var_1276_equation_0, values = (var_1042_cast, var_999_cast))[name = tensor<string, []>("op_1276_cast")];
            tensor<fp16, []> var_1277_to_fp16 = const()[name = tensor<string, []>("op_1277_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_227_cast = mul(x = var_1276_cast, y = var_1277_to_fp16)[name = tensor<string, []>("aw_chunk_227_cast")];
            tensor<string, []> var_1280_equation_0 = const()[name = tensor<string, []>("op_1280_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1280_cast = einsum(equation = var_1280_equation_0, values = (var_1042_cast, var_1000_cast))[name = tensor<string, []>("op_1280_cast")];
            tensor<fp16, []> var_1281_to_fp16 = const()[name = tensor<string, []>("op_1281_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_229_cast = mul(x = var_1280_cast, y = var_1281_to_fp16)[name = tensor<string, []>("aw_chunk_229_cast")];
            tensor<string, []> var_1284_equation_0 = const()[name = tensor<string, []>("op_1284_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1284_cast = einsum(equation = var_1284_equation_0, values = (var_1042_cast, var_1001_cast))[name = tensor<string, []>("op_1284_cast")];
            tensor<fp16, []> var_1285_to_fp16 = const()[name = tensor<string, []>("op_1285_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_231_cast = mul(x = var_1284_cast, y = var_1285_to_fp16)[name = tensor<string, []>("aw_chunk_231_cast")];
            tensor<string, []> var_1288_equation_0 = const()[name = tensor<string, []>("op_1288_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1288_cast = einsum(equation = var_1288_equation_0, values = (var_1042_cast, var_1002_cast))[name = tensor<string, []>("op_1288_cast")];
            tensor<fp16, []> var_1289_to_fp16 = const()[name = tensor<string, []>("op_1289_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_233_cast = mul(x = var_1288_cast, y = var_1289_to_fp16)[name = tensor<string, []>("aw_chunk_233_cast")];
            tensor<string, []> var_1292_equation_0 = const()[name = tensor<string, []>("op_1292_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1292_cast = einsum(equation = var_1292_equation_0, values = (var_1042_cast, var_1003_cast))[name = tensor<string, []>("op_1292_cast")];
            tensor<fp16, []> var_1293_to_fp16 = const()[name = tensor<string, []>("op_1293_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_235_cast = mul(x = var_1292_cast, y = var_1293_to_fp16)[name = tensor<string, []>("aw_chunk_235_cast")];
            tensor<string, []> var_1296_equation_0 = const()[name = tensor<string, []>("op_1296_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1296_cast = einsum(equation = var_1296_equation_0, values = (var_1042_cast, var_1004_cast))[name = tensor<string, []>("op_1296_cast")];
            tensor<fp16, []> var_1297_to_fp16 = const()[name = tensor<string, []>("op_1297_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_237_cast = mul(x = var_1296_cast, y = var_1297_to_fp16)[name = tensor<string, []>("aw_chunk_237_cast")];
            tensor<string, []> var_1300_equation_0 = const()[name = tensor<string, []>("op_1300_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1300_cast = einsum(equation = var_1300_equation_0, values = (var_1042_cast, var_1005_cast))[name = tensor<string, []>("op_1300_cast")];
            tensor<fp16, []> var_1301_to_fp16 = const()[name = tensor<string, []>("op_1301_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_239_cast = mul(x = var_1300_cast, y = var_1301_to_fp16)[name = tensor<string, []>("aw_chunk_239_cast")];
            tensor<string, []> var_1304_equation_0 = const()[name = tensor<string, []>("op_1304_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1304_cast = einsum(equation = var_1304_equation_0, values = (var_1046_cast, var_1006_cast))[name = tensor<string, []>("op_1304_cast")];
            tensor<fp16, []> var_1305_to_fp16 = const()[name = tensor<string, []>("op_1305_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_241_cast = mul(x = var_1304_cast, y = var_1305_to_fp16)[name = tensor<string, []>("aw_chunk_241_cast")];
            tensor<string, []> var_1308_equation_0 = const()[name = tensor<string, []>("op_1308_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1308_cast = einsum(equation = var_1308_equation_0, values = (var_1046_cast, var_1007_cast))[name = tensor<string, []>("op_1308_cast")];
            tensor<fp16, []> var_1309_to_fp16 = const()[name = tensor<string, []>("op_1309_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_243_cast = mul(x = var_1308_cast, y = var_1309_to_fp16)[name = tensor<string, []>("aw_chunk_243_cast")];
            tensor<string, []> var_1312_equation_0 = const()[name = tensor<string, []>("op_1312_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1312_cast = einsum(equation = var_1312_equation_0, values = (var_1046_cast, var_1008_cast))[name = tensor<string, []>("op_1312_cast")];
            tensor<fp16, []> var_1313_to_fp16 = const()[name = tensor<string, []>("op_1313_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_245_cast = mul(x = var_1312_cast, y = var_1313_to_fp16)[name = tensor<string, []>("aw_chunk_245_cast")];
            tensor<string, []> var_1316_equation_0 = const()[name = tensor<string, []>("op_1316_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1316_cast = einsum(equation = var_1316_equation_0, values = (var_1046_cast, var_1009_cast))[name = tensor<string, []>("op_1316_cast")];
            tensor<fp16, []> var_1317_to_fp16 = const()[name = tensor<string, []>("op_1317_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_247_cast = mul(x = var_1316_cast, y = var_1317_to_fp16)[name = tensor<string, []>("aw_chunk_247_cast")];
            tensor<string, []> var_1320_equation_0 = const()[name = tensor<string, []>("op_1320_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1320_cast = einsum(equation = var_1320_equation_0, values = (var_1046_cast, var_1010_cast))[name = tensor<string, []>("op_1320_cast")];
            tensor<fp16, []> var_1321_to_fp16 = const()[name = tensor<string, []>("op_1321_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_249_cast = mul(x = var_1320_cast, y = var_1321_to_fp16)[name = tensor<string, []>("aw_chunk_249_cast")];
            tensor<string, []> var_1324_equation_0 = const()[name = tensor<string, []>("op_1324_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1324_cast = einsum(equation = var_1324_equation_0, values = (var_1046_cast, var_1011_cast))[name = tensor<string, []>("op_1324_cast")];
            tensor<fp16, []> var_1325_to_fp16 = const()[name = tensor<string, []>("op_1325_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_251_cast = mul(x = var_1324_cast, y = var_1325_to_fp16)[name = tensor<string, []>("aw_chunk_251_cast")];
            tensor<string, []> var_1328_equation_0 = const()[name = tensor<string, []>("op_1328_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1328_cast = einsum(equation = var_1328_equation_0, values = (var_1046_cast, var_1012_cast))[name = tensor<string, []>("op_1328_cast")];
            tensor<fp16, []> var_1329_to_fp16 = const()[name = tensor<string, []>("op_1329_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_253_cast = mul(x = var_1328_cast, y = var_1329_to_fp16)[name = tensor<string, []>("aw_chunk_253_cast")];
            tensor<string, []> var_1332_equation_0 = const()[name = tensor<string, []>("op_1332_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1332_cast = einsum(equation = var_1332_equation_0, values = (var_1046_cast, var_1013_cast))[name = tensor<string, []>("op_1332_cast")];
            tensor<fp16, []> var_1333_to_fp16 = const()[name = tensor<string, []>("op_1333_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_255_cast = mul(x = var_1332_cast, y = var_1333_to_fp16)[name = tensor<string, []>("aw_chunk_255_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1335_cast = softmax(axis = var_123, x = aw_chunk_129_cast)[name = tensor<string, []>("op_1335_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1336_cast = softmax(axis = var_123, x = aw_chunk_131_cast)[name = tensor<string, []>("op_1336_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1337_cast = softmax(axis = var_123, x = aw_chunk_133_cast)[name = tensor<string, []>("op_1337_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1338_cast = softmax(axis = var_123, x = aw_chunk_135_cast)[name = tensor<string, []>("op_1338_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1339_cast = softmax(axis = var_123, x = aw_chunk_137_cast)[name = tensor<string, []>("op_1339_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1340_cast = softmax(axis = var_123, x = aw_chunk_139_cast)[name = tensor<string, []>("op_1340_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1341_cast = softmax(axis = var_123, x = aw_chunk_141_cast)[name = tensor<string, []>("op_1341_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1342_cast = softmax(axis = var_123, x = aw_chunk_143_cast)[name = tensor<string, []>("op_1342_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1343_cast = softmax(axis = var_123, x = aw_chunk_145_cast)[name = tensor<string, []>("op_1343_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1344_cast = softmax(axis = var_123, x = aw_chunk_147_cast)[name = tensor<string, []>("op_1344_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1345_cast = softmax(axis = var_123, x = aw_chunk_149_cast)[name = tensor<string, []>("op_1345_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1346_cast = softmax(axis = var_123, x = aw_chunk_151_cast)[name = tensor<string, []>("op_1346_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1347_cast = softmax(axis = var_123, x = aw_chunk_153_cast)[name = tensor<string, []>("op_1347_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1348_cast = softmax(axis = var_123, x = aw_chunk_155_cast)[name = tensor<string, []>("op_1348_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1349_cast = softmax(axis = var_123, x = aw_chunk_157_cast)[name = tensor<string, []>("op_1349_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1350_cast = softmax(axis = var_123, x = aw_chunk_159_cast)[name = tensor<string, []>("op_1350_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1351_cast = softmax(axis = var_123, x = aw_chunk_161_cast)[name = tensor<string, []>("op_1351_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1352_cast = softmax(axis = var_123, x = aw_chunk_163_cast)[name = tensor<string, []>("op_1352_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1353_cast = softmax(axis = var_123, x = aw_chunk_165_cast)[name = tensor<string, []>("op_1353_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1354_cast = softmax(axis = var_123, x = aw_chunk_167_cast)[name = tensor<string, []>("op_1354_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1355_cast = softmax(axis = var_123, x = aw_chunk_169_cast)[name = tensor<string, []>("op_1355_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1356_cast = softmax(axis = var_123, x = aw_chunk_171_cast)[name = tensor<string, []>("op_1356_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1357_cast = softmax(axis = var_123, x = aw_chunk_173_cast)[name = tensor<string, []>("op_1357_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1358_cast = softmax(axis = var_123, x = aw_chunk_175_cast)[name = tensor<string, []>("op_1358_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1359_cast = softmax(axis = var_123, x = aw_chunk_177_cast)[name = tensor<string, []>("op_1359_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1360_cast = softmax(axis = var_123, x = aw_chunk_179_cast)[name = tensor<string, []>("op_1360_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1361_cast = softmax(axis = var_123, x = aw_chunk_181_cast)[name = tensor<string, []>("op_1361_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1362_cast = softmax(axis = var_123, x = aw_chunk_183_cast)[name = tensor<string, []>("op_1362_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1363_cast = softmax(axis = var_123, x = aw_chunk_185_cast)[name = tensor<string, []>("op_1363_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1364_cast = softmax(axis = var_123, x = aw_chunk_187_cast)[name = tensor<string, []>("op_1364_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1365_cast = softmax(axis = var_123, x = aw_chunk_189_cast)[name = tensor<string, []>("op_1365_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1366_cast = softmax(axis = var_123, x = aw_chunk_191_cast)[name = tensor<string, []>("op_1366_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1367_cast = softmax(axis = var_123, x = aw_chunk_193_cast)[name = tensor<string, []>("op_1367_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1368_cast = softmax(axis = var_123, x = aw_chunk_195_cast)[name = tensor<string, []>("op_1368_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1369_cast = softmax(axis = var_123, x = aw_chunk_197_cast)[name = tensor<string, []>("op_1369_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1370_cast = softmax(axis = var_123, x = aw_chunk_199_cast)[name = tensor<string, []>("op_1370_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1371_cast = softmax(axis = var_123, x = aw_chunk_201_cast)[name = tensor<string, []>("op_1371_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1372_cast = softmax(axis = var_123, x = aw_chunk_203_cast)[name = tensor<string, []>("op_1372_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1373_cast = softmax(axis = var_123, x = aw_chunk_205_cast)[name = tensor<string, []>("op_1373_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1374_cast = softmax(axis = var_123, x = aw_chunk_207_cast)[name = tensor<string, []>("op_1374_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1375_cast = softmax(axis = var_123, x = aw_chunk_209_cast)[name = tensor<string, []>("op_1375_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1376_cast = softmax(axis = var_123, x = aw_chunk_211_cast)[name = tensor<string, []>("op_1376_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1377_cast = softmax(axis = var_123, x = aw_chunk_213_cast)[name = tensor<string, []>("op_1377_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1378_cast = softmax(axis = var_123, x = aw_chunk_215_cast)[name = tensor<string, []>("op_1378_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1379_cast = softmax(axis = var_123, x = aw_chunk_217_cast)[name = tensor<string, []>("op_1379_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1380_cast = softmax(axis = var_123, x = aw_chunk_219_cast)[name = tensor<string, []>("op_1380_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1381_cast = softmax(axis = var_123, x = aw_chunk_221_cast)[name = tensor<string, []>("op_1381_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1382_cast = softmax(axis = var_123, x = aw_chunk_223_cast)[name = tensor<string, []>("op_1382_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1383_cast = softmax(axis = var_123, x = aw_chunk_225_cast)[name = tensor<string, []>("op_1383_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1384_cast = softmax(axis = var_123, x = aw_chunk_227_cast)[name = tensor<string, []>("op_1384_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1385_cast = softmax(axis = var_123, x = aw_chunk_229_cast)[name = tensor<string, []>("op_1385_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1386_cast = softmax(axis = var_123, x = aw_chunk_231_cast)[name = tensor<string, []>("op_1386_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1387_cast = softmax(axis = var_123, x = aw_chunk_233_cast)[name = tensor<string, []>("op_1387_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1388_cast = softmax(axis = var_123, x = aw_chunk_235_cast)[name = tensor<string, []>("op_1388_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1389_cast = softmax(axis = var_123, x = aw_chunk_237_cast)[name = tensor<string, []>("op_1389_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1390_cast = softmax(axis = var_123, x = aw_chunk_239_cast)[name = tensor<string, []>("op_1390_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1391_cast = softmax(axis = var_123, x = aw_chunk_241_cast)[name = tensor<string, []>("op_1391_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1392_cast = softmax(axis = var_123, x = aw_chunk_243_cast)[name = tensor<string, []>("op_1392_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1393_cast = softmax(axis = var_123, x = aw_chunk_245_cast)[name = tensor<string, []>("op_1393_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1394_cast = softmax(axis = var_123, x = aw_chunk_247_cast)[name = tensor<string, []>("op_1394_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1395_cast = softmax(axis = var_123, x = aw_chunk_249_cast)[name = tensor<string, []>("op_1395_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1396_cast = softmax(axis = var_123, x = aw_chunk_251_cast)[name = tensor<string, []>("op_1396_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1397_cast = softmax(axis = var_123, x = aw_chunk_253_cast)[name = tensor<string, []>("op_1397_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1398_cast = softmax(axis = var_123, x = aw_chunk_255_cast)[name = tensor<string, []>("op_1398_cast")];
            tensor<string, []> var_1400_equation_0 = const()[name = tensor<string, []>("op_1400_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1400_cast = einsum(equation = var_1400_equation_0, values = (var_1048_cast, var_1335_cast))[name = tensor<string, []>("op_1400_cast")];
            tensor<string, []> var_1402_equation_0 = const()[name = tensor<string, []>("op_1402_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1402_cast = einsum(equation = var_1402_equation_0, values = (var_1048_cast, var_1336_cast))[name = tensor<string, []>("op_1402_cast")];
            tensor<string, []> var_1404_equation_0 = const()[name = tensor<string, []>("op_1404_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1404_cast = einsum(equation = var_1404_equation_0, values = (var_1048_cast, var_1337_cast))[name = tensor<string, []>("op_1404_cast")];
            tensor<string, []> var_1406_equation_0 = const()[name = tensor<string, []>("op_1406_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1406_cast = einsum(equation = var_1406_equation_0, values = (var_1048_cast, var_1338_cast))[name = tensor<string, []>("op_1406_cast")];
            tensor<string, []> var_1408_equation_0 = const()[name = tensor<string, []>("op_1408_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1408_cast = einsum(equation = var_1408_equation_0, values = (var_1048_cast, var_1339_cast))[name = tensor<string, []>("op_1408_cast")];
            tensor<string, []> var_1410_equation_0 = const()[name = tensor<string, []>("op_1410_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1410_cast = einsum(equation = var_1410_equation_0, values = (var_1048_cast, var_1340_cast))[name = tensor<string, []>("op_1410_cast")];
            tensor<string, []> var_1412_equation_0 = const()[name = tensor<string, []>("op_1412_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1412_cast = einsum(equation = var_1412_equation_0, values = (var_1048_cast, var_1341_cast))[name = tensor<string, []>("op_1412_cast")];
            tensor<string, []> var_1414_equation_0 = const()[name = tensor<string, []>("op_1414_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1414_cast = einsum(equation = var_1414_equation_0, values = (var_1048_cast, var_1342_cast))[name = tensor<string, []>("op_1414_cast")];
            tensor<string, []> var_1416_equation_0 = const()[name = tensor<string, []>("op_1416_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1416_cast = einsum(equation = var_1416_equation_0, values = (var_1052_cast, var_1343_cast))[name = tensor<string, []>("op_1416_cast")];
            tensor<string, []> var_1418_equation_0 = const()[name = tensor<string, []>("op_1418_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1418_cast = einsum(equation = var_1418_equation_0, values = (var_1052_cast, var_1344_cast))[name = tensor<string, []>("op_1418_cast")];
            tensor<string, []> var_1420_equation_0 = const()[name = tensor<string, []>("op_1420_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1420_cast = einsum(equation = var_1420_equation_0, values = (var_1052_cast, var_1345_cast))[name = tensor<string, []>("op_1420_cast")];
            tensor<string, []> var_1422_equation_0 = const()[name = tensor<string, []>("op_1422_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1422_cast = einsum(equation = var_1422_equation_0, values = (var_1052_cast, var_1346_cast))[name = tensor<string, []>("op_1422_cast")];
            tensor<string, []> var_1424_equation_0 = const()[name = tensor<string, []>("op_1424_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1424_cast = einsum(equation = var_1424_equation_0, values = (var_1052_cast, var_1347_cast))[name = tensor<string, []>("op_1424_cast")];
            tensor<string, []> var_1426_equation_0 = const()[name = tensor<string, []>("op_1426_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1426_cast = einsum(equation = var_1426_equation_0, values = (var_1052_cast, var_1348_cast))[name = tensor<string, []>("op_1426_cast")];
            tensor<string, []> var_1428_equation_0 = const()[name = tensor<string, []>("op_1428_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1428_cast = einsum(equation = var_1428_equation_0, values = (var_1052_cast, var_1349_cast))[name = tensor<string, []>("op_1428_cast")];
            tensor<string, []> var_1430_equation_0 = const()[name = tensor<string, []>("op_1430_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1430_cast = einsum(equation = var_1430_equation_0, values = (var_1052_cast, var_1350_cast))[name = tensor<string, []>("op_1430_cast")];
            tensor<string, []> var_1432_equation_0 = const()[name = tensor<string, []>("op_1432_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1432_cast = einsum(equation = var_1432_equation_0, values = (var_1056_cast, var_1351_cast))[name = tensor<string, []>("op_1432_cast")];
            tensor<string, []> var_1434_equation_0 = const()[name = tensor<string, []>("op_1434_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1434_cast = einsum(equation = var_1434_equation_0, values = (var_1056_cast, var_1352_cast))[name = tensor<string, []>("op_1434_cast")];
            tensor<string, []> var_1436_equation_0 = const()[name = tensor<string, []>("op_1436_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1436_cast = einsum(equation = var_1436_equation_0, values = (var_1056_cast, var_1353_cast))[name = tensor<string, []>("op_1436_cast")];
            tensor<string, []> var_1438_equation_0 = const()[name = tensor<string, []>("op_1438_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1438_cast = einsum(equation = var_1438_equation_0, values = (var_1056_cast, var_1354_cast))[name = tensor<string, []>("op_1438_cast")];
            tensor<string, []> var_1440_equation_0 = const()[name = tensor<string, []>("op_1440_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1440_cast = einsum(equation = var_1440_equation_0, values = (var_1056_cast, var_1355_cast))[name = tensor<string, []>("op_1440_cast")];
            tensor<string, []> var_1442_equation_0 = const()[name = tensor<string, []>("op_1442_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1442_cast = einsum(equation = var_1442_equation_0, values = (var_1056_cast, var_1356_cast))[name = tensor<string, []>("op_1442_cast")];
            tensor<string, []> var_1444_equation_0 = const()[name = tensor<string, []>("op_1444_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1444_cast = einsum(equation = var_1444_equation_0, values = (var_1056_cast, var_1357_cast))[name = tensor<string, []>("op_1444_cast")];
            tensor<string, []> var_1446_equation_0 = const()[name = tensor<string, []>("op_1446_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1446_cast = einsum(equation = var_1446_equation_0, values = (var_1056_cast, var_1358_cast))[name = tensor<string, []>("op_1446_cast")];
            tensor<string, []> var_1448_equation_0 = const()[name = tensor<string, []>("op_1448_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1448_cast = einsum(equation = var_1448_equation_0, values = (var_1060_cast, var_1359_cast))[name = tensor<string, []>("op_1448_cast")];
            tensor<string, []> var_1450_equation_0 = const()[name = tensor<string, []>("op_1450_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1450_cast = einsum(equation = var_1450_equation_0, values = (var_1060_cast, var_1360_cast))[name = tensor<string, []>("op_1450_cast")];
            tensor<string, []> var_1452_equation_0 = const()[name = tensor<string, []>("op_1452_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1452_cast = einsum(equation = var_1452_equation_0, values = (var_1060_cast, var_1361_cast))[name = tensor<string, []>("op_1452_cast")];
            tensor<string, []> var_1454_equation_0 = const()[name = tensor<string, []>("op_1454_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1454_cast = einsum(equation = var_1454_equation_0, values = (var_1060_cast, var_1362_cast))[name = tensor<string, []>("op_1454_cast")];
            tensor<string, []> var_1456_equation_0 = const()[name = tensor<string, []>("op_1456_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1456_cast = einsum(equation = var_1456_equation_0, values = (var_1060_cast, var_1363_cast))[name = tensor<string, []>("op_1456_cast")];
            tensor<string, []> var_1458_equation_0 = const()[name = tensor<string, []>("op_1458_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1458_cast = einsum(equation = var_1458_equation_0, values = (var_1060_cast, var_1364_cast))[name = tensor<string, []>("op_1458_cast")];
            tensor<string, []> var_1460_equation_0 = const()[name = tensor<string, []>("op_1460_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1460_cast = einsum(equation = var_1460_equation_0, values = (var_1060_cast, var_1365_cast))[name = tensor<string, []>("op_1460_cast")];
            tensor<string, []> var_1462_equation_0 = const()[name = tensor<string, []>("op_1462_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1462_cast = einsum(equation = var_1462_equation_0, values = (var_1060_cast, var_1366_cast))[name = tensor<string, []>("op_1462_cast")];
            tensor<string, []> var_1464_equation_0 = const()[name = tensor<string, []>("op_1464_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1464_cast = einsum(equation = var_1464_equation_0, values = (var_1064_cast, var_1367_cast))[name = tensor<string, []>("op_1464_cast")];
            tensor<string, []> var_1466_equation_0 = const()[name = tensor<string, []>("op_1466_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1466_cast = einsum(equation = var_1466_equation_0, values = (var_1064_cast, var_1368_cast))[name = tensor<string, []>("op_1466_cast")];
            tensor<string, []> var_1468_equation_0 = const()[name = tensor<string, []>("op_1468_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1468_cast = einsum(equation = var_1468_equation_0, values = (var_1064_cast, var_1369_cast))[name = tensor<string, []>("op_1468_cast")];
            tensor<string, []> var_1470_equation_0 = const()[name = tensor<string, []>("op_1470_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1470_cast = einsum(equation = var_1470_equation_0, values = (var_1064_cast, var_1370_cast))[name = tensor<string, []>("op_1470_cast")];
            tensor<string, []> var_1472_equation_0 = const()[name = tensor<string, []>("op_1472_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1472_cast = einsum(equation = var_1472_equation_0, values = (var_1064_cast, var_1371_cast))[name = tensor<string, []>("op_1472_cast")];
            tensor<string, []> var_1474_equation_0 = const()[name = tensor<string, []>("op_1474_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1474_cast = einsum(equation = var_1474_equation_0, values = (var_1064_cast, var_1372_cast))[name = tensor<string, []>("op_1474_cast")];
            tensor<string, []> var_1476_equation_0 = const()[name = tensor<string, []>("op_1476_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1476_cast = einsum(equation = var_1476_equation_0, values = (var_1064_cast, var_1373_cast))[name = tensor<string, []>("op_1476_cast")];
            tensor<string, []> var_1478_equation_0 = const()[name = tensor<string, []>("op_1478_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1478_cast = einsum(equation = var_1478_equation_0, values = (var_1064_cast, var_1374_cast))[name = tensor<string, []>("op_1478_cast")];
            tensor<string, []> var_1480_equation_0 = const()[name = tensor<string, []>("op_1480_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1480_cast = einsum(equation = var_1480_equation_0, values = (var_1068_cast, var_1375_cast))[name = tensor<string, []>("op_1480_cast")];
            tensor<string, []> var_1482_equation_0 = const()[name = tensor<string, []>("op_1482_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1482_cast = einsum(equation = var_1482_equation_0, values = (var_1068_cast, var_1376_cast))[name = tensor<string, []>("op_1482_cast")];
            tensor<string, []> var_1484_equation_0 = const()[name = tensor<string, []>("op_1484_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1484_cast = einsum(equation = var_1484_equation_0, values = (var_1068_cast, var_1377_cast))[name = tensor<string, []>("op_1484_cast")];
            tensor<string, []> var_1486_equation_0 = const()[name = tensor<string, []>("op_1486_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1486_cast = einsum(equation = var_1486_equation_0, values = (var_1068_cast, var_1378_cast))[name = tensor<string, []>("op_1486_cast")];
            tensor<string, []> var_1488_equation_0 = const()[name = tensor<string, []>("op_1488_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1488_cast = einsum(equation = var_1488_equation_0, values = (var_1068_cast, var_1379_cast))[name = tensor<string, []>("op_1488_cast")];
            tensor<string, []> var_1490_equation_0 = const()[name = tensor<string, []>("op_1490_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1490_cast = einsum(equation = var_1490_equation_0, values = (var_1068_cast, var_1380_cast))[name = tensor<string, []>("op_1490_cast")];
            tensor<string, []> var_1492_equation_0 = const()[name = tensor<string, []>("op_1492_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1492_cast = einsum(equation = var_1492_equation_0, values = (var_1068_cast, var_1381_cast))[name = tensor<string, []>("op_1492_cast")];
            tensor<string, []> var_1494_equation_0 = const()[name = tensor<string, []>("op_1494_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1494_cast = einsum(equation = var_1494_equation_0, values = (var_1068_cast, var_1382_cast))[name = tensor<string, []>("op_1494_cast")];
            tensor<string, []> var_1496_equation_0 = const()[name = tensor<string, []>("op_1496_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1496_cast = einsum(equation = var_1496_equation_0, values = (var_1072_cast, var_1383_cast))[name = tensor<string, []>("op_1496_cast")];
            tensor<string, []> var_1498_equation_0 = const()[name = tensor<string, []>("op_1498_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1498_cast = einsum(equation = var_1498_equation_0, values = (var_1072_cast, var_1384_cast))[name = tensor<string, []>("op_1498_cast")];
            tensor<string, []> var_1500_equation_0 = const()[name = tensor<string, []>("op_1500_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1500_cast = einsum(equation = var_1500_equation_0, values = (var_1072_cast, var_1385_cast))[name = tensor<string, []>("op_1500_cast")];
            tensor<string, []> var_1502_equation_0 = const()[name = tensor<string, []>("op_1502_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1502_cast = einsum(equation = var_1502_equation_0, values = (var_1072_cast, var_1386_cast))[name = tensor<string, []>("op_1502_cast")];
            tensor<string, []> var_1504_equation_0 = const()[name = tensor<string, []>("op_1504_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1504_cast = einsum(equation = var_1504_equation_0, values = (var_1072_cast, var_1387_cast))[name = tensor<string, []>("op_1504_cast")];
            tensor<string, []> var_1506_equation_0 = const()[name = tensor<string, []>("op_1506_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1506_cast = einsum(equation = var_1506_equation_0, values = (var_1072_cast, var_1388_cast))[name = tensor<string, []>("op_1506_cast")];
            tensor<string, []> var_1508_equation_0 = const()[name = tensor<string, []>("op_1508_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1508_cast = einsum(equation = var_1508_equation_0, values = (var_1072_cast, var_1389_cast))[name = tensor<string, []>("op_1508_cast")];
            tensor<string, []> var_1510_equation_0 = const()[name = tensor<string, []>("op_1510_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1510_cast = einsum(equation = var_1510_equation_0, values = (var_1072_cast, var_1390_cast))[name = tensor<string, []>("op_1510_cast")];
            tensor<string, []> var_1512_equation_0 = const()[name = tensor<string, []>("op_1512_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1512_cast = einsum(equation = var_1512_equation_0, values = (var_1076_cast, var_1391_cast))[name = tensor<string, []>("op_1512_cast")];
            tensor<string, []> var_1514_equation_0 = const()[name = tensor<string, []>("op_1514_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1514_cast = einsum(equation = var_1514_equation_0, values = (var_1076_cast, var_1392_cast))[name = tensor<string, []>("op_1514_cast")];
            tensor<string, []> var_1516_equation_0 = const()[name = tensor<string, []>("op_1516_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1516_cast = einsum(equation = var_1516_equation_0, values = (var_1076_cast, var_1393_cast))[name = tensor<string, []>("op_1516_cast")];
            tensor<string, []> var_1518_equation_0 = const()[name = tensor<string, []>("op_1518_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1518_cast = einsum(equation = var_1518_equation_0, values = (var_1076_cast, var_1394_cast))[name = tensor<string, []>("op_1518_cast")];
            tensor<string, []> var_1520_equation_0 = const()[name = tensor<string, []>("op_1520_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1520_cast = einsum(equation = var_1520_equation_0, values = (var_1076_cast, var_1395_cast))[name = tensor<string, []>("op_1520_cast")];
            tensor<string, []> var_1522_equation_0 = const()[name = tensor<string, []>("op_1522_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1522_cast = einsum(equation = var_1522_equation_0, values = (var_1076_cast, var_1396_cast))[name = tensor<string, []>("op_1522_cast")];
            tensor<string, []> var_1524_equation_0 = const()[name = tensor<string, []>("op_1524_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1524_cast = einsum(equation = var_1524_equation_0, values = (var_1076_cast, var_1397_cast))[name = tensor<string, []>("op_1524_cast")];
            tensor<string, []> var_1526_equation_0 = const()[name = tensor<string, []>("op_1526_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_1526_cast = einsum(equation = var_1526_equation_0, values = (var_1076_cast, var_1398_cast))[name = tensor<string, []>("op_1526_cast")];
            tensor<bool, []> var_1528_interleave_0 = const()[name = tensor<string, []>("op_1528_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_1528_cast = concat(axis = var_95, interleave = var_1528_interleave_0, values = (var_1400_cast, var_1402_cast, var_1404_cast, var_1406_cast, var_1408_cast, var_1410_cast, var_1412_cast, var_1414_cast))[name = tensor<string, []>("op_1528_cast")];
            tensor<bool, []> var_1530_interleave_0 = const()[name = tensor<string, []>("op_1530_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_1530_cast = concat(axis = var_95, interleave = var_1530_interleave_0, values = (var_1416_cast, var_1418_cast, var_1420_cast, var_1422_cast, var_1424_cast, var_1426_cast, var_1428_cast, var_1430_cast))[name = tensor<string, []>("op_1530_cast")];
            tensor<bool, []> var_1532_interleave_0 = const()[name = tensor<string, []>("op_1532_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_1532_cast = concat(axis = var_95, interleave = var_1532_interleave_0, values = (var_1432_cast, var_1434_cast, var_1436_cast, var_1438_cast, var_1440_cast, var_1442_cast, var_1444_cast, var_1446_cast))[name = tensor<string, []>("op_1532_cast")];
            tensor<bool, []> var_1534_interleave_0 = const()[name = tensor<string, []>("op_1534_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_1534_cast = concat(axis = var_95, interleave = var_1534_interleave_0, values = (var_1448_cast, var_1450_cast, var_1452_cast, var_1454_cast, var_1456_cast, var_1458_cast, var_1460_cast, var_1462_cast))[name = tensor<string, []>("op_1534_cast")];
            tensor<bool, []> var_1536_interleave_0 = const()[name = tensor<string, []>("op_1536_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_1536_cast = concat(axis = var_95, interleave = var_1536_interleave_0, values = (var_1464_cast, var_1466_cast, var_1468_cast, var_1470_cast, var_1472_cast, var_1474_cast, var_1476_cast, var_1478_cast))[name = tensor<string, []>("op_1536_cast")];
            tensor<bool, []> var_1538_interleave_0 = const()[name = tensor<string, []>("op_1538_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_1538_cast = concat(axis = var_95, interleave = var_1538_interleave_0, values = (var_1480_cast, var_1482_cast, var_1484_cast, var_1486_cast, var_1488_cast, var_1490_cast, var_1492_cast, var_1494_cast))[name = tensor<string, []>("op_1538_cast")];
            tensor<bool, []> var_1540_interleave_0 = const()[name = tensor<string, []>("op_1540_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_1540_cast = concat(axis = var_95, interleave = var_1540_interleave_0, values = (var_1496_cast, var_1498_cast, var_1500_cast, var_1502_cast, var_1504_cast, var_1506_cast, var_1508_cast, var_1510_cast))[name = tensor<string, []>("op_1540_cast")];
            tensor<bool, []> var_1542_interleave_0 = const()[name = tensor<string, []>("op_1542_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_1542_cast = concat(axis = var_95, interleave = var_1542_interleave_0, values = (var_1512_cast, var_1514_cast, var_1516_cast, var_1518_cast, var_1520_cast, var_1522_cast, var_1524_cast, var_1526_cast))[name = tensor<string, []>("op_1542_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_123, interleave = input_27_interleave_0, values = (var_1528_cast, var_1530_cast, var_1532_cast, var_1534_cast, var_1536_cast, var_1538_cast, var_1540_cast, var_1542_cast))[name = tensor<string, []>("input_27_cast")];
            tensor<int32, [2]> var_1548 = const()[name = tensor<string, []>("op_1548"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1550 = const()[name = tensor<string, []>("op_1550"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_1552_pad_type_0 = const()[name = tensor<string, []>("op_1552_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_1552_pad_0 = const()[name = tensor<string, []>("op_1552_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4085632))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4162496))), name = tensor<string, []>("down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            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, []>(4162688)))];
            tensor<fp16, [2, 320, 1, 4096]> var_1552_cast = conv(bias = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_1550, groups = var_123, pad = var_1552_pad_0, pad_type = var_1552_pad_type_0, strides = var_1548, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_27_cast)[name = tensor<string, []>("op_1552_cast")];
            tensor<fp16, [2, 320, 1, 4096]> inputs_5_cast = add(x = var_1552_cast, y = inputs_3_cast)[name = tensor<string, []>("inputs_5_cast")];
            tensor<int32, [1]> var_1556 = const()[name = tensor<string, []>("op_1556"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_5_cast = reduce_mean(axes = var_1556, keep_dims = var_118, x = inputs_5_cast)[name = tensor<string, []>("channels_mean_5_cast")];
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_5_cast = sub(x = inputs_5_cast, y = channels_mean_5_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_5_cast")];
            tensor<int32, [1]> var_1560 = const()[name = tensor<string, []>("op_1560"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_1561_cast = reduce_mean(axes = var_1560, keep_dims = var_118, x = zero_mean_sq_5_cast)[name = tensor<string, []>("op_1561_cast")];
            tensor<fp16, []> var_1562_to_fp16 = const()[name = tensor<string, []>("op_1562_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_1563_cast = add(x = var_1561_cast, y = var_1562_to_fp16)[name = tensor<string, []>("op_1563_cast")];
            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_1563_cast)[name = tensor<string, []>("denom_5_cast")];
            tensor<fp16, [2, 320, 1, 4096]> out_5_cast = mul(x = zero_mean_5_cast, y = denom_5_cast)[name = tensor<string, []>("out_5_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_1567_to_fp16 = const()[name = tensor<string, []>("op_1567_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4163392)))];
            tensor<fp16, [2, 320, 1, 4096]> var_1568_cast = add(x = out_5_cast, y = var_1567_to_fp16)[name = tensor<string, []>("op_1568_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_1570_to_fp16 = const()[name = tensor<string, []>("op_1570_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4164096)))];
            tensor<fp16, [2, 320, 1, 4096]> input_29_cast = mul(x = var_1568_cast, y = var_1570_to_fp16)[name = tensor<string, []>("input_29_cast")];
            tensor<int32, [2]> var_1578 = const()[name = tensor<string, []>("op_1578"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1580 = const()[name = tensor<string, []>("op_1580"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_1582_pad_type_0 = const()[name = tensor<string, []>("op_1582_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_1582_pad_0 = const()[name = tensor<string, []>("op_1582_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [614400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4164800))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4779264))), name = tensor<string, []>("down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([2560, 320, 1, 1])];
            tensor<fp16, [2560]> down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4779456))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4781440))), name = tensor<string, []>("down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([2560])];
            tensor<fp16, [2, 2560, 1, 4096]> var_1582_cast = conv(bias = down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_1580, groups = var_123, pad = var_1582_pad_0, pad_type = var_1582_pad_type_0, strides = var_1578, weight = down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_29_cast)[name = tensor<string, []>("op_1582_cast")];
            tensor<int32, [2]> var_1583_split_sizes_0 = const()[name = tensor<string, []>("op_1583_split_sizes_0"), val = tensor<int32, [2]>([1280, 1280])];
            tensor<int32, []> var_1583_axis_0 = const()[name = tensor<string, []>("op_1583_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 1280, 1, 4096]> var_1583_cast_0, tensor<fp16, [2, 1280, 1, 4096]> var_1583_cast_1 = split(axis = var_1583_axis_0, split_sizes = var_1583_split_sizes_0, x = var_1582_cast)[name = tensor<string, []>("op_1583_cast")];
            tensor<string, []> var_1585_mode_0 = const()[name = tensor<string, []>("op_1585_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 1280, 1, 4096]> var_1585_cast = gelu(mode = var_1585_mode_0, x = var_1583_cast_1)[name = tensor<string, []>("op_1585_cast")];
            tensor<fp16, [2, 1280, 1, 4096]> input_31_cast = mul(x = var_1583_cast_0, y = var_1585_cast)[name = tensor<string, []>("input_31_cast")];
            tensor<int32, [2]> var_1589 = const()[name = tensor<string, []>("op_1589"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1591 = const()[name = tensor<string, []>("op_1591"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_1593_pad_type_0 = const()[name = tensor<string, []>("op_1593_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_1593_pad_0 = const()[name = tensor<string, []>("op_1593_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4781632))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5088896))), name = tensor<string, []>("down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 1280, 1, 1])];
            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, []>(5089088)))];
            tensor<fp16, [2, 320, 1, 4096]> var_1593_cast = conv(bias = down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_1591, groups = var_123, pad = var_1593_pad_0, pad_type = var_1593_pad_type_0, strides = var_1589, weight = down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_31_cast)[name = tensor<string, []>("op_1593_cast")];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_15_cast = add(x = var_1593_cast, y = inputs_5_cast)[name = tensor<string, []>("hidden_states_15_cast")];
            tensor<int32, [4]> var_1595 = const()[name = tensor<string, []>("op_1595"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> input_33_cast = reshape(shape = var_1595, x = hidden_states_15_cast)[name = tensor<string, []>("input_33_cast")];
            tensor<int32, [2]> var_1599 = const()[name = tensor<string, []>("op_1599"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1601 = const()[name = tensor<string, []>("op_1601"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5089792))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5166656))), name = tensor<string, []>("down_blocks_0_attentions_0_proj_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            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, []>(5166848)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_17_cast = conv(bias = down_blocks_0_attentions_0_proj_out_bias_to_fp16, dilations = var_1601, groups = var_123, pad = hidden_states_17_pad_0, pad_type = hidden_states_17_pad_type_0, strides = var_1599, weight = down_blocks_0_attentions_0_proj_out_weight_to_fp16_palettized, x = input_33_cast)[name = tensor<string, []>("hidden_states_17_cast")];
            tensor<fp16, [2, 320, 64, 64]> input_35_cast = add(x = hidden_states_17_cast, y = hidden_states_5_cast)[name = tensor<string, []>("input_35_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)[name = tensor<string, []>("reshape_12_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)[name = tensor<string, []>("reduce_mean_9_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> sub_6_cast = sub(x = reshape_12_cast, y = reduce_mean_9_cast)[name = tensor<string, []>("sub_6_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> square_3_cast = square(x = sub_6_cast)[name = tensor<string, []>("square_3_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)[name = tensor<string, []>("reduce_mean_11_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)[name = tensor<string, []>("add_6_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_3_cast = sqrt(x = add_6_cast)[name = tensor<string, []>("sqrt_3_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> real_div_3_cast = real_div(x = sub_6_cast, y = sqrt_3_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_13_cast")];
            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, []>(5167552)))];
            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, []>(5168256)))];
            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_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_13_cast)[name = tensor<string, []>("add_7_cast")];
            tensor<fp16, [2, 320, 64, 64]> input_39_cast = silu(x = add_7_cast)[name = tensor<string, []>("input_39_cast")];
            tensor<int32, [2]> var_1616 = const()[name = tensor<string, []>("op_1616"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1618 = const()[name = tensor<string, []>("op_1618"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [691200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5168960))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5860224))), name = tensor<string, []>("down_blocks_0_resnets_1_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 3, 3])];
            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, []>(5860416)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_19_cast = conv(bias = down_blocks_0_resnets_1_conv1_bias_to_fp16, dilations = var_1618, groups = var_123, pad = hidden_states_19_pad_0, pad_type = hidden_states_19_pad_type_0, strides = var_1616, weight = down_blocks_0_resnets_1_conv1_weight_to_fp16_palettized, x = input_39_cast)[name = tensor<string, []>("hidden_states_19_cast")];
            tensor<int32, [2]> var_1624 = const()[name = tensor<string, []>("op_1624"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1626 = const()[name = tensor<string, []>("op_1626"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5861120))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6168384))), name = tensor<string, []>("down_blocks_0_resnets_1_time_emb_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 1280, 1, 1])];
            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, []>(6168576)))];
            tensor<fp16, [2, 320, 1, 1]> temb_3_cast = conv(bias = down_blocks_0_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_1626, groups = var_123, pad = temb_3_pad_0, pad_type = temb_3_pad_type_0, strides = var_1624, weight = down_blocks_0_resnets_1_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor<string, []>("temb_3_cast")];
            tensor<fp16, [2, 320, 64, 64]> input_43_cast = add(x = hidden_states_19_cast, y = temb_3_cast)[name = tensor<string, []>("input_43_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)[name = tensor<string, []>("reshape_16_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)[name = tensor<string, []>("reduce_mean_12_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> sub_8_cast = sub(x = reshape_16_cast, y = reduce_mean_12_cast)[name = tensor<string, []>("sub_8_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> square_4_cast = square(x = sub_8_cast)[name = tensor<string, []>("square_4_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)[name = tensor<string, []>("reduce_mean_14_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)[name = tensor<string, []>("add_8_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_4_cast = sqrt(x = add_8_cast)[name = tensor<string, []>("sqrt_4_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> real_div_4_cast = real_div(x = sub_8_cast, y = sqrt_4_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_17_cast")];
            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, []>(6169280)))];
            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, []>(6169984)))];
            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_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_17_cast)[name = tensor<string, []>("add_9_cast")];
            tensor<fp16, [2, 320, 64, 64]> input_47_cast = silu(x = add_9_cast)[name = tensor<string, []>("input_47_cast")];
            tensor<int32, [2]> var_1636 = const()[name = tensor<string, []>("op_1636"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1638 = const()[name = tensor<string, []>("op_1638"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [691200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6170688))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6861952))), name = tensor<string, []>("down_blocks_0_resnets_1_conv2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 3, 3])];
            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, []>(6862144)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_21_cast = conv(bias = down_blocks_0_resnets_1_conv2_bias_to_fp16, dilations = var_1638, groups = var_123, pad = hidden_states_21_pad_0, pad_type = hidden_states_21_pad_type_0, strides = var_1636, weight = down_blocks_0_resnets_1_conv2_weight_to_fp16_palettized, x = input_47_cast)[name = tensor<string, []>("hidden_states_21_cast")];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_23_cast = add(x = input_35_cast, y = hidden_states_21_cast)[name = tensor<string, []>("hidden_states_23_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)[name = tensor<string, []>("reshape_20_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)[name = tensor<string, []>("reduce_mean_15_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> sub_10_cast = sub(x = reshape_20_cast, y = reduce_mean_15_cast)[name = tensor<string, []>("sub_10_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> square_5_cast = square(x = sub_10_cast)[name = tensor<string, []>("square_5_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)[name = tensor<string, []>("reduce_mean_17_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)[name = tensor<string, []>("add_10_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_5_cast = sqrt(x = add_10_cast)[name = tensor<string, []>("sqrt_5_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> real_div_5_cast = real_div(x = sub_10_cast, y = sqrt_5_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_21_cast")];
            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, []>(6862848)))];
            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, []>(6863552)))];
            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_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_21_cast)[name = tensor<string, []>("add_11_cast")];
            tensor<int32, [2]> var_1658 = const()[name = tensor<string, []>("op_1658"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1660 = const()[name = tensor<string, []>("op_1660"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6864256))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6941120))), name = tensor<string, []>("down_blocks_0_attentions_1_proj_in_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            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, []>(6941312)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_25_cast = conv(bias = down_blocks_0_attentions_1_proj_in_bias_to_fp16, dilations = var_1660, groups = var_123, pad = hidden_states_25_pad_0, pad_type = hidden_states_25_pad_type_0, strides = var_1658, weight = down_blocks_0_attentions_1_proj_in_weight_to_fp16_palettized, x = add_11_cast)[name = tensor<string, []>("hidden_states_25_cast")];
            tensor<int32, [4]> var_1665 = const()[name = tensor<string, []>("op_1665"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<fp16, [2, 320, 1, 4096]> inputs_7_cast = reshape(shape = var_1665, x = hidden_states_25_cast)[name = tensor<string, []>("inputs_7_cast")];
            tensor<int32, [1]> var_1675 = const()[name = tensor<string, []>("op_1675"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_7_cast = reduce_mean(axes = var_1675, keep_dims = var_118, x = inputs_7_cast)[name = tensor<string, []>("channels_mean_7_cast")];
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_7_cast = sub(x = inputs_7_cast, y = channels_mean_7_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_7_cast")];
            tensor<int32, [1]> var_1679 = const()[name = tensor<string, []>("op_1679"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_1680_cast = reduce_mean(axes = var_1679, keep_dims = var_118, x = zero_mean_sq_7_cast)[name = tensor<string, []>("op_1680_cast")];
            tensor<fp16, []> var_1681_to_fp16 = const()[name = tensor<string, []>("op_1681_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_1682_cast = add(x = var_1680_cast, y = var_1681_to_fp16)[name = tensor<string, []>("op_1682_cast")];
            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_1682_cast)[name = tensor<string, []>("denom_7_cast")];
            tensor<fp16, [2, 320, 1, 4096]> out_7_cast = mul(x = zero_mean_7_cast, y = denom_7_cast)[name = tensor<string, []>("out_7_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_1686_to_fp16 = const()[name = tensor<string, []>("op_1686_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6942016)))];
            tensor<fp16, [2, 320, 1, 4096]> var_1687_cast = add(x = out_7_cast, y = var_1686_to_fp16)[name = tensor<string, []>("op_1687_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_1689_to_fp16 = const()[name = tensor<string, []>("op_1689_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6942720)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_27_cast = mul(x = var_1687_cast, y = var_1689_to_fp16)[name = tensor<string, []>("hidden_states_27_cast")];
            tensor<int32, [2]> var_1696 = const()[name = tensor<string, []>("op_1696"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1698 = const()[name = tensor<string, []>("op_1698"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6943424))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7020288))), name = tensor<string, []>("down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            tensor<fp16, [2, 320, 1, 4096]> q_5_cast = conv(dilations = var_1698, groups = var_123, pad = q_5_pad_0, pad_type = q_5_pad_type_0, strides = var_1696, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_27_cast)[name = tensor<string, []>("q_5_cast")];
            tensor<int32, [2]> var_1702 = const()[name = tensor<string, []>("op_1702"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1704 = const()[name = tensor<string, []>("op_1704"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7020480))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7097344))), name = tensor<string, []>("down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            tensor<fp16, [2, 320, 1, 4096]> k_9_cast = conv(dilations = var_1704, groups = var_123, pad = k_9_pad_0, pad_type = k_9_pad_type_0, strides = var_1702, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_27_cast)[name = tensor<string, []>("k_9_cast")];
            tensor<int32, [2]> var_1708 = const()[name = tensor<string, []>("op_1708"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1710 = const()[name = tensor<string, []>("op_1710"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7097536))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7174400))), name = tensor<string, []>("down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            tensor<fp16, [2, 320, 1, 4096]> v_5_cast = conv(dilations = var_1710, groups = var_123, pad = v_5_pad_0, pad_type = v_5_pad_type_0, strides = var_1708, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_27_cast)[name = tensor<string, []>("v_5_cast")];
            tensor<int32, [4]> var_1714_begin_0 = const()[name = tensor<string, []>("op_1714_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1714_end_0 = const()[name = tensor<string, []>("op_1714_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_1714_end_mask_0 = const()[name = tensor<string, []>("op_1714_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_1714_cast = slice_by_index(begin = var_1714_begin_0, end = var_1714_end_0, end_mask = var_1714_end_mask_0, x = q_5_cast)[name = tensor<string, []>("op_1714_cast")];
            tensor<int32, [4]> var_1718_begin_0 = const()[name = tensor<string, []>("op_1718_begin_0"), val = tensor<int32, [4]>([0, 40, 0, 0])];
            tensor<int32, [4]> var_1718_end_0 = const()[name = tensor<string, []>("op_1718_end_0"), val = tensor<int32, [4]>([2, 80, 1, 4096])];
            tensor<bool, [4]> var_1718_end_mask_0 = const()[name = tensor<string, []>("op_1718_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_1718_cast = slice_by_index(begin = var_1718_begin_0, end = var_1718_end_0, end_mask = var_1718_end_mask_0, x = q_5_cast)[name = tensor<string, []>("op_1718_cast")];
            tensor<int32, [4]> var_1722_begin_0 = const()[name = tensor<string, []>("op_1722_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_1722_end_0 = const()[name = tensor<string, []>("op_1722_end_0"), val = tensor<int32, [4]>([2, 120, 1, 4096])];
            tensor<bool, [4]> var_1722_end_mask_0 = const()[name = tensor<string, []>("op_1722_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_1722_cast = slice_by_index(begin = var_1722_begin_0, end = var_1722_end_0, end_mask = var_1722_end_mask_0, x = q_5_cast)[name = tensor<string, []>("op_1722_cast")];
            tensor<int32, [4]> var_1726_begin_0 = const()[name = tensor<string, []>("op_1726_begin_0"), val = tensor<int32, [4]>([0, 120, 0, 0])];
            tensor<int32, [4]> var_1726_end_0 = const()[name = tensor<string, []>("op_1726_end_0"), val = tensor<int32, [4]>([2, 160, 1, 4096])];
            tensor<bool, [4]> var_1726_end_mask_0 = const()[name = tensor<string, []>("op_1726_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_1726_cast = slice_by_index(begin = var_1726_begin_0, end = var_1726_end_0, end_mask = var_1726_end_mask_0, x = q_5_cast)[name = tensor<string, []>("op_1726_cast")];
            tensor<int32, [4]> var_1730_begin_0 = const()[name = tensor<string, []>("op_1730_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_1730_end_0 = const()[name = tensor<string, []>("op_1730_end_0"), val = tensor<int32, [4]>([2, 200, 1, 4096])];
            tensor<bool, [4]> var_1730_end_mask_0 = const()[name = tensor<string, []>("op_1730_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_1730_cast = slice_by_index(begin = var_1730_begin_0, end = var_1730_end_0, end_mask = var_1730_end_mask_0, x = q_5_cast)[name = tensor<string, []>("op_1730_cast")];
            tensor<int32, [4]> var_1734_begin_0 = const()[name = tensor<string, []>("op_1734_begin_0"), val = tensor<int32, [4]>([0, 200, 0, 0])];
            tensor<int32, [4]> var_1734_end_0 = const()[name = tensor<string, []>("op_1734_end_0"), val = tensor<int32, [4]>([2, 240, 1, 4096])];
            tensor<bool, [4]> var_1734_end_mask_0 = const()[name = tensor<string, []>("op_1734_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_1734_cast = slice_by_index(begin = var_1734_begin_0, end = var_1734_end_0, end_mask = var_1734_end_mask_0, x = q_5_cast)[name = tensor<string, []>("op_1734_cast")];
            tensor<int32, [4]> var_1738_begin_0 = const()[name = tensor<string, []>("op_1738_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_1738_end_0 = const()[name = tensor<string, []>("op_1738_end_0"), val = tensor<int32, [4]>([2, 280, 1, 4096])];
            tensor<bool, [4]> var_1738_end_mask_0 = const()[name = tensor<string, []>("op_1738_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_1738_cast = slice_by_index(begin = var_1738_begin_0, end = var_1738_end_0, end_mask = var_1738_end_mask_0, x = q_5_cast)[name = tensor<string, []>("op_1738_cast")];
            tensor<int32, [4]> var_1742_begin_0 = const()[name = tensor<string, []>("op_1742_begin_0"), val = tensor<int32, [4]>([0, 280, 0, 0])];
            tensor<int32, [4]> var_1742_end_0 = const()[name = tensor<string, []>("op_1742_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_1742_end_mask_0 = const()[name = tensor<string, []>("op_1742_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_1742_cast = slice_by_index(begin = var_1742_begin_0, end = var_1742_end_0, end_mask = var_1742_end_mask_0, x = q_5_cast)[name = tensor<string, []>("op_1742_cast")];
            tensor<int32, [4]> var_1745_begin_0 = const()[name = tensor<string, []>("op_1745_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1745_end_0 = const()[name = tensor<string, []>("op_1745_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_1745_end_mask_0 = const()[name = tensor<string, []>("op_1745_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1745_cast = slice_by_index(begin = var_1745_begin_0, end = var_1745_end_0, end_mask = var_1745_end_mask_0, x = var_1714_cast)[name = tensor<string, []>("op_1745_cast")];
            tensor<int32, [4]> var_1746_begin_0 = const()[name = tensor<string, []>("op_1746_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_1746_end_0 = const()[name = tensor<string, []>("op_1746_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_1746_end_mask_0 = const()[name = tensor<string, []>("op_1746_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1746_cast = slice_by_index(begin = var_1746_begin_0, end = var_1746_end_0, end_mask = var_1746_end_mask_0, x = var_1714_cast)[name = tensor<string, []>("op_1746_cast")];
            tensor<int32, [4]> var_1747_begin_0 = const()[name = tensor<string, []>("op_1747_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_1747_end_0 = const()[name = tensor<string, []>("op_1747_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_1747_end_mask_0 = const()[name = tensor<string, []>("op_1747_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1747_cast = slice_by_index(begin = var_1747_begin_0, end = var_1747_end_0, end_mask = var_1747_end_mask_0, x = var_1714_cast)[name = tensor<string, []>("op_1747_cast")];
            tensor<int32, [4]> var_1748_begin_0 = const()[name = tensor<string, []>("op_1748_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_1748_end_0 = const()[name = tensor<string, []>("op_1748_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_1748_end_mask_0 = const()[name = tensor<string, []>("op_1748_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1748_cast = slice_by_index(begin = var_1748_begin_0, end = var_1748_end_0, end_mask = var_1748_end_mask_0, x = var_1714_cast)[name = tensor<string, []>("op_1748_cast")];
            tensor<int32, [4]> var_1749_begin_0 = const()[name = tensor<string, []>("op_1749_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_1749_end_0 = const()[name = tensor<string, []>("op_1749_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_1749_end_mask_0 = const()[name = tensor<string, []>("op_1749_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1749_cast = slice_by_index(begin = var_1749_begin_0, end = var_1749_end_0, end_mask = var_1749_end_mask_0, x = var_1714_cast)[name = tensor<string, []>("op_1749_cast")];
            tensor<int32, [4]> var_1750_begin_0 = const()[name = tensor<string, []>("op_1750_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_1750_end_0 = const()[name = tensor<string, []>("op_1750_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_1750_end_mask_0 = const()[name = tensor<string, []>("op_1750_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1750_cast = slice_by_index(begin = var_1750_begin_0, end = var_1750_end_0, end_mask = var_1750_end_mask_0, x = var_1714_cast)[name = tensor<string, []>("op_1750_cast")];
            tensor<int32, [4]> var_1751_begin_0 = const()[name = tensor<string, []>("op_1751_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_1751_end_0 = const()[name = tensor<string, []>("op_1751_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_1751_end_mask_0 = const()[name = tensor<string, []>("op_1751_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1751_cast = slice_by_index(begin = var_1751_begin_0, end = var_1751_end_0, end_mask = var_1751_end_mask_0, x = var_1714_cast)[name = tensor<string, []>("op_1751_cast")];
            tensor<int32, [4]> var_1752_begin_0 = const()[name = tensor<string, []>("op_1752_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_1752_end_0 = const()[name = tensor<string, []>("op_1752_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_1752_end_mask_0 = const()[name = tensor<string, []>("op_1752_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1752_cast = slice_by_index(begin = var_1752_begin_0, end = var_1752_end_0, end_mask = var_1752_end_mask_0, x = var_1714_cast)[name = tensor<string, []>("op_1752_cast")];
            tensor<int32, [4]> var_1753_begin_0 = const()[name = tensor<string, []>("op_1753_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1753_end_0 = const()[name = tensor<string, []>("op_1753_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_1753_end_mask_0 = const()[name = tensor<string, []>("op_1753_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1753_cast = slice_by_index(begin = var_1753_begin_0, end = var_1753_end_0, end_mask = var_1753_end_mask_0, x = var_1718_cast)[name = tensor<string, []>("op_1753_cast")];
            tensor<int32, [4]> var_1754_begin_0 = const()[name = tensor<string, []>("op_1754_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_1754_end_0 = const()[name = tensor<string, []>("op_1754_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_1754_end_mask_0 = const()[name = tensor<string, []>("op_1754_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1754_cast = slice_by_index(begin = var_1754_begin_0, end = var_1754_end_0, end_mask = var_1754_end_mask_0, x = var_1718_cast)[name = tensor<string, []>("op_1754_cast")];
            tensor<int32, [4]> var_1755_begin_0 = const()[name = tensor<string, []>("op_1755_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_1755_end_0 = const()[name = tensor<string, []>("op_1755_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_1755_end_mask_0 = const()[name = tensor<string, []>("op_1755_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1755_cast = slice_by_index(begin = var_1755_begin_0, end = var_1755_end_0, end_mask = var_1755_end_mask_0, x = var_1718_cast)[name = tensor<string, []>("op_1755_cast")];
            tensor<int32, [4]> var_1756_begin_0 = const()[name = tensor<string, []>("op_1756_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_1756_end_0 = const()[name = tensor<string, []>("op_1756_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_1756_end_mask_0 = const()[name = tensor<string, []>("op_1756_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1756_cast = slice_by_index(begin = var_1756_begin_0, end = var_1756_end_0, end_mask = var_1756_end_mask_0, x = var_1718_cast)[name = tensor<string, []>("op_1756_cast")];
            tensor<int32, [4]> var_1757_begin_0 = const()[name = tensor<string, []>("op_1757_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_1757_end_0 = const()[name = tensor<string, []>("op_1757_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_1757_end_mask_0 = const()[name = tensor<string, []>("op_1757_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1757_cast = slice_by_index(begin = var_1757_begin_0, end = var_1757_end_0, end_mask = var_1757_end_mask_0, x = var_1718_cast)[name = tensor<string, []>("op_1757_cast")];
            tensor<int32, [4]> var_1758_begin_0 = const()[name = tensor<string, []>("op_1758_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_1758_end_0 = const()[name = tensor<string, []>("op_1758_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            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, 40, 1, 512]> var_1758_cast = slice_by_index(begin = var_1758_begin_0, end = var_1758_end_0, end_mask = var_1758_end_mask_0, x = var_1718_cast)[name = tensor<string, []>("op_1758_cast")];
            tensor<int32, [4]> var_1759_begin_0 = const()[name = tensor<string, []>("op_1759_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_1759_end_0 = const()[name = tensor<string, []>("op_1759_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_1759_end_mask_0 = const()[name = tensor<string, []>("op_1759_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1759_cast = slice_by_index(begin = var_1759_begin_0, end = var_1759_end_0, end_mask = var_1759_end_mask_0, x = var_1718_cast)[name = tensor<string, []>("op_1759_cast")];
            tensor<int32, [4]> var_1760_begin_0 = const()[name = tensor<string, []>("op_1760_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_1760_end_0 = const()[name = tensor<string, []>("op_1760_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_1760_end_mask_0 = const()[name = tensor<string, []>("op_1760_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1760_cast = slice_by_index(begin = var_1760_begin_0, end = var_1760_end_0, end_mask = var_1760_end_mask_0, x = var_1718_cast)[name = tensor<string, []>("op_1760_cast")];
            tensor<int32, [4]> var_1761_begin_0 = const()[name = tensor<string, []>("op_1761_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1761_end_0 = const()[name = tensor<string, []>("op_1761_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_1761_end_mask_0 = const()[name = tensor<string, []>("op_1761_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1761_cast = slice_by_index(begin = var_1761_begin_0, end = var_1761_end_0, end_mask = var_1761_end_mask_0, x = var_1722_cast)[name = tensor<string, []>("op_1761_cast")];
            tensor<int32, [4]> var_1762_begin_0 = const()[name = tensor<string, []>("op_1762_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_1762_end_0 = const()[name = tensor<string, []>("op_1762_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            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, 40, 1, 512]> var_1762_cast = slice_by_index(begin = var_1762_begin_0, end = var_1762_end_0, end_mask = var_1762_end_mask_0, x = var_1722_cast)[name = tensor<string, []>("op_1762_cast")];
            tensor<int32, [4]> var_1763_begin_0 = const()[name = tensor<string, []>("op_1763_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_1763_end_0 = const()[name = tensor<string, []>("op_1763_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_1763_end_mask_0 = const()[name = tensor<string, []>("op_1763_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1763_cast = slice_by_index(begin = var_1763_begin_0, end = var_1763_end_0, end_mask = var_1763_end_mask_0, x = var_1722_cast)[name = tensor<string, []>("op_1763_cast")];
            tensor<int32, [4]> var_1764_begin_0 = const()[name = tensor<string, []>("op_1764_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_1764_end_0 = const()[name = tensor<string, []>("op_1764_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_1764_end_mask_0 = const()[name = tensor<string, []>("op_1764_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1764_cast = slice_by_index(begin = var_1764_begin_0, end = var_1764_end_0, end_mask = var_1764_end_mask_0, x = var_1722_cast)[name = tensor<string, []>("op_1764_cast")];
            tensor<int32, [4]> var_1765_begin_0 = const()[name = tensor<string, []>("op_1765_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_1765_end_0 = const()[name = tensor<string, []>("op_1765_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_1765_end_mask_0 = const()[name = tensor<string, []>("op_1765_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1765_cast = slice_by_index(begin = var_1765_begin_0, end = var_1765_end_0, end_mask = var_1765_end_mask_0, x = var_1722_cast)[name = tensor<string, []>("op_1765_cast")];
            tensor<int32, [4]> var_1766_begin_0 = const()[name = tensor<string, []>("op_1766_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_1766_end_0 = const()[name = tensor<string, []>("op_1766_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            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, 40, 1, 512]> var_1766_cast = slice_by_index(begin = var_1766_begin_0, end = var_1766_end_0, end_mask = var_1766_end_mask_0, x = var_1722_cast)[name = tensor<string, []>("op_1766_cast")];
            tensor<int32, [4]> var_1767_begin_0 = const()[name = tensor<string, []>("op_1767_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_1767_end_0 = const()[name = tensor<string, []>("op_1767_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_1767_end_mask_0 = const()[name = tensor<string, []>("op_1767_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1767_cast = slice_by_index(begin = var_1767_begin_0, end = var_1767_end_0, end_mask = var_1767_end_mask_0, x = var_1722_cast)[name = tensor<string, []>("op_1767_cast")];
            tensor<int32, [4]> var_1768_begin_0 = const()[name = tensor<string, []>("op_1768_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_1768_end_0 = const()[name = tensor<string, []>("op_1768_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_1768_end_mask_0 = const()[name = tensor<string, []>("op_1768_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1768_cast = slice_by_index(begin = var_1768_begin_0, end = var_1768_end_0, end_mask = var_1768_end_mask_0, x = var_1722_cast)[name = tensor<string, []>("op_1768_cast")];
            tensor<int32, [4]> var_1769_begin_0 = const()[name = tensor<string, []>("op_1769_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1769_end_0 = const()[name = tensor<string, []>("op_1769_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_1769_end_mask_0 = const()[name = tensor<string, []>("op_1769_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1769_cast = slice_by_index(begin = var_1769_begin_0, end = var_1769_end_0, end_mask = var_1769_end_mask_0, x = var_1726_cast)[name = tensor<string, []>("op_1769_cast")];
            tensor<int32, [4]> var_1770_begin_0 = const()[name = tensor<string, []>("op_1770_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_1770_end_0 = const()[name = tensor<string, []>("op_1770_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            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, 40, 1, 512]> var_1770_cast = slice_by_index(begin = var_1770_begin_0, end = var_1770_end_0, end_mask = var_1770_end_mask_0, x = var_1726_cast)[name = tensor<string, []>("op_1770_cast")];
            tensor<int32, [4]> var_1771_begin_0 = const()[name = tensor<string, []>("op_1771_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_1771_end_0 = const()[name = tensor<string, []>("op_1771_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_1771_end_mask_0 = const()[name = tensor<string, []>("op_1771_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1771_cast = slice_by_index(begin = var_1771_begin_0, end = var_1771_end_0, end_mask = var_1771_end_mask_0, x = var_1726_cast)[name = tensor<string, []>("op_1771_cast")];
            tensor<int32, [4]> var_1772_begin_0 = const()[name = tensor<string, []>("op_1772_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_1772_end_0 = const()[name = tensor<string, []>("op_1772_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_1772_end_mask_0 = const()[name = tensor<string, []>("op_1772_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1772_cast = slice_by_index(begin = var_1772_begin_0, end = var_1772_end_0, end_mask = var_1772_end_mask_0, x = var_1726_cast)[name = tensor<string, []>("op_1772_cast")];
            tensor<int32, [4]> var_1773_begin_0 = const()[name = tensor<string, []>("op_1773_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_1773_end_0 = const()[name = tensor<string, []>("op_1773_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_1773_end_mask_0 = const()[name = tensor<string, []>("op_1773_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1773_cast = slice_by_index(begin = var_1773_begin_0, end = var_1773_end_0, end_mask = var_1773_end_mask_0, x = var_1726_cast)[name = tensor<string, []>("op_1773_cast")];
            tensor<int32, [4]> var_1774_begin_0 = const()[name = tensor<string, []>("op_1774_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_1774_end_0 = const()[name = tensor<string, []>("op_1774_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            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, 40, 1, 512]> var_1774_cast = slice_by_index(begin = var_1774_begin_0, end = var_1774_end_0, end_mask = var_1774_end_mask_0, x = var_1726_cast)[name = tensor<string, []>("op_1774_cast")];
            tensor<int32, [4]> var_1775_begin_0 = const()[name = tensor<string, []>("op_1775_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_1775_end_0 = const()[name = tensor<string, []>("op_1775_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_1775_end_mask_0 = const()[name = tensor<string, []>("op_1775_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1775_cast = slice_by_index(begin = var_1775_begin_0, end = var_1775_end_0, end_mask = var_1775_end_mask_0, x = var_1726_cast)[name = tensor<string, []>("op_1775_cast")];
            tensor<int32, [4]> var_1776_begin_0 = const()[name = tensor<string, []>("op_1776_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_1776_end_0 = const()[name = tensor<string, []>("op_1776_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_1776_end_mask_0 = const()[name = tensor<string, []>("op_1776_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1776_cast = slice_by_index(begin = var_1776_begin_0, end = var_1776_end_0, end_mask = var_1776_end_mask_0, x = var_1726_cast)[name = tensor<string, []>("op_1776_cast")];
            tensor<int32, [4]> var_1777_begin_0 = const()[name = tensor<string, []>("op_1777_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1777_end_0 = const()[name = tensor<string, []>("op_1777_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_1777_end_mask_0 = const()[name = tensor<string, []>("op_1777_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1777_cast = slice_by_index(begin = var_1777_begin_0, end = var_1777_end_0, end_mask = var_1777_end_mask_0, x = var_1730_cast)[name = tensor<string, []>("op_1777_cast")];
            tensor<int32, [4]> var_1778_begin_0 = const()[name = tensor<string, []>("op_1778_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_1778_end_0 = const()[name = tensor<string, []>("op_1778_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            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, 40, 1, 512]> var_1778_cast = slice_by_index(begin = var_1778_begin_0, end = var_1778_end_0, end_mask = var_1778_end_mask_0, x = var_1730_cast)[name = tensor<string, []>("op_1778_cast")];
            tensor<int32, [4]> var_1779_begin_0 = const()[name = tensor<string, []>("op_1779_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_1779_end_0 = const()[name = tensor<string, []>("op_1779_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_1779_end_mask_0 = const()[name = tensor<string, []>("op_1779_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1779_cast = slice_by_index(begin = var_1779_begin_0, end = var_1779_end_0, end_mask = var_1779_end_mask_0, x = var_1730_cast)[name = tensor<string, []>("op_1779_cast")];
            tensor<int32, [4]> var_1780_begin_0 = const()[name = tensor<string, []>("op_1780_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_1780_end_0 = const()[name = tensor<string, []>("op_1780_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_1780_end_mask_0 = const()[name = tensor<string, []>("op_1780_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1780_cast = slice_by_index(begin = var_1780_begin_0, end = var_1780_end_0, end_mask = var_1780_end_mask_0, x = var_1730_cast)[name = tensor<string, []>("op_1780_cast")];
            tensor<int32, [4]> var_1781_begin_0 = const()[name = tensor<string, []>("op_1781_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_1781_end_0 = const()[name = tensor<string, []>("op_1781_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_1781_end_mask_0 = const()[name = tensor<string, []>("op_1781_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1781_cast = slice_by_index(begin = var_1781_begin_0, end = var_1781_end_0, end_mask = var_1781_end_mask_0, x = var_1730_cast)[name = tensor<string, []>("op_1781_cast")];
            tensor<int32, [4]> var_1782_begin_0 = const()[name = tensor<string, []>("op_1782_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_1782_end_0 = const()[name = tensor<string, []>("op_1782_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            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, 40, 1, 512]> var_1782_cast = slice_by_index(begin = var_1782_begin_0, end = var_1782_end_0, end_mask = var_1782_end_mask_0, x = var_1730_cast)[name = tensor<string, []>("op_1782_cast")];
            tensor<int32, [4]> var_1783_begin_0 = const()[name = tensor<string, []>("op_1783_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_1783_end_0 = const()[name = tensor<string, []>("op_1783_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_1783_end_mask_0 = const()[name = tensor<string, []>("op_1783_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1783_cast = slice_by_index(begin = var_1783_begin_0, end = var_1783_end_0, end_mask = var_1783_end_mask_0, x = var_1730_cast)[name = tensor<string, []>("op_1783_cast")];
            tensor<int32, [4]> var_1784_begin_0 = const()[name = tensor<string, []>("op_1784_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_1784_end_0 = const()[name = tensor<string, []>("op_1784_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_1784_end_mask_0 = const()[name = tensor<string, []>("op_1784_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1784_cast = slice_by_index(begin = var_1784_begin_0, end = var_1784_end_0, end_mask = var_1784_end_mask_0, x = var_1730_cast)[name = tensor<string, []>("op_1784_cast")];
            tensor<int32, [4]> var_1785_begin_0 = const()[name = tensor<string, []>("op_1785_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1785_end_0 = const()[name = tensor<string, []>("op_1785_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_1785_end_mask_0 = const()[name = tensor<string, []>("op_1785_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1785_cast = slice_by_index(begin = var_1785_begin_0, end = var_1785_end_0, end_mask = var_1785_end_mask_0, x = var_1734_cast)[name = tensor<string, []>("op_1785_cast")];
            tensor<int32, [4]> var_1786_begin_0 = const()[name = tensor<string, []>("op_1786_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_1786_end_0 = const()[name = tensor<string, []>("op_1786_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            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, 40, 1, 512]> var_1786_cast = slice_by_index(begin = var_1786_begin_0, end = var_1786_end_0, end_mask = var_1786_end_mask_0, x = var_1734_cast)[name = tensor<string, []>("op_1786_cast")];
            tensor<int32, [4]> var_1787_begin_0 = const()[name = tensor<string, []>("op_1787_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_1787_end_0 = const()[name = tensor<string, []>("op_1787_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_1787_end_mask_0 = const()[name = tensor<string, []>("op_1787_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1787_cast = slice_by_index(begin = var_1787_begin_0, end = var_1787_end_0, end_mask = var_1787_end_mask_0, x = var_1734_cast)[name = tensor<string, []>("op_1787_cast")];
            tensor<int32, [4]> var_1788_begin_0 = const()[name = tensor<string, []>("op_1788_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_1788_end_0 = const()[name = tensor<string, []>("op_1788_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_1788_end_mask_0 = const()[name = tensor<string, []>("op_1788_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1788_cast = slice_by_index(begin = var_1788_begin_0, end = var_1788_end_0, end_mask = var_1788_end_mask_0, x = var_1734_cast)[name = tensor<string, []>("op_1788_cast")];
            tensor<int32, [4]> var_1789_begin_0 = const()[name = tensor<string, []>("op_1789_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_1789_end_0 = const()[name = tensor<string, []>("op_1789_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_1789_end_mask_0 = const()[name = tensor<string, []>("op_1789_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1789_cast = slice_by_index(begin = var_1789_begin_0, end = var_1789_end_0, end_mask = var_1789_end_mask_0, x = var_1734_cast)[name = tensor<string, []>("op_1789_cast")];
            tensor<int32, [4]> var_1790_begin_0 = const()[name = tensor<string, []>("op_1790_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_1790_end_0 = const()[name = tensor<string, []>("op_1790_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            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, 40, 1, 512]> var_1790_cast = slice_by_index(begin = var_1790_begin_0, end = var_1790_end_0, end_mask = var_1790_end_mask_0, x = var_1734_cast)[name = tensor<string, []>("op_1790_cast")];
            tensor<int32, [4]> var_1791_begin_0 = const()[name = tensor<string, []>("op_1791_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_1791_end_0 = const()[name = tensor<string, []>("op_1791_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_1791_end_mask_0 = const()[name = tensor<string, []>("op_1791_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1791_cast = slice_by_index(begin = var_1791_begin_0, end = var_1791_end_0, end_mask = var_1791_end_mask_0, x = var_1734_cast)[name = tensor<string, []>("op_1791_cast")];
            tensor<int32, [4]> var_1792_begin_0 = const()[name = tensor<string, []>("op_1792_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_1792_end_0 = const()[name = tensor<string, []>("op_1792_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_1792_end_mask_0 = const()[name = tensor<string, []>("op_1792_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1792_cast = slice_by_index(begin = var_1792_begin_0, end = var_1792_end_0, end_mask = var_1792_end_mask_0, x = var_1734_cast)[name = tensor<string, []>("op_1792_cast")];
            tensor<int32, [4]> var_1793_begin_0 = const()[name = tensor<string, []>("op_1793_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1793_end_0 = const()[name = tensor<string, []>("op_1793_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_1793_end_mask_0 = const()[name = tensor<string, []>("op_1793_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1793_cast = slice_by_index(begin = var_1793_begin_0, end = var_1793_end_0, end_mask = var_1793_end_mask_0, x = var_1738_cast)[name = tensor<string, []>("op_1793_cast")];
            tensor<int32, [4]> var_1794_begin_0 = const()[name = tensor<string, []>("op_1794_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_1794_end_0 = const()[name = tensor<string, []>("op_1794_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            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, 40, 1, 512]> var_1794_cast = slice_by_index(begin = var_1794_begin_0, end = var_1794_end_0, end_mask = var_1794_end_mask_0, x = var_1738_cast)[name = tensor<string, []>("op_1794_cast")];
            tensor<int32, [4]> var_1795_begin_0 = const()[name = tensor<string, []>("op_1795_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_1795_end_0 = const()[name = tensor<string, []>("op_1795_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_1795_end_mask_0 = const()[name = tensor<string, []>("op_1795_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1795_cast = slice_by_index(begin = var_1795_begin_0, end = var_1795_end_0, end_mask = var_1795_end_mask_0, x = var_1738_cast)[name = tensor<string, []>("op_1795_cast")];
            tensor<int32, [4]> var_1796_begin_0 = const()[name = tensor<string, []>("op_1796_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_1796_end_0 = const()[name = tensor<string, []>("op_1796_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_1796_end_mask_0 = const()[name = tensor<string, []>("op_1796_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1796_cast = slice_by_index(begin = var_1796_begin_0, end = var_1796_end_0, end_mask = var_1796_end_mask_0, x = var_1738_cast)[name = tensor<string, []>("op_1796_cast")];
            tensor<int32, [4]> var_1797_begin_0 = const()[name = tensor<string, []>("op_1797_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_1797_end_0 = const()[name = tensor<string, []>("op_1797_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_1797_end_mask_0 = const()[name = tensor<string, []>("op_1797_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1797_cast = slice_by_index(begin = var_1797_begin_0, end = var_1797_end_0, end_mask = var_1797_end_mask_0, x = var_1738_cast)[name = tensor<string, []>("op_1797_cast")];
            tensor<int32, [4]> var_1798_begin_0 = const()[name = tensor<string, []>("op_1798_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_1798_end_0 = const()[name = tensor<string, []>("op_1798_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_1798_end_mask_0 = const()[name = tensor<string, []>("op_1798_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1798_cast = slice_by_index(begin = var_1798_begin_0, end = var_1798_end_0, end_mask = var_1798_end_mask_0, x = var_1738_cast)[name = tensor<string, []>("op_1798_cast")];
            tensor<int32, [4]> var_1799_begin_0 = const()[name = tensor<string, []>("op_1799_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_1799_end_0 = const()[name = tensor<string, []>("op_1799_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_1799_end_mask_0 = const()[name = tensor<string, []>("op_1799_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1799_cast = slice_by_index(begin = var_1799_begin_0, end = var_1799_end_0, end_mask = var_1799_end_mask_0, x = var_1738_cast)[name = tensor<string, []>("op_1799_cast")];
            tensor<int32, [4]> var_1800_begin_0 = const()[name = tensor<string, []>("op_1800_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_1800_end_0 = const()[name = tensor<string, []>("op_1800_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_1800_end_mask_0 = const()[name = tensor<string, []>("op_1800_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1800_cast = slice_by_index(begin = var_1800_begin_0, end = var_1800_end_0, end_mask = var_1800_end_mask_0, x = var_1738_cast)[name = tensor<string, []>("op_1800_cast")];
            tensor<int32, [4]> var_1801_begin_0 = const()[name = tensor<string, []>("op_1801_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1801_end_0 = const()[name = tensor<string, []>("op_1801_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_1801_end_mask_0 = const()[name = tensor<string, []>("op_1801_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1801_cast = slice_by_index(begin = var_1801_begin_0, end = var_1801_end_0, end_mask = var_1801_end_mask_0, x = var_1742_cast)[name = tensor<string, []>("op_1801_cast")];
            tensor<int32, [4]> var_1802_begin_0 = const()[name = tensor<string, []>("op_1802_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_1802_end_0 = const()[name = tensor<string, []>("op_1802_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_1802_end_mask_0 = const()[name = tensor<string, []>("op_1802_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1802_cast = slice_by_index(begin = var_1802_begin_0, end = var_1802_end_0, end_mask = var_1802_end_mask_0, x = var_1742_cast)[name = tensor<string, []>("op_1802_cast")];
            tensor<int32, [4]> var_1803_begin_0 = const()[name = tensor<string, []>("op_1803_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_1803_end_0 = const()[name = tensor<string, []>("op_1803_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_1803_end_mask_0 = const()[name = tensor<string, []>("op_1803_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1803_cast = slice_by_index(begin = var_1803_begin_0, end = var_1803_end_0, end_mask = var_1803_end_mask_0, x = var_1742_cast)[name = tensor<string, []>("op_1803_cast")];
            tensor<int32, [4]> var_1804_begin_0 = const()[name = tensor<string, []>("op_1804_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_1804_end_0 = const()[name = tensor<string, []>("op_1804_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_1804_end_mask_0 = const()[name = tensor<string, []>("op_1804_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1804_cast = slice_by_index(begin = var_1804_begin_0, end = var_1804_end_0, end_mask = var_1804_end_mask_0, x = var_1742_cast)[name = tensor<string, []>("op_1804_cast")];
            tensor<int32, [4]> var_1805_begin_0 = const()[name = tensor<string, []>("op_1805_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_1805_end_0 = const()[name = tensor<string, []>("op_1805_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_1805_end_mask_0 = const()[name = tensor<string, []>("op_1805_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1805_cast = slice_by_index(begin = var_1805_begin_0, end = var_1805_end_0, end_mask = var_1805_end_mask_0, x = var_1742_cast)[name = tensor<string, []>("op_1805_cast")];
            tensor<int32, [4]> var_1806_begin_0 = const()[name = tensor<string, []>("op_1806_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_1806_end_0 = const()[name = tensor<string, []>("op_1806_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_1806_end_mask_0 = const()[name = tensor<string, []>("op_1806_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1806_cast = slice_by_index(begin = var_1806_begin_0, end = var_1806_end_0, end_mask = var_1806_end_mask_0, x = var_1742_cast)[name = tensor<string, []>("op_1806_cast")];
            tensor<int32, [4]> var_1807_begin_0 = const()[name = tensor<string, []>("op_1807_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_1807_end_0 = const()[name = tensor<string, []>("op_1807_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_1807_end_mask_0 = const()[name = tensor<string, []>("op_1807_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1807_cast = slice_by_index(begin = var_1807_begin_0, end = var_1807_end_0, end_mask = var_1807_end_mask_0, x = var_1742_cast)[name = tensor<string, []>("op_1807_cast")];
            tensor<int32, [4]> var_1808_begin_0 = const()[name = tensor<string, []>("op_1808_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_1808_end_0 = const()[name = tensor<string, []>("op_1808_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_1808_end_mask_0 = const()[name = tensor<string, []>("op_1808_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_1808_cast = slice_by_index(begin = var_1808_begin_0, end = var_1808_end_0, end_mask = var_1808_end_mask_0, x = var_1742_cast)[name = tensor<string, []>("op_1808_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_1813_begin_0 = const()[name = tensor<string, []>("op_1813_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1813_end_0 = const()[name = tensor<string, []>("op_1813_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 40])];
            tensor<bool, [4]> var_1813_end_mask_0 = const()[name = tensor<string, []>("op_1813_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)[name = tensor<string, []>("transpose_29")];
            tensor<fp16, [2, 4096, 1, 40]> var_1813_cast = slice_by_index(begin = var_1813_begin_0, end = var_1813_end_0, end_mask = var_1813_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_1813_cast")];
            tensor<int32, [4]> var_1817_begin_0 = const()[name = tensor<string, []>("op_1817_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 40])];
            tensor<int32, [4]> var_1817_end_0 = const()[name = tensor<string, []>("op_1817_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 80])];
            tensor<bool, [4]> var_1817_end_mask_0 = const()[name = tensor<string, []>("op_1817_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_1817_cast = slice_by_index(begin = var_1817_begin_0, end = var_1817_end_0, end_mask = var_1817_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_1817_cast")];
            tensor<int32, [4]> var_1821_begin_0 = const()[name = tensor<string, []>("op_1821_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 80])];
            tensor<int32, [4]> var_1821_end_0 = const()[name = tensor<string, []>("op_1821_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 120])];
            tensor<bool, [4]> var_1821_end_mask_0 = const()[name = tensor<string, []>("op_1821_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_1821_cast = slice_by_index(begin = var_1821_begin_0, end = var_1821_end_0, end_mask = var_1821_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_1821_cast")];
            tensor<int32, [4]> var_1825_begin_0 = const()[name = tensor<string, []>("op_1825_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 120])];
            tensor<int32, [4]> var_1825_end_0 = const()[name = tensor<string, []>("op_1825_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 160])];
            tensor<bool, [4]> var_1825_end_mask_0 = const()[name = tensor<string, []>("op_1825_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_1825_cast = slice_by_index(begin = var_1825_begin_0, end = var_1825_end_0, end_mask = var_1825_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_1825_cast")];
            tensor<int32, [4]> var_1829_begin_0 = const()[name = tensor<string, []>("op_1829_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_1829_end_0 = const()[name = tensor<string, []>("op_1829_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 200])];
            tensor<bool, [4]> var_1829_end_mask_0 = const()[name = tensor<string, []>("op_1829_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_1829_cast = slice_by_index(begin = var_1829_begin_0, end = var_1829_end_0, end_mask = var_1829_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_1829_cast")];
            tensor<int32, [4]> var_1833_begin_0 = const()[name = tensor<string, []>("op_1833_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 200])];
            tensor<int32, [4]> var_1833_end_0 = const()[name = tensor<string, []>("op_1833_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 240])];
            tensor<bool, [4]> var_1833_end_mask_0 = const()[name = tensor<string, []>("op_1833_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_1833_cast = slice_by_index(begin = var_1833_begin_0, end = var_1833_end_0, end_mask = var_1833_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_1833_cast")];
            tensor<int32, [4]> var_1837_begin_0 = const()[name = tensor<string, []>("op_1837_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 240])];
            tensor<int32, [4]> var_1837_end_0 = const()[name = tensor<string, []>("op_1837_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 280])];
            tensor<bool, [4]> var_1837_end_mask_0 = const()[name = tensor<string, []>("op_1837_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_1837_cast = slice_by_index(begin = var_1837_begin_0, end = var_1837_end_0, end_mask = var_1837_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_1837_cast")];
            tensor<int32, [4]> var_1841_begin_0 = const()[name = tensor<string, []>("op_1841_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 280])];
            tensor<int32, [4]> var_1841_end_0 = const()[name = tensor<string, []>("op_1841_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_1841_end_mask_0 = const()[name = tensor<string, []>("op_1841_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_1841_cast = slice_by_index(begin = var_1841_begin_0, end = var_1841_end_0, end_mask = var_1841_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_1841_cast")];
            tensor<int32, [4]> var_1843_begin_0 = const()[name = tensor<string, []>("op_1843_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1843_end_0 = const()[name = tensor<string, []>("op_1843_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_1843_end_mask_0 = const()[name = tensor<string, []>("op_1843_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_1843_cast = slice_by_index(begin = var_1843_begin_0, end = var_1843_end_0, end_mask = var_1843_end_mask_0, x = v_5_cast)[name = tensor<string, []>("op_1843_cast")];
            tensor<int32, [4]> var_1847_begin_0 = const()[name = tensor<string, []>("op_1847_begin_0"), val = tensor<int32, [4]>([0, 40, 0, 0])];
            tensor<int32, [4]> var_1847_end_0 = const()[name = tensor<string, []>("op_1847_end_0"), val = tensor<int32, [4]>([2, 80, 1, 4096])];
            tensor<bool, [4]> var_1847_end_mask_0 = const()[name = tensor<string, []>("op_1847_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_1847_cast = slice_by_index(begin = var_1847_begin_0, end = var_1847_end_0, end_mask = var_1847_end_mask_0, x = v_5_cast)[name = tensor<string, []>("op_1847_cast")];
            tensor<int32, [4]> var_1851_begin_0 = const()[name = tensor<string, []>("op_1851_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_1851_end_0 = const()[name = tensor<string, []>("op_1851_end_0"), val = tensor<int32, [4]>([2, 120, 1, 4096])];
            tensor<bool, [4]> var_1851_end_mask_0 = const()[name = tensor<string, []>("op_1851_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_1851_cast = slice_by_index(begin = var_1851_begin_0, end = var_1851_end_0, end_mask = var_1851_end_mask_0, x = v_5_cast)[name = tensor<string, []>("op_1851_cast")];
            tensor<int32, [4]> var_1855_begin_0 = const()[name = tensor<string, []>("op_1855_begin_0"), val = tensor<int32, [4]>([0, 120, 0, 0])];
            tensor<int32, [4]> var_1855_end_0 = const()[name = tensor<string, []>("op_1855_end_0"), val = tensor<int32, [4]>([2, 160, 1, 4096])];
            tensor<bool, [4]> var_1855_end_mask_0 = const()[name = tensor<string, []>("op_1855_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_1855_cast = slice_by_index(begin = var_1855_begin_0, end = var_1855_end_0, end_mask = var_1855_end_mask_0, x = v_5_cast)[name = tensor<string, []>("op_1855_cast")];
            tensor<int32, [4]> var_1859_begin_0 = const()[name = tensor<string, []>("op_1859_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_1859_end_0 = const()[name = tensor<string, []>("op_1859_end_0"), val = tensor<int32, [4]>([2, 200, 1, 4096])];
            tensor<bool, [4]> var_1859_end_mask_0 = const()[name = tensor<string, []>("op_1859_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_1859_cast = slice_by_index(begin = var_1859_begin_0, end = var_1859_end_0, end_mask = var_1859_end_mask_0, x = v_5_cast)[name = tensor<string, []>("op_1859_cast")];
            tensor<int32, [4]> var_1863_begin_0 = const()[name = tensor<string, []>("op_1863_begin_0"), val = tensor<int32, [4]>([0, 200, 0, 0])];
            tensor<int32, [4]> var_1863_end_0 = const()[name = tensor<string, []>("op_1863_end_0"), val = tensor<int32, [4]>([2, 240, 1, 4096])];
            tensor<bool, [4]> var_1863_end_mask_0 = const()[name = tensor<string, []>("op_1863_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_1863_cast = slice_by_index(begin = var_1863_begin_0, end = var_1863_end_0, end_mask = var_1863_end_mask_0, x = v_5_cast)[name = tensor<string, []>("op_1863_cast")];
            tensor<int32, [4]> var_1867_begin_0 = const()[name = tensor<string, []>("op_1867_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_1867_end_0 = const()[name = tensor<string, []>("op_1867_end_0"), val = tensor<int32, [4]>([2, 280, 1, 4096])];
            tensor<bool, [4]> var_1867_end_mask_0 = const()[name = tensor<string, []>("op_1867_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_1867_cast = slice_by_index(begin = var_1867_begin_0, end = var_1867_end_0, end_mask = var_1867_end_mask_0, x = v_5_cast)[name = tensor<string, []>("op_1867_cast")];
            tensor<int32, [4]> var_1871_begin_0 = const()[name = tensor<string, []>("op_1871_begin_0"), val = tensor<int32, [4]>([0, 280, 0, 0])];
            tensor<int32, [4]> var_1871_end_0 = const()[name = tensor<string, []>("op_1871_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_1871_end_mask_0 = const()[name = tensor<string, []>("op_1871_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_1871_cast = slice_by_index(begin = var_1871_begin_0, end = var_1871_end_0, end_mask = var_1871_end_mask_0, x = v_5_cast)[name = tensor<string, []>("op_1871_cast")];
            tensor<string, []> var_1875_equation_0 = const()[name = tensor<string, []>("op_1875_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1875_cast = einsum(equation = var_1875_equation_0, values = (var_1813_cast, var_1745_cast))[name = tensor<string, []>("op_1875_cast")];
            tensor<fp16, []> var_1876_to_fp16 = const()[name = tensor<string, []>("op_1876_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_257_cast = mul(x = var_1875_cast, y = var_1876_to_fp16)[name = tensor<string, []>("aw_chunk_257_cast")];
            tensor<string, []> var_1879_equation_0 = const()[name = tensor<string, []>("op_1879_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1879_cast = einsum(equation = var_1879_equation_0, values = (var_1813_cast, var_1746_cast))[name = tensor<string, []>("op_1879_cast")];
            tensor<fp16, []> var_1880_to_fp16 = const()[name = tensor<string, []>("op_1880_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_259_cast = mul(x = var_1879_cast, y = var_1880_to_fp16)[name = tensor<string, []>("aw_chunk_259_cast")];
            tensor<string, []> var_1883_equation_0 = const()[name = tensor<string, []>("op_1883_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1883_cast = einsum(equation = var_1883_equation_0, values = (var_1813_cast, var_1747_cast))[name = tensor<string, []>("op_1883_cast")];
            tensor<fp16, []> var_1884_to_fp16 = const()[name = tensor<string, []>("op_1884_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_261_cast = mul(x = var_1883_cast, y = var_1884_to_fp16)[name = tensor<string, []>("aw_chunk_261_cast")];
            tensor<string, []> var_1887_equation_0 = const()[name = tensor<string, []>("op_1887_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1887_cast = einsum(equation = var_1887_equation_0, values = (var_1813_cast, var_1748_cast))[name = tensor<string, []>("op_1887_cast")];
            tensor<fp16, []> var_1888_to_fp16 = const()[name = tensor<string, []>("op_1888_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_263_cast = mul(x = var_1887_cast, y = var_1888_to_fp16)[name = tensor<string, []>("aw_chunk_263_cast")];
            tensor<string, []> var_1891_equation_0 = const()[name = tensor<string, []>("op_1891_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1891_cast = einsum(equation = var_1891_equation_0, values = (var_1813_cast, var_1749_cast))[name = tensor<string, []>("op_1891_cast")];
            tensor<fp16, []> var_1892_to_fp16 = const()[name = tensor<string, []>("op_1892_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_265_cast = mul(x = var_1891_cast, y = var_1892_to_fp16)[name = tensor<string, []>("aw_chunk_265_cast")];
            tensor<string, []> var_1895_equation_0 = const()[name = tensor<string, []>("op_1895_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1895_cast = einsum(equation = var_1895_equation_0, values = (var_1813_cast, var_1750_cast))[name = tensor<string, []>("op_1895_cast")];
            tensor<fp16, []> var_1896_to_fp16 = const()[name = tensor<string, []>("op_1896_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_267_cast = mul(x = var_1895_cast, y = var_1896_to_fp16)[name = tensor<string, []>("aw_chunk_267_cast")];
            tensor<string, []> var_1899_equation_0 = const()[name = tensor<string, []>("op_1899_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1899_cast = einsum(equation = var_1899_equation_0, values = (var_1813_cast, var_1751_cast))[name = tensor<string, []>("op_1899_cast")];
            tensor<fp16, []> var_1900_to_fp16 = const()[name = tensor<string, []>("op_1900_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_269_cast = mul(x = var_1899_cast, y = var_1900_to_fp16)[name = tensor<string, []>("aw_chunk_269_cast")];
            tensor<string, []> var_1903_equation_0 = const()[name = tensor<string, []>("op_1903_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1903_cast = einsum(equation = var_1903_equation_0, values = (var_1813_cast, var_1752_cast))[name = tensor<string, []>("op_1903_cast")];
            tensor<fp16, []> var_1904_to_fp16 = const()[name = tensor<string, []>("op_1904_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_271_cast = mul(x = var_1903_cast, y = var_1904_to_fp16)[name = tensor<string, []>("aw_chunk_271_cast")];
            tensor<string, []> var_1907_equation_0 = const()[name = tensor<string, []>("op_1907_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1907_cast = einsum(equation = var_1907_equation_0, values = (var_1817_cast, var_1753_cast))[name = tensor<string, []>("op_1907_cast")];
            tensor<fp16, []> var_1908_to_fp16 = const()[name = tensor<string, []>("op_1908_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_273_cast = mul(x = var_1907_cast, y = var_1908_to_fp16)[name = tensor<string, []>("aw_chunk_273_cast")];
            tensor<string, []> var_1911_equation_0 = const()[name = tensor<string, []>("op_1911_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1911_cast = einsum(equation = var_1911_equation_0, values = (var_1817_cast, var_1754_cast))[name = tensor<string, []>("op_1911_cast")];
            tensor<fp16, []> var_1912_to_fp16 = const()[name = tensor<string, []>("op_1912_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_275_cast = mul(x = var_1911_cast, y = var_1912_to_fp16)[name = tensor<string, []>("aw_chunk_275_cast")];
            tensor<string, []> var_1915_equation_0 = const()[name = tensor<string, []>("op_1915_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1915_cast = einsum(equation = var_1915_equation_0, values = (var_1817_cast, var_1755_cast))[name = tensor<string, []>("op_1915_cast")];
            tensor<fp16, []> var_1916_to_fp16 = const()[name = tensor<string, []>("op_1916_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_277_cast = mul(x = var_1915_cast, y = var_1916_to_fp16)[name = tensor<string, []>("aw_chunk_277_cast")];
            tensor<string, []> var_1919_equation_0 = const()[name = tensor<string, []>("op_1919_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1919_cast = einsum(equation = var_1919_equation_0, values = (var_1817_cast, var_1756_cast))[name = tensor<string, []>("op_1919_cast")];
            tensor<fp16, []> var_1920_to_fp16 = const()[name = tensor<string, []>("op_1920_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_279_cast = mul(x = var_1919_cast, y = var_1920_to_fp16)[name = tensor<string, []>("aw_chunk_279_cast")];
            tensor<string, []> var_1923_equation_0 = const()[name = tensor<string, []>("op_1923_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1923_cast = einsum(equation = var_1923_equation_0, values = (var_1817_cast, var_1757_cast))[name = tensor<string, []>("op_1923_cast")];
            tensor<fp16, []> var_1924_to_fp16 = const()[name = tensor<string, []>("op_1924_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_281_cast = mul(x = var_1923_cast, y = var_1924_to_fp16)[name = tensor<string, []>("aw_chunk_281_cast")];
            tensor<string, []> var_1927_equation_0 = const()[name = tensor<string, []>("op_1927_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1927_cast = einsum(equation = var_1927_equation_0, values = (var_1817_cast, var_1758_cast))[name = tensor<string, []>("op_1927_cast")];
            tensor<fp16, []> var_1928_to_fp16 = const()[name = tensor<string, []>("op_1928_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_283_cast = mul(x = var_1927_cast, y = var_1928_to_fp16)[name = tensor<string, []>("aw_chunk_283_cast")];
            tensor<string, []> var_1931_equation_0 = const()[name = tensor<string, []>("op_1931_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1931_cast = einsum(equation = var_1931_equation_0, values = (var_1817_cast, var_1759_cast))[name = tensor<string, []>("op_1931_cast")];
            tensor<fp16, []> var_1932_to_fp16 = const()[name = tensor<string, []>("op_1932_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_285_cast = mul(x = var_1931_cast, y = var_1932_to_fp16)[name = tensor<string, []>("aw_chunk_285_cast")];
            tensor<string, []> var_1935_equation_0 = const()[name = tensor<string, []>("op_1935_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1935_cast = einsum(equation = var_1935_equation_0, values = (var_1817_cast, var_1760_cast))[name = tensor<string, []>("op_1935_cast")];
            tensor<fp16, []> var_1936_to_fp16 = const()[name = tensor<string, []>("op_1936_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_287_cast = mul(x = var_1935_cast, y = var_1936_to_fp16)[name = tensor<string, []>("aw_chunk_287_cast")];
            tensor<string, []> var_1939_equation_0 = const()[name = tensor<string, []>("op_1939_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1939_cast = einsum(equation = var_1939_equation_0, values = (var_1821_cast, var_1761_cast))[name = tensor<string, []>("op_1939_cast")];
            tensor<fp16, []> var_1940_to_fp16 = const()[name = tensor<string, []>("op_1940_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_289_cast = mul(x = var_1939_cast, y = var_1940_to_fp16)[name = tensor<string, []>("aw_chunk_289_cast")];
            tensor<string, []> var_1943_equation_0 = const()[name = tensor<string, []>("op_1943_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1943_cast = einsum(equation = var_1943_equation_0, values = (var_1821_cast, var_1762_cast))[name = tensor<string, []>("op_1943_cast")];
            tensor<fp16, []> var_1944_to_fp16 = const()[name = tensor<string, []>("op_1944_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_291_cast = mul(x = var_1943_cast, y = var_1944_to_fp16)[name = tensor<string, []>("aw_chunk_291_cast")];
            tensor<string, []> var_1947_equation_0 = const()[name = tensor<string, []>("op_1947_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1947_cast = einsum(equation = var_1947_equation_0, values = (var_1821_cast, var_1763_cast))[name = tensor<string, []>("op_1947_cast")];
            tensor<fp16, []> var_1948_to_fp16 = const()[name = tensor<string, []>("op_1948_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_293_cast = mul(x = var_1947_cast, y = var_1948_to_fp16)[name = tensor<string, []>("aw_chunk_293_cast")];
            tensor<string, []> var_1951_equation_0 = const()[name = tensor<string, []>("op_1951_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1951_cast = einsum(equation = var_1951_equation_0, values = (var_1821_cast, var_1764_cast))[name = tensor<string, []>("op_1951_cast")];
            tensor<fp16, []> var_1952_to_fp16 = const()[name = tensor<string, []>("op_1952_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_295_cast = mul(x = var_1951_cast, y = var_1952_to_fp16)[name = tensor<string, []>("aw_chunk_295_cast")];
            tensor<string, []> var_1955_equation_0 = const()[name = tensor<string, []>("op_1955_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1955_cast = einsum(equation = var_1955_equation_0, values = (var_1821_cast, var_1765_cast))[name = tensor<string, []>("op_1955_cast")];
            tensor<fp16, []> var_1956_to_fp16 = const()[name = tensor<string, []>("op_1956_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_297_cast = mul(x = var_1955_cast, y = var_1956_to_fp16)[name = tensor<string, []>("aw_chunk_297_cast")];
            tensor<string, []> var_1959_equation_0 = const()[name = tensor<string, []>("op_1959_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1959_cast = einsum(equation = var_1959_equation_0, values = (var_1821_cast, var_1766_cast))[name = tensor<string, []>("op_1959_cast")];
            tensor<fp16, []> var_1960_to_fp16 = const()[name = tensor<string, []>("op_1960_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_299_cast = mul(x = var_1959_cast, y = var_1960_to_fp16)[name = tensor<string, []>("aw_chunk_299_cast")];
            tensor<string, []> var_1963_equation_0 = const()[name = tensor<string, []>("op_1963_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1963_cast = einsum(equation = var_1963_equation_0, values = (var_1821_cast, var_1767_cast))[name = tensor<string, []>("op_1963_cast")];
            tensor<fp16, []> var_1964_to_fp16 = const()[name = tensor<string, []>("op_1964_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_301_cast = mul(x = var_1963_cast, y = var_1964_to_fp16)[name = tensor<string, []>("aw_chunk_301_cast")];
            tensor<string, []> var_1967_equation_0 = const()[name = tensor<string, []>("op_1967_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1967_cast = einsum(equation = var_1967_equation_0, values = (var_1821_cast, var_1768_cast))[name = tensor<string, []>("op_1967_cast")];
            tensor<fp16, []> var_1968_to_fp16 = const()[name = tensor<string, []>("op_1968_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_303_cast = mul(x = var_1967_cast, y = var_1968_to_fp16)[name = tensor<string, []>("aw_chunk_303_cast")];
            tensor<string, []> var_1971_equation_0 = const()[name = tensor<string, []>("op_1971_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1971_cast = einsum(equation = var_1971_equation_0, values = (var_1825_cast, var_1769_cast))[name = tensor<string, []>("op_1971_cast")];
            tensor<fp16, []> var_1972_to_fp16 = const()[name = tensor<string, []>("op_1972_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_305_cast = mul(x = var_1971_cast, y = var_1972_to_fp16)[name = tensor<string, []>("aw_chunk_305_cast")];
            tensor<string, []> var_1975_equation_0 = const()[name = tensor<string, []>("op_1975_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1975_cast = einsum(equation = var_1975_equation_0, values = (var_1825_cast, var_1770_cast))[name = tensor<string, []>("op_1975_cast")];
            tensor<fp16, []> var_1976_to_fp16 = const()[name = tensor<string, []>("op_1976_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_307_cast = mul(x = var_1975_cast, y = var_1976_to_fp16)[name = tensor<string, []>("aw_chunk_307_cast")];
            tensor<string, []> var_1979_equation_0 = const()[name = tensor<string, []>("op_1979_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1979_cast = einsum(equation = var_1979_equation_0, values = (var_1825_cast, var_1771_cast))[name = tensor<string, []>("op_1979_cast")];
            tensor<fp16, []> var_1980_to_fp16 = const()[name = tensor<string, []>("op_1980_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_309_cast = mul(x = var_1979_cast, y = var_1980_to_fp16)[name = tensor<string, []>("aw_chunk_309_cast")];
            tensor<string, []> var_1983_equation_0 = const()[name = tensor<string, []>("op_1983_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1983_cast = einsum(equation = var_1983_equation_0, values = (var_1825_cast, var_1772_cast))[name = tensor<string, []>("op_1983_cast")];
            tensor<fp16, []> var_1984_to_fp16 = const()[name = tensor<string, []>("op_1984_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_311_cast = mul(x = var_1983_cast, y = var_1984_to_fp16)[name = tensor<string, []>("aw_chunk_311_cast")];
            tensor<string, []> var_1987_equation_0 = const()[name = tensor<string, []>("op_1987_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1987_cast = einsum(equation = var_1987_equation_0, values = (var_1825_cast, var_1773_cast))[name = tensor<string, []>("op_1987_cast")];
            tensor<fp16, []> var_1988_to_fp16 = const()[name = tensor<string, []>("op_1988_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_313_cast = mul(x = var_1987_cast, y = var_1988_to_fp16)[name = tensor<string, []>("aw_chunk_313_cast")];
            tensor<string, []> var_1991_equation_0 = const()[name = tensor<string, []>("op_1991_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1991_cast = einsum(equation = var_1991_equation_0, values = (var_1825_cast, var_1774_cast))[name = tensor<string, []>("op_1991_cast")];
            tensor<fp16, []> var_1992_to_fp16 = const()[name = tensor<string, []>("op_1992_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_315_cast = mul(x = var_1991_cast, y = var_1992_to_fp16)[name = tensor<string, []>("aw_chunk_315_cast")];
            tensor<string, []> var_1995_equation_0 = const()[name = tensor<string, []>("op_1995_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1995_cast = einsum(equation = var_1995_equation_0, values = (var_1825_cast, var_1775_cast))[name = tensor<string, []>("op_1995_cast")];
            tensor<fp16, []> var_1996_to_fp16 = const()[name = tensor<string, []>("op_1996_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_317_cast = mul(x = var_1995_cast, y = var_1996_to_fp16)[name = tensor<string, []>("aw_chunk_317_cast")];
            tensor<string, []> var_1999_equation_0 = const()[name = tensor<string, []>("op_1999_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1999_cast = einsum(equation = var_1999_equation_0, values = (var_1825_cast, var_1776_cast))[name = tensor<string, []>("op_1999_cast")];
            tensor<fp16, []> var_2000_to_fp16 = const()[name = tensor<string, []>("op_2000_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_319_cast = mul(x = var_1999_cast, y = var_2000_to_fp16)[name = tensor<string, []>("aw_chunk_319_cast")];
            tensor<string, []> var_2003_equation_0 = const()[name = tensor<string, []>("op_2003_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2003_cast = einsum(equation = var_2003_equation_0, values = (var_1829_cast, var_1777_cast))[name = tensor<string, []>("op_2003_cast")];
            tensor<fp16, []> var_2004_to_fp16 = const()[name = tensor<string, []>("op_2004_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_321_cast = mul(x = var_2003_cast, y = var_2004_to_fp16)[name = tensor<string, []>("aw_chunk_321_cast")];
            tensor<string, []> var_2007_equation_0 = const()[name = tensor<string, []>("op_2007_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2007_cast = einsum(equation = var_2007_equation_0, values = (var_1829_cast, var_1778_cast))[name = tensor<string, []>("op_2007_cast")];
            tensor<fp16, []> var_2008_to_fp16 = const()[name = tensor<string, []>("op_2008_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_323_cast = mul(x = var_2007_cast, y = var_2008_to_fp16)[name = tensor<string, []>("aw_chunk_323_cast")];
            tensor<string, []> var_2011_equation_0 = const()[name = tensor<string, []>("op_2011_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2011_cast = einsum(equation = var_2011_equation_0, values = (var_1829_cast, var_1779_cast))[name = tensor<string, []>("op_2011_cast")];
            tensor<fp16, []> var_2012_to_fp16 = const()[name = tensor<string, []>("op_2012_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_325_cast = mul(x = var_2011_cast, y = var_2012_to_fp16)[name = tensor<string, []>("aw_chunk_325_cast")];
            tensor<string, []> var_2015_equation_0 = const()[name = tensor<string, []>("op_2015_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2015_cast = einsum(equation = var_2015_equation_0, values = (var_1829_cast, var_1780_cast))[name = tensor<string, []>("op_2015_cast")];
            tensor<fp16, []> var_2016_to_fp16 = const()[name = tensor<string, []>("op_2016_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_327_cast = mul(x = var_2015_cast, y = var_2016_to_fp16)[name = tensor<string, []>("aw_chunk_327_cast")];
            tensor<string, []> var_2019_equation_0 = const()[name = tensor<string, []>("op_2019_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2019_cast = einsum(equation = var_2019_equation_0, values = (var_1829_cast, var_1781_cast))[name = tensor<string, []>("op_2019_cast")];
            tensor<fp16, []> var_2020_to_fp16 = const()[name = tensor<string, []>("op_2020_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_329_cast = mul(x = var_2019_cast, y = var_2020_to_fp16)[name = tensor<string, []>("aw_chunk_329_cast")];
            tensor<string, []> var_2023_equation_0 = const()[name = tensor<string, []>("op_2023_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2023_cast = einsum(equation = var_2023_equation_0, values = (var_1829_cast, var_1782_cast))[name = tensor<string, []>("op_2023_cast")];
            tensor<fp16, []> var_2024_to_fp16 = const()[name = tensor<string, []>("op_2024_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_331_cast = mul(x = var_2023_cast, y = var_2024_to_fp16)[name = tensor<string, []>("aw_chunk_331_cast")];
            tensor<string, []> var_2027_equation_0 = const()[name = tensor<string, []>("op_2027_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2027_cast = einsum(equation = var_2027_equation_0, values = (var_1829_cast, var_1783_cast))[name = tensor<string, []>("op_2027_cast")];
            tensor<fp16, []> var_2028_to_fp16 = const()[name = tensor<string, []>("op_2028_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_333_cast = mul(x = var_2027_cast, y = var_2028_to_fp16)[name = tensor<string, []>("aw_chunk_333_cast")];
            tensor<string, []> var_2031_equation_0 = const()[name = tensor<string, []>("op_2031_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2031_cast = einsum(equation = var_2031_equation_0, values = (var_1829_cast, var_1784_cast))[name = tensor<string, []>("op_2031_cast")];
            tensor<fp16, []> var_2032_to_fp16 = const()[name = tensor<string, []>("op_2032_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_335_cast = mul(x = var_2031_cast, y = var_2032_to_fp16)[name = tensor<string, []>("aw_chunk_335_cast")];
            tensor<string, []> var_2035_equation_0 = const()[name = tensor<string, []>("op_2035_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2035_cast = einsum(equation = var_2035_equation_0, values = (var_1833_cast, var_1785_cast))[name = tensor<string, []>("op_2035_cast")];
            tensor<fp16, []> var_2036_to_fp16 = const()[name = tensor<string, []>("op_2036_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_337_cast = mul(x = var_2035_cast, y = var_2036_to_fp16)[name = tensor<string, []>("aw_chunk_337_cast")];
            tensor<string, []> var_2039_equation_0 = const()[name = tensor<string, []>("op_2039_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2039_cast = einsum(equation = var_2039_equation_0, values = (var_1833_cast, var_1786_cast))[name = tensor<string, []>("op_2039_cast")];
            tensor<fp16, []> var_2040_to_fp16 = const()[name = tensor<string, []>("op_2040_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_339_cast = mul(x = var_2039_cast, y = var_2040_to_fp16)[name = tensor<string, []>("aw_chunk_339_cast")];
            tensor<string, []> var_2043_equation_0 = const()[name = tensor<string, []>("op_2043_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2043_cast = einsum(equation = var_2043_equation_0, values = (var_1833_cast, var_1787_cast))[name = tensor<string, []>("op_2043_cast")];
            tensor<fp16, []> var_2044_to_fp16 = const()[name = tensor<string, []>("op_2044_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_341_cast = mul(x = var_2043_cast, y = var_2044_to_fp16)[name = tensor<string, []>("aw_chunk_341_cast")];
            tensor<string, []> var_2047_equation_0 = const()[name = tensor<string, []>("op_2047_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2047_cast = einsum(equation = var_2047_equation_0, values = (var_1833_cast, var_1788_cast))[name = tensor<string, []>("op_2047_cast")];
            tensor<fp16, []> var_2048_to_fp16 = const()[name = tensor<string, []>("op_2048_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_343_cast = mul(x = var_2047_cast, y = var_2048_to_fp16)[name = tensor<string, []>("aw_chunk_343_cast")];
            tensor<string, []> var_2051_equation_0 = const()[name = tensor<string, []>("op_2051_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2051_cast = einsum(equation = var_2051_equation_0, values = (var_1833_cast, var_1789_cast))[name = tensor<string, []>("op_2051_cast")];
            tensor<fp16, []> var_2052_to_fp16 = const()[name = tensor<string, []>("op_2052_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_345_cast = mul(x = var_2051_cast, y = var_2052_to_fp16)[name = tensor<string, []>("aw_chunk_345_cast")];
            tensor<string, []> var_2055_equation_0 = const()[name = tensor<string, []>("op_2055_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2055_cast = einsum(equation = var_2055_equation_0, values = (var_1833_cast, var_1790_cast))[name = tensor<string, []>("op_2055_cast")];
            tensor<fp16, []> var_2056_to_fp16 = const()[name = tensor<string, []>("op_2056_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_347_cast = mul(x = var_2055_cast, y = var_2056_to_fp16)[name = tensor<string, []>("aw_chunk_347_cast")];
            tensor<string, []> var_2059_equation_0 = const()[name = tensor<string, []>("op_2059_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2059_cast = einsum(equation = var_2059_equation_0, values = (var_1833_cast, var_1791_cast))[name = tensor<string, []>("op_2059_cast")];
            tensor<fp16, []> var_2060_to_fp16 = const()[name = tensor<string, []>("op_2060_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_349_cast = mul(x = var_2059_cast, y = var_2060_to_fp16)[name = tensor<string, []>("aw_chunk_349_cast")];
            tensor<string, []> var_2063_equation_0 = const()[name = tensor<string, []>("op_2063_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2063_cast = einsum(equation = var_2063_equation_0, values = (var_1833_cast, var_1792_cast))[name = tensor<string, []>("op_2063_cast")];
            tensor<fp16, []> var_2064_to_fp16 = const()[name = tensor<string, []>("op_2064_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_351_cast = mul(x = var_2063_cast, y = var_2064_to_fp16)[name = tensor<string, []>("aw_chunk_351_cast")];
            tensor<string, []> var_2067_equation_0 = const()[name = tensor<string, []>("op_2067_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2067_cast = einsum(equation = var_2067_equation_0, values = (var_1837_cast, var_1793_cast))[name = tensor<string, []>("op_2067_cast")];
            tensor<fp16, []> var_2068_to_fp16 = const()[name = tensor<string, []>("op_2068_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_353_cast = mul(x = var_2067_cast, y = var_2068_to_fp16)[name = tensor<string, []>("aw_chunk_353_cast")];
            tensor<string, []> var_2071_equation_0 = const()[name = tensor<string, []>("op_2071_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2071_cast = einsum(equation = var_2071_equation_0, values = (var_1837_cast, var_1794_cast))[name = tensor<string, []>("op_2071_cast")];
            tensor<fp16, []> var_2072_to_fp16 = const()[name = tensor<string, []>("op_2072_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_355_cast = mul(x = var_2071_cast, y = var_2072_to_fp16)[name = tensor<string, []>("aw_chunk_355_cast")];
            tensor<string, []> var_2075_equation_0 = const()[name = tensor<string, []>("op_2075_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2075_cast = einsum(equation = var_2075_equation_0, values = (var_1837_cast, var_1795_cast))[name = tensor<string, []>("op_2075_cast")];
            tensor<fp16, []> var_2076_to_fp16 = const()[name = tensor<string, []>("op_2076_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_357_cast = mul(x = var_2075_cast, y = var_2076_to_fp16)[name = tensor<string, []>("aw_chunk_357_cast")];
            tensor<string, []> var_2079_equation_0 = const()[name = tensor<string, []>("op_2079_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2079_cast = einsum(equation = var_2079_equation_0, values = (var_1837_cast, var_1796_cast))[name = tensor<string, []>("op_2079_cast")];
            tensor<fp16, []> var_2080_to_fp16 = const()[name = tensor<string, []>("op_2080_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_359_cast = mul(x = var_2079_cast, y = var_2080_to_fp16)[name = tensor<string, []>("aw_chunk_359_cast")];
            tensor<string, []> var_2083_equation_0 = const()[name = tensor<string, []>("op_2083_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2083_cast = einsum(equation = var_2083_equation_0, values = (var_1837_cast, var_1797_cast))[name = tensor<string, []>("op_2083_cast")];
            tensor<fp16, []> var_2084_to_fp16 = const()[name = tensor<string, []>("op_2084_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_361_cast = mul(x = var_2083_cast, y = var_2084_to_fp16)[name = tensor<string, []>("aw_chunk_361_cast")];
            tensor<string, []> var_2087_equation_0 = const()[name = tensor<string, []>("op_2087_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2087_cast = einsum(equation = var_2087_equation_0, values = (var_1837_cast, var_1798_cast))[name = tensor<string, []>("op_2087_cast")];
            tensor<fp16, []> var_2088_to_fp16 = const()[name = tensor<string, []>("op_2088_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_363_cast = mul(x = var_2087_cast, y = var_2088_to_fp16)[name = tensor<string, []>("aw_chunk_363_cast")];
            tensor<string, []> var_2091_equation_0 = const()[name = tensor<string, []>("op_2091_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2091_cast = einsum(equation = var_2091_equation_0, values = (var_1837_cast, var_1799_cast))[name = tensor<string, []>("op_2091_cast")];
            tensor<fp16, []> var_2092_to_fp16 = const()[name = tensor<string, []>("op_2092_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_365_cast = mul(x = var_2091_cast, y = var_2092_to_fp16)[name = tensor<string, []>("aw_chunk_365_cast")];
            tensor<string, []> var_2095_equation_0 = const()[name = tensor<string, []>("op_2095_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2095_cast = einsum(equation = var_2095_equation_0, values = (var_1837_cast, var_1800_cast))[name = tensor<string, []>("op_2095_cast")];
            tensor<fp16, []> var_2096_to_fp16 = const()[name = tensor<string, []>("op_2096_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_367_cast = mul(x = var_2095_cast, y = var_2096_to_fp16)[name = tensor<string, []>("aw_chunk_367_cast")];
            tensor<string, []> var_2099_equation_0 = const()[name = tensor<string, []>("op_2099_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2099_cast = einsum(equation = var_2099_equation_0, values = (var_1841_cast, var_1801_cast))[name = tensor<string, []>("op_2099_cast")];
            tensor<fp16, []> var_2100_to_fp16 = const()[name = tensor<string, []>("op_2100_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_369_cast = mul(x = var_2099_cast, y = var_2100_to_fp16)[name = tensor<string, []>("aw_chunk_369_cast")];
            tensor<string, []> var_2103_equation_0 = const()[name = tensor<string, []>("op_2103_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2103_cast = einsum(equation = var_2103_equation_0, values = (var_1841_cast, var_1802_cast))[name = tensor<string, []>("op_2103_cast")];
            tensor<fp16, []> var_2104_to_fp16 = const()[name = tensor<string, []>("op_2104_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_371_cast = mul(x = var_2103_cast, y = var_2104_to_fp16)[name = tensor<string, []>("aw_chunk_371_cast")];
            tensor<string, []> var_2107_equation_0 = const()[name = tensor<string, []>("op_2107_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2107_cast = einsum(equation = var_2107_equation_0, values = (var_1841_cast, var_1803_cast))[name = tensor<string, []>("op_2107_cast")];
            tensor<fp16, []> var_2108_to_fp16 = const()[name = tensor<string, []>("op_2108_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_373_cast = mul(x = var_2107_cast, y = var_2108_to_fp16)[name = tensor<string, []>("aw_chunk_373_cast")];
            tensor<string, []> var_2111_equation_0 = const()[name = tensor<string, []>("op_2111_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2111_cast = einsum(equation = var_2111_equation_0, values = (var_1841_cast, var_1804_cast))[name = tensor<string, []>("op_2111_cast")];
            tensor<fp16, []> var_2112_to_fp16 = const()[name = tensor<string, []>("op_2112_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_375_cast = mul(x = var_2111_cast, y = var_2112_to_fp16)[name = tensor<string, []>("aw_chunk_375_cast")];
            tensor<string, []> var_2115_equation_0 = const()[name = tensor<string, []>("op_2115_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2115_cast = einsum(equation = var_2115_equation_0, values = (var_1841_cast, var_1805_cast))[name = tensor<string, []>("op_2115_cast")];
            tensor<fp16, []> var_2116_to_fp16 = const()[name = tensor<string, []>("op_2116_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_377_cast = mul(x = var_2115_cast, y = var_2116_to_fp16)[name = tensor<string, []>("aw_chunk_377_cast")];
            tensor<string, []> var_2119_equation_0 = const()[name = tensor<string, []>("op_2119_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2119_cast = einsum(equation = var_2119_equation_0, values = (var_1841_cast, var_1806_cast))[name = tensor<string, []>("op_2119_cast")];
            tensor<fp16, []> var_2120_to_fp16 = const()[name = tensor<string, []>("op_2120_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_379_cast = mul(x = var_2119_cast, y = var_2120_to_fp16)[name = tensor<string, []>("aw_chunk_379_cast")];
            tensor<string, []> var_2123_equation_0 = const()[name = tensor<string, []>("op_2123_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2123_cast = einsum(equation = var_2123_equation_0, values = (var_1841_cast, var_1807_cast))[name = tensor<string, []>("op_2123_cast")];
            tensor<fp16, []> var_2124_to_fp16 = const()[name = tensor<string, []>("op_2124_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_381_cast = mul(x = var_2123_cast, y = var_2124_to_fp16)[name = tensor<string, []>("aw_chunk_381_cast")];
            tensor<string, []> var_2127_equation_0 = const()[name = tensor<string, []>("op_2127_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_2127_cast = einsum(equation = var_2127_equation_0, values = (var_1841_cast, var_1808_cast))[name = tensor<string, []>("op_2127_cast")];
            tensor<fp16, []> var_2128_to_fp16 = const()[name = tensor<string, []>("op_2128_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_383_cast = mul(x = var_2127_cast, y = var_2128_to_fp16)[name = tensor<string, []>("aw_chunk_383_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2130_cast = softmax(axis = var_123, x = aw_chunk_257_cast)[name = tensor<string, []>("op_2130_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2131_cast = softmax(axis = var_123, x = aw_chunk_259_cast)[name = tensor<string, []>("op_2131_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2132_cast = softmax(axis = var_123, x = aw_chunk_261_cast)[name = tensor<string, []>("op_2132_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2133_cast = softmax(axis = var_123, x = aw_chunk_263_cast)[name = tensor<string, []>("op_2133_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2134_cast = softmax(axis = var_123, x = aw_chunk_265_cast)[name = tensor<string, []>("op_2134_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2135_cast = softmax(axis = var_123, x = aw_chunk_267_cast)[name = tensor<string, []>("op_2135_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2136_cast = softmax(axis = var_123, x = aw_chunk_269_cast)[name = tensor<string, []>("op_2136_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2137_cast = softmax(axis = var_123, x = aw_chunk_271_cast)[name = tensor<string, []>("op_2137_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2138_cast = softmax(axis = var_123, x = aw_chunk_273_cast)[name = tensor<string, []>("op_2138_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2139_cast = softmax(axis = var_123, x = aw_chunk_275_cast)[name = tensor<string, []>("op_2139_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2140_cast = softmax(axis = var_123, x = aw_chunk_277_cast)[name = tensor<string, []>("op_2140_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2141_cast = softmax(axis = var_123, x = aw_chunk_279_cast)[name = tensor<string, []>("op_2141_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2142_cast = softmax(axis = var_123, x = aw_chunk_281_cast)[name = tensor<string, []>("op_2142_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2143_cast = softmax(axis = var_123, x = aw_chunk_283_cast)[name = tensor<string, []>("op_2143_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2144_cast = softmax(axis = var_123, x = aw_chunk_285_cast)[name = tensor<string, []>("op_2144_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2145_cast = softmax(axis = var_123, x = aw_chunk_287_cast)[name = tensor<string, []>("op_2145_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2146_cast = softmax(axis = var_123, x = aw_chunk_289_cast)[name = tensor<string, []>("op_2146_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2147_cast = softmax(axis = var_123, x = aw_chunk_291_cast)[name = tensor<string, []>("op_2147_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2148_cast = softmax(axis = var_123, x = aw_chunk_293_cast)[name = tensor<string, []>("op_2148_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2149_cast = softmax(axis = var_123, x = aw_chunk_295_cast)[name = tensor<string, []>("op_2149_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2150_cast = softmax(axis = var_123, x = aw_chunk_297_cast)[name = tensor<string, []>("op_2150_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2151_cast = softmax(axis = var_123, x = aw_chunk_299_cast)[name = tensor<string, []>("op_2151_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2152_cast = softmax(axis = var_123, x = aw_chunk_301_cast)[name = tensor<string, []>("op_2152_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2153_cast = softmax(axis = var_123, x = aw_chunk_303_cast)[name = tensor<string, []>("op_2153_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2154_cast = softmax(axis = var_123, x = aw_chunk_305_cast)[name = tensor<string, []>("op_2154_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2155_cast = softmax(axis = var_123, x = aw_chunk_307_cast)[name = tensor<string, []>("op_2155_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2156_cast = softmax(axis = var_123, x = aw_chunk_309_cast)[name = tensor<string, []>("op_2156_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2157_cast = softmax(axis = var_123, x = aw_chunk_311_cast)[name = tensor<string, []>("op_2157_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2158_cast = softmax(axis = var_123, x = aw_chunk_313_cast)[name = tensor<string, []>("op_2158_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2159_cast = softmax(axis = var_123, x = aw_chunk_315_cast)[name = tensor<string, []>("op_2159_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2160_cast = softmax(axis = var_123, x = aw_chunk_317_cast)[name = tensor<string, []>("op_2160_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2161_cast = softmax(axis = var_123, x = aw_chunk_319_cast)[name = tensor<string, []>("op_2161_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2162_cast = softmax(axis = var_123, x = aw_chunk_321_cast)[name = tensor<string, []>("op_2162_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2163_cast = softmax(axis = var_123, x = aw_chunk_323_cast)[name = tensor<string, []>("op_2163_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2164_cast = softmax(axis = var_123, x = aw_chunk_325_cast)[name = tensor<string, []>("op_2164_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2165_cast = softmax(axis = var_123, x = aw_chunk_327_cast)[name = tensor<string, []>("op_2165_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2166_cast = softmax(axis = var_123, x = aw_chunk_329_cast)[name = tensor<string, []>("op_2166_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2167_cast = softmax(axis = var_123, x = aw_chunk_331_cast)[name = tensor<string, []>("op_2167_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2168_cast = softmax(axis = var_123, x = aw_chunk_333_cast)[name = tensor<string, []>("op_2168_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2169_cast = softmax(axis = var_123, x = aw_chunk_335_cast)[name = tensor<string, []>("op_2169_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2170_cast = softmax(axis = var_123, x = aw_chunk_337_cast)[name = tensor<string, []>("op_2170_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2171_cast = softmax(axis = var_123, x = aw_chunk_339_cast)[name = tensor<string, []>("op_2171_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2172_cast = softmax(axis = var_123, x = aw_chunk_341_cast)[name = tensor<string, []>("op_2172_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2173_cast = softmax(axis = var_123, x = aw_chunk_343_cast)[name = tensor<string, []>("op_2173_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2174_cast = softmax(axis = var_123, x = aw_chunk_345_cast)[name = tensor<string, []>("op_2174_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2175_cast = softmax(axis = var_123, x = aw_chunk_347_cast)[name = tensor<string, []>("op_2175_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2176_cast = softmax(axis = var_123, x = aw_chunk_349_cast)[name = tensor<string, []>("op_2176_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2177_cast = softmax(axis = var_123, x = aw_chunk_351_cast)[name = tensor<string, []>("op_2177_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2178_cast = softmax(axis = var_123, x = aw_chunk_353_cast)[name = tensor<string, []>("op_2178_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2179_cast = softmax(axis = var_123, x = aw_chunk_355_cast)[name = tensor<string, []>("op_2179_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2180_cast = softmax(axis = var_123, x = aw_chunk_357_cast)[name = tensor<string, []>("op_2180_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2181_cast = softmax(axis = var_123, x = aw_chunk_359_cast)[name = tensor<string, []>("op_2181_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2182_cast = softmax(axis = var_123, x = aw_chunk_361_cast)[name = tensor<string, []>("op_2182_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2183_cast = softmax(axis = var_123, x = aw_chunk_363_cast)[name = tensor<string, []>("op_2183_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2184_cast = softmax(axis = var_123, x = aw_chunk_365_cast)[name = tensor<string, []>("op_2184_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2185_cast = softmax(axis = var_123, x = aw_chunk_367_cast)[name = tensor<string, []>("op_2185_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2186_cast = softmax(axis = var_123, x = aw_chunk_369_cast)[name = tensor<string, []>("op_2186_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2187_cast = softmax(axis = var_123, x = aw_chunk_371_cast)[name = tensor<string, []>("op_2187_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2188_cast = softmax(axis = var_123, x = aw_chunk_373_cast)[name = tensor<string, []>("op_2188_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2189_cast = softmax(axis = var_123, x = aw_chunk_375_cast)[name = tensor<string, []>("op_2189_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2190_cast = softmax(axis = var_123, x = aw_chunk_377_cast)[name = tensor<string, []>("op_2190_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2191_cast = softmax(axis = var_123, x = aw_chunk_379_cast)[name = tensor<string, []>("op_2191_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2192_cast = softmax(axis = var_123, x = aw_chunk_381_cast)[name = tensor<string, []>("op_2192_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_2193_cast = softmax(axis = var_123, x = aw_chunk_383_cast)[name = tensor<string, []>("op_2193_cast")];
            tensor<string, []> var_2195_equation_0 = const()[name = tensor<string, []>("op_2195_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2195_cast = einsum(equation = var_2195_equation_0, values = (var_1843_cast, var_2130_cast))[name = tensor<string, []>("op_2195_cast")];
            tensor<string, []> var_2197_equation_0 = const()[name = tensor<string, []>("op_2197_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2197_cast = einsum(equation = var_2197_equation_0, values = (var_1843_cast, var_2131_cast))[name = tensor<string, []>("op_2197_cast")];
            tensor<string, []> var_2199_equation_0 = const()[name = tensor<string, []>("op_2199_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2199_cast = einsum(equation = var_2199_equation_0, values = (var_1843_cast, var_2132_cast))[name = tensor<string, []>("op_2199_cast")];
            tensor<string, []> var_2201_equation_0 = const()[name = tensor<string, []>("op_2201_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2201_cast = einsum(equation = var_2201_equation_0, values = (var_1843_cast, var_2133_cast))[name = tensor<string, []>("op_2201_cast")];
            tensor<string, []> var_2203_equation_0 = const()[name = tensor<string, []>("op_2203_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2203_cast = einsum(equation = var_2203_equation_0, values = (var_1843_cast, var_2134_cast))[name = tensor<string, []>("op_2203_cast")];
            tensor<string, []> var_2205_equation_0 = const()[name = tensor<string, []>("op_2205_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2205_cast = einsum(equation = var_2205_equation_0, values = (var_1843_cast, var_2135_cast))[name = tensor<string, []>("op_2205_cast")];
            tensor<string, []> var_2207_equation_0 = const()[name = tensor<string, []>("op_2207_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2207_cast = einsum(equation = var_2207_equation_0, values = (var_1843_cast, var_2136_cast))[name = tensor<string, []>("op_2207_cast")];
            tensor<string, []> var_2209_equation_0 = const()[name = tensor<string, []>("op_2209_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2209_cast = einsum(equation = var_2209_equation_0, values = (var_1843_cast, var_2137_cast))[name = tensor<string, []>("op_2209_cast")];
            tensor<string, []> var_2211_equation_0 = const()[name = tensor<string, []>("op_2211_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2211_cast = einsum(equation = var_2211_equation_0, values = (var_1847_cast, var_2138_cast))[name = tensor<string, []>("op_2211_cast")];
            tensor<string, []> var_2213_equation_0 = const()[name = tensor<string, []>("op_2213_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2213_cast = einsum(equation = var_2213_equation_0, values = (var_1847_cast, var_2139_cast))[name = tensor<string, []>("op_2213_cast")];
            tensor<string, []> var_2215_equation_0 = const()[name = tensor<string, []>("op_2215_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2215_cast = einsum(equation = var_2215_equation_0, values = (var_1847_cast, var_2140_cast))[name = tensor<string, []>("op_2215_cast")];
            tensor<string, []> var_2217_equation_0 = const()[name = tensor<string, []>("op_2217_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2217_cast = einsum(equation = var_2217_equation_0, values = (var_1847_cast, var_2141_cast))[name = tensor<string, []>("op_2217_cast")];
            tensor<string, []> var_2219_equation_0 = const()[name = tensor<string, []>("op_2219_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2219_cast = einsum(equation = var_2219_equation_0, values = (var_1847_cast, var_2142_cast))[name = tensor<string, []>("op_2219_cast")];
            tensor<string, []> var_2221_equation_0 = const()[name = tensor<string, []>("op_2221_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2221_cast = einsum(equation = var_2221_equation_0, values = (var_1847_cast, var_2143_cast))[name = tensor<string, []>("op_2221_cast")];
            tensor<string, []> var_2223_equation_0 = const()[name = tensor<string, []>("op_2223_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2223_cast = einsum(equation = var_2223_equation_0, values = (var_1847_cast, var_2144_cast))[name = tensor<string, []>("op_2223_cast")];
            tensor<string, []> var_2225_equation_0 = const()[name = tensor<string, []>("op_2225_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2225_cast = einsum(equation = var_2225_equation_0, values = (var_1847_cast, var_2145_cast))[name = tensor<string, []>("op_2225_cast")];
            tensor<string, []> var_2227_equation_0 = const()[name = tensor<string, []>("op_2227_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2227_cast = einsum(equation = var_2227_equation_0, values = (var_1851_cast, var_2146_cast))[name = tensor<string, []>("op_2227_cast")];
            tensor<string, []> var_2229_equation_0 = const()[name = tensor<string, []>("op_2229_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2229_cast = einsum(equation = var_2229_equation_0, values = (var_1851_cast, var_2147_cast))[name = tensor<string, []>("op_2229_cast")];
            tensor<string, []> var_2231_equation_0 = const()[name = tensor<string, []>("op_2231_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2231_cast = einsum(equation = var_2231_equation_0, values = (var_1851_cast, var_2148_cast))[name = tensor<string, []>("op_2231_cast")];
            tensor<string, []> var_2233_equation_0 = const()[name = tensor<string, []>("op_2233_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2233_cast = einsum(equation = var_2233_equation_0, values = (var_1851_cast, var_2149_cast))[name = tensor<string, []>("op_2233_cast")];
            tensor<string, []> var_2235_equation_0 = const()[name = tensor<string, []>("op_2235_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2235_cast = einsum(equation = var_2235_equation_0, values = (var_1851_cast, var_2150_cast))[name = tensor<string, []>("op_2235_cast")];
            tensor<string, []> var_2237_equation_0 = const()[name = tensor<string, []>("op_2237_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2237_cast = einsum(equation = var_2237_equation_0, values = (var_1851_cast, var_2151_cast))[name = tensor<string, []>("op_2237_cast")];
            tensor<string, []> var_2239_equation_0 = const()[name = tensor<string, []>("op_2239_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2239_cast = einsum(equation = var_2239_equation_0, values = (var_1851_cast, var_2152_cast))[name = tensor<string, []>("op_2239_cast")];
            tensor<string, []> var_2241_equation_0 = const()[name = tensor<string, []>("op_2241_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2241_cast = einsum(equation = var_2241_equation_0, values = (var_1851_cast, var_2153_cast))[name = tensor<string, []>("op_2241_cast")];
            tensor<string, []> var_2243_equation_0 = const()[name = tensor<string, []>("op_2243_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2243_cast = einsum(equation = var_2243_equation_0, values = (var_1855_cast, var_2154_cast))[name = tensor<string, []>("op_2243_cast")];
            tensor<string, []> var_2245_equation_0 = const()[name = tensor<string, []>("op_2245_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2245_cast = einsum(equation = var_2245_equation_0, values = (var_1855_cast, var_2155_cast))[name = tensor<string, []>("op_2245_cast")];
            tensor<string, []> var_2247_equation_0 = const()[name = tensor<string, []>("op_2247_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2247_cast = einsum(equation = var_2247_equation_0, values = (var_1855_cast, var_2156_cast))[name = tensor<string, []>("op_2247_cast")];
            tensor<string, []> var_2249_equation_0 = const()[name = tensor<string, []>("op_2249_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2249_cast = einsum(equation = var_2249_equation_0, values = (var_1855_cast, var_2157_cast))[name = tensor<string, []>("op_2249_cast")];
            tensor<string, []> var_2251_equation_0 = const()[name = tensor<string, []>("op_2251_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2251_cast = einsum(equation = var_2251_equation_0, values = (var_1855_cast, var_2158_cast))[name = tensor<string, []>("op_2251_cast")];
            tensor<string, []> var_2253_equation_0 = const()[name = tensor<string, []>("op_2253_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2253_cast = einsum(equation = var_2253_equation_0, values = (var_1855_cast, var_2159_cast))[name = tensor<string, []>("op_2253_cast")];
            tensor<string, []> var_2255_equation_0 = const()[name = tensor<string, []>("op_2255_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2255_cast = einsum(equation = var_2255_equation_0, values = (var_1855_cast, var_2160_cast))[name = tensor<string, []>("op_2255_cast")];
            tensor<string, []> var_2257_equation_0 = const()[name = tensor<string, []>("op_2257_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2257_cast = einsum(equation = var_2257_equation_0, values = (var_1855_cast, var_2161_cast))[name = tensor<string, []>("op_2257_cast")];
            tensor<string, []> var_2259_equation_0 = const()[name = tensor<string, []>("op_2259_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2259_cast = einsum(equation = var_2259_equation_0, values = (var_1859_cast, var_2162_cast))[name = tensor<string, []>("op_2259_cast")];
            tensor<string, []> var_2261_equation_0 = const()[name = tensor<string, []>("op_2261_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2261_cast = einsum(equation = var_2261_equation_0, values = (var_1859_cast, var_2163_cast))[name = tensor<string, []>("op_2261_cast")];
            tensor<string, []> var_2263_equation_0 = const()[name = tensor<string, []>("op_2263_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2263_cast = einsum(equation = var_2263_equation_0, values = (var_1859_cast, var_2164_cast))[name = tensor<string, []>("op_2263_cast")];
            tensor<string, []> var_2265_equation_0 = const()[name = tensor<string, []>("op_2265_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2265_cast = einsum(equation = var_2265_equation_0, values = (var_1859_cast, var_2165_cast))[name = tensor<string, []>("op_2265_cast")];
            tensor<string, []> var_2267_equation_0 = const()[name = tensor<string, []>("op_2267_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2267_cast = einsum(equation = var_2267_equation_0, values = (var_1859_cast, var_2166_cast))[name = tensor<string, []>("op_2267_cast")];
            tensor<string, []> var_2269_equation_0 = const()[name = tensor<string, []>("op_2269_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2269_cast = einsum(equation = var_2269_equation_0, values = (var_1859_cast, var_2167_cast))[name = tensor<string, []>("op_2269_cast")];
            tensor<string, []> var_2271_equation_0 = const()[name = tensor<string, []>("op_2271_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2271_cast = einsum(equation = var_2271_equation_0, values = (var_1859_cast, var_2168_cast))[name = tensor<string, []>("op_2271_cast")];
            tensor<string, []> var_2273_equation_0 = const()[name = tensor<string, []>("op_2273_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2273_cast = einsum(equation = var_2273_equation_0, values = (var_1859_cast, var_2169_cast))[name = tensor<string, []>("op_2273_cast")];
            tensor<string, []> var_2275_equation_0 = const()[name = tensor<string, []>("op_2275_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2275_cast = einsum(equation = var_2275_equation_0, values = (var_1863_cast, var_2170_cast))[name = tensor<string, []>("op_2275_cast")];
            tensor<string, []> var_2277_equation_0 = const()[name = tensor<string, []>("op_2277_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2277_cast = einsum(equation = var_2277_equation_0, values = (var_1863_cast, var_2171_cast))[name = tensor<string, []>("op_2277_cast")];
            tensor<string, []> var_2279_equation_0 = const()[name = tensor<string, []>("op_2279_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2279_cast = einsum(equation = var_2279_equation_0, values = (var_1863_cast, var_2172_cast))[name = tensor<string, []>("op_2279_cast")];
            tensor<string, []> var_2281_equation_0 = const()[name = tensor<string, []>("op_2281_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2281_cast = einsum(equation = var_2281_equation_0, values = (var_1863_cast, var_2173_cast))[name = tensor<string, []>("op_2281_cast")];
            tensor<string, []> var_2283_equation_0 = const()[name = tensor<string, []>("op_2283_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2283_cast = einsum(equation = var_2283_equation_0, values = (var_1863_cast, var_2174_cast))[name = tensor<string, []>("op_2283_cast")];
            tensor<string, []> var_2285_equation_0 = const()[name = tensor<string, []>("op_2285_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2285_cast = einsum(equation = var_2285_equation_0, values = (var_1863_cast, var_2175_cast))[name = tensor<string, []>("op_2285_cast")];
            tensor<string, []> var_2287_equation_0 = const()[name = tensor<string, []>("op_2287_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2287_cast = einsum(equation = var_2287_equation_0, values = (var_1863_cast, var_2176_cast))[name = tensor<string, []>("op_2287_cast")];
            tensor<string, []> var_2289_equation_0 = const()[name = tensor<string, []>("op_2289_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2289_cast = einsum(equation = var_2289_equation_0, values = (var_1863_cast, var_2177_cast))[name = tensor<string, []>("op_2289_cast")];
            tensor<string, []> var_2291_equation_0 = const()[name = tensor<string, []>("op_2291_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2291_cast = einsum(equation = var_2291_equation_0, values = (var_1867_cast, var_2178_cast))[name = tensor<string, []>("op_2291_cast")];
            tensor<string, []> var_2293_equation_0 = const()[name = tensor<string, []>("op_2293_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2293_cast = einsum(equation = var_2293_equation_0, values = (var_1867_cast, var_2179_cast))[name = tensor<string, []>("op_2293_cast")];
            tensor<string, []> var_2295_equation_0 = const()[name = tensor<string, []>("op_2295_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2295_cast = einsum(equation = var_2295_equation_0, values = (var_1867_cast, var_2180_cast))[name = tensor<string, []>("op_2295_cast")];
            tensor<string, []> var_2297_equation_0 = const()[name = tensor<string, []>("op_2297_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2297_cast = einsum(equation = var_2297_equation_0, values = (var_1867_cast, var_2181_cast))[name = tensor<string, []>("op_2297_cast")];
            tensor<string, []> var_2299_equation_0 = const()[name = tensor<string, []>("op_2299_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2299_cast = einsum(equation = var_2299_equation_0, values = (var_1867_cast, var_2182_cast))[name = tensor<string, []>("op_2299_cast")];
            tensor<string, []> var_2301_equation_0 = const()[name = tensor<string, []>("op_2301_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2301_cast = einsum(equation = var_2301_equation_0, values = (var_1867_cast, var_2183_cast))[name = tensor<string, []>("op_2301_cast")];
            tensor<string, []> var_2303_equation_0 = const()[name = tensor<string, []>("op_2303_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2303_cast = einsum(equation = var_2303_equation_0, values = (var_1867_cast, var_2184_cast))[name = tensor<string, []>("op_2303_cast")];
            tensor<string, []> var_2305_equation_0 = const()[name = tensor<string, []>("op_2305_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2305_cast = einsum(equation = var_2305_equation_0, values = (var_1867_cast, var_2185_cast))[name = tensor<string, []>("op_2305_cast")];
            tensor<string, []> var_2307_equation_0 = const()[name = tensor<string, []>("op_2307_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2307_cast = einsum(equation = var_2307_equation_0, values = (var_1871_cast, var_2186_cast))[name = tensor<string, []>("op_2307_cast")];
            tensor<string, []> var_2309_equation_0 = const()[name = tensor<string, []>("op_2309_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2309_cast = einsum(equation = var_2309_equation_0, values = (var_1871_cast, var_2187_cast))[name = tensor<string, []>("op_2309_cast")];
            tensor<string, []> var_2311_equation_0 = const()[name = tensor<string, []>("op_2311_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2311_cast = einsum(equation = var_2311_equation_0, values = (var_1871_cast, var_2188_cast))[name = tensor<string, []>("op_2311_cast")];
            tensor<string, []> var_2313_equation_0 = const()[name = tensor<string, []>("op_2313_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2313_cast = einsum(equation = var_2313_equation_0, values = (var_1871_cast, var_2189_cast))[name = tensor<string, []>("op_2313_cast")];
            tensor<string, []> var_2315_equation_0 = const()[name = tensor<string, []>("op_2315_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2315_cast = einsum(equation = var_2315_equation_0, values = (var_1871_cast, var_2190_cast))[name = tensor<string, []>("op_2315_cast")];
            tensor<string, []> var_2317_equation_0 = const()[name = tensor<string, []>("op_2317_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2317_cast = einsum(equation = var_2317_equation_0, values = (var_1871_cast, var_2191_cast))[name = tensor<string, []>("op_2317_cast")];
            tensor<string, []> var_2319_equation_0 = const()[name = tensor<string, []>("op_2319_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2319_cast = einsum(equation = var_2319_equation_0, values = (var_1871_cast, var_2192_cast))[name = tensor<string, []>("op_2319_cast")];
            tensor<string, []> var_2321_equation_0 = const()[name = tensor<string, []>("op_2321_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2321_cast = einsum(equation = var_2321_equation_0, values = (var_1871_cast, var_2193_cast))[name = tensor<string, []>("op_2321_cast")];
            tensor<bool, []> var_2323_interleave_0 = const()[name = tensor<string, []>("op_2323_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_2323_cast = concat(axis = var_95, interleave = var_2323_interleave_0, values = (var_2195_cast, var_2197_cast, var_2199_cast, var_2201_cast, var_2203_cast, var_2205_cast, var_2207_cast, var_2209_cast))[name = tensor<string, []>("op_2323_cast")];
            tensor<bool, []> var_2325_interleave_0 = const()[name = tensor<string, []>("op_2325_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_2325_cast = concat(axis = var_95, interleave = var_2325_interleave_0, values = (var_2211_cast, var_2213_cast, var_2215_cast, var_2217_cast, var_2219_cast, var_2221_cast, var_2223_cast, var_2225_cast))[name = tensor<string, []>("op_2325_cast")];
            tensor<bool, []> var_2327_interleave_0 = const()[name = tensor<string, []>("op_2327_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_2327_cast = concat(axis = var_95, interleave = var_2327_interleave_0, values = (var_2227_cast, var_2229_cast, var_2231_cast, var_2233_cast, var_2235_cast, var_2237_cast, var_2239_cast, var_2241_cast))[name = tensor<string, []>("op_2327_cast")];
            tensor<bool, []> var_2329_interleave_0 = const()[name = tensor<string, []>("op_2329_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_2329_cast = concat(axis = var_95, interleave = var_2329_interleave_0, values = (var_2243_cast, var_2245_cast, var_2247_cast, var_2249_cast, var_2251_cast, var_2253_cast, var_2255_cast, var_2257_cast))[name = tensor<string, []>("op_2329_cast")];
            tensor<bool, []> var_2331_interleave_0 = const()[name = tensor<string, []>("op_2331_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_2331_cast = concat(axis = var_95, interleave = var_2331_interleave_0, values = (var_2259_cast, var_2261_cast, var_2263_cast, var_2265_cast, var_2267_cast, var_2269_cast, var_2271_cast, var_2273_cast))[name = tensor<string, []>("op_2331_cast")];
            tensor<bool, []> var_2333_interleave_0 = const()[name = tensor<string, []>("op_2333_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_2333_cast = concat(axis = var_95, interleave = var_2333_interleave_0, values = (var_2275_cast, var_2277_cast, var_2279_cast, var_2281_cast, var_2283_cast, var_2285_cast, var_2287_cast, var_2289_cast))[name = tensor<string, []>("op_2333_cast")];
            tensor<bool, []> var_2335_interleave_0 = const()[name = tensor<string, []>("op_2335_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_2335_cast = concat(axis = var_95, interleave = var_2335_interleave_0, values = (var_2291_cast, var_2293_cast, var_2295_cast, var_2297_cast, var_2299_cast, var_2301_cast, var_2303_cast, var_2305_cast))[name = tensor<string, []>("op_2335_cast")];
            tensor<bool, []> var_2337_interleave_0 = const()[name = tensor<string, []>("op_2337_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_2337_cast = concat(axis = var_95, interleave = var_2337_interleave_0, values = (var_2307_cast, var_2309_cast, var_2311_cast, var_2313_cast, var_2315_cast, var_2317_cast, var_2319_cast, var_2321_cast))[name = tensor<string, []>("op_2337_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_123, interleave = input_51_interleave_0, values = (var_2323_cast, var_2325_cast, var_2327_cast, var_2329_cast, var_2331_cast, var_2333_cast, var_2335_cast, var_2337_cast))[name = tensor<string, []>("input_51_cast")];
            tensor<int32, [2]> var_2343 = const()[name = tensor<string, []>("op_2343"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2345 = const()[name = tensor<string, []>("op_2345"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_2347_pad_type_0 = const()[name = tensor<string, []>("op_2347_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_2347_pad_0 = const()[name = tensor<string, []>("op_2347_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7174592))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7251456))), name = tensor<string, []>("down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            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, []>(7251648)))];
            tensor<fp16, [2, 320, 1, 4096]> var_2347_cast = conv(bias = down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_2345, groups = var_123, pad = var_2347_pad_0, pad_type = var_2347_pad_type_0, strides = var_2343, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_51_cast)[name = tensor<string, []>("op_2347_cast")];
            tensor<fp16, [2, 320, 1, 4096]> inputs_9_cast = add(x = var_2347_cast, y = inputs_7_cast)[name = tensor<string, []>("inputs_9_cast")];
            tensor<int32, [1]> var_2351 = const()[name = tensor<string, []>("op_2351"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_9_cast = reduce_mean(axes = var_2351, keep_dims = var_118, x = inputs_9_cast)[name = tensor<string, []>("channels_mean_9_cast")];
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_9_cast = sub(x = inputs_9_cast, y = channels_mean_9_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_9_cast")];
            tensor<int32, [1]> var_2355 = const()[name = tensor<string, []>("op_2355"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_2356_cast = reduce_mean(axes = var_2355, keep_dims = var_118, x = zero_mean_sq_9_cast)[name = tensor<string, []>("op_2356_cast")];
            tensor<fp16, []> var_2357_to_fp16 = const()[name = tensor<string, []>("op_2357_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_2358_cast = add(x = var_2356_cast, y = var_2357_to_fp16)[name = tensor<string, []>("op_2358_cast")];
            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_2358_cast)[name = tensor<string, []>("denom_9_cast")];
            tensor<fp16, [2, 320, 1, 4096]> out_9_cast = mul(x = zero_mean_9_cast, y = denom_9_cast)[name = tensor<string, []>("out_9_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_2362_to_fp16 = const()[name = tensor<string, []>("op_2362_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7252352)))];
            tensor<fp16, [2, 320, 1, 4096]> var_2363_cast = add(x = out_9_cast, y = var_2362_to_fp16)[name = tensor<string, []>("op_2363_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_2365_to_fp16 = const()[name = tensor<string, []>("op_2365_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7253056)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_29_cast = mul(x = var_2363_cast, y = var_2365_to_fp16)[name = tensor<string, []>("hidden_states_29_cast")];
            tensor<int32, [2]> var_2372 = const()[name = tensor<string, []>("op_2372"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2374 = const()[name = tensor<string, []>("op_2374"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7253760))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7330624))), name = tensor<string, []>("down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            tensor<fp16, [2, 320, 1, 4096]> q_7_cast = conv(dilations = var_2374, groups = var_123, pad = q_7_pad_0, pad_type = q_7_pad_type_0, strides = var_2372, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_29_cast)[name = tensor<string, []>("q_7_cast")];
            tensor<int32, [2]> var_2378 = const()[name = tensor<string, []>("op_2378"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2380 = const()[name = tensor<string, []>("op_2380"), 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, 768, 1, 1]> down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [184320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7330816))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7515200))), name = tensor<string, []>("down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 768, 1, 1])];
            tensor<fp16, [2, 320, 1, 77]> k_13_cast = conv(dilations = var_2380, groups = var_123, pad = k_13_pad_0, pad_type = k_13_pad_type_0, strides = var_2378, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_13_cast")];
            tensor<int32, [2]> var_2384 = const()[name = tensor<string, []>("op_2384"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2386 = const()[name = tensor<string, []>("op_2386"), 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, 768, 1, 1]> down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [184320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7515392))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7699776))), name = tensor<string, []>("down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 768, 1, 1])];
            tensor<fp16, [2, 320, 1, 77]> v_7_cast = conv(dilations = var_2386, groups = var_123, pad = v_7_pad_0, pad_type = v_7_pad_type_0, strides = var_2384, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_7_cast")];
            tensor<int32, [4]> var_2390_begin_0 = const()[name = tensor<string, []>("op_2390_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2390_end_0 = const()[name = tensor<string, []>("op_2390_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            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, 40, 1, 4096]> 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_7_cast)[name = tensor<string, []>("op_2390_cast")];
            tensor<int32, [4]> var_2394_begin_0 = const()[name = tensor<string, []>("op_2394_begin_0"), val = tensor<int32, [4]>([0, 40, 0, 0])];
            tensor<int32, [4]> var_2394_end_0 = const()[name = tensor<string, []>("op_2394_end_0"), val = tensor<int32, [4]>([2, 80, 1, 4096])];
            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, 40, 1, 4096]> 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_7_cast)[name = tensor<string, []>("op_2394_cast")];
            tensor<int32, [4]> var_2398_begin_0 = const()[name = tensor<string, []>("op_2398_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_2398_end_0 = const()[name = tensor<string, []>("op_2398_end_0"), val = tensor<int32, [4]>([2, 120, 1, 4096])];
            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, 40, 1, 4096]> 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_7_cast)[name = tensor<string, []>("op_2398_cast")];
            tensor<int32, [4]> var_2402_begin_0 = const()[name = tensor<string, []>("op_2402_begin_0"), val = tensor<int32, [4]>([0, 120, 0, 0])];
            tensor<int32, [4]> var_2402_end_0 = const()[name = tensor<string, []>("op_2402_end_0"), val = tensor<int32, [4]>([2, 160, 1, 4096])];
            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, 40, 1, 4096]> 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_7_cast)[name = tensor<string, []>("op_2402_cast")];
            tensor<int32, [4]> var_2406_begin_0 = const()[name = tensor<string, []>("op_2406_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_2406_end_0 = const()[name = tensor<string, []>("op_2406_end_0"), val = tensor<int32, [4]>([2, 200, 1, 4096])];
            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, 40, 1, 4096]> 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_7_cast)[name = tensor<string, []>("op_2406_cast")];
            tensor<int32, [4]> var_2410_begin_0 = const()[name = tensor<string, []>("op_2410_begin_0"), val = tensor<int32, [4]>([0, 200, 0, 0])];
            tensor<int32, [4]> var_2410_end_0 = const()[name = tensor<string, []>("op_2410_end_0"), val = tensor<int32, [4]>([2, 240, 1, 4096])];
            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, 40, 1, 4096]> 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_7_cast)[name = tensor<string, []>("op_2410_cast")];
            tensor<int32, [4]> var_2414_begin_0 = const()[name = tensor<string, []>("op_2414_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_2414_end_0 = const()[name = tensor<string, []>("op_2414_end_0"), val = tensor<int32, [4]>([2, 280, 1, 4096])];
            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, 40, 1, 4096]> 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_7_cast)[name = tensor<string, []>("op_2414_cast")];
            tensor<int32, [4]> var_2418_begin_0 = const()[name = tensor<string, []>("op_2418_begin_0"), val = tensor<int32, [4]>([0, 280, 0, 0])];
            tensor<int32, [4]> var_2418_end_0 = const()[name = tensor<string, []>("op_2418_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            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, 40, 1, 4096]> 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_7_cast)[name = tensor<string, []>("op_2418_cast")];
            tensor<int32, [4]> var_2421_begin_0 = const()[name = tensor<string, []>("op_2421_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2421_end_0 = const()[name = tensor<string, []>("op_2421_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_2421_end_mask_0 = const()[name = tensor<string, []>("op_2421_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2421_cast = slice_by_index(begin = var_2421_begin_0, end = var_2421_end_0, end_mask = var_2421_end_mask_0, x = var_2390_cast)[name = tensor<string, []>("op_2421_cast")];
            tensor<int32, [4]> var_2422_begin_0 = const()[name = tensor<string, []>("op_2422_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2422_end_0 = const()[name = tensor<string, []>("op_2422_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_2422_end_mask_0 = const()[name = tensor<string, []>("op_2422_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2422_cast = slice_by_index(begin = var_2422_begin_0, end = var_2422_end_0, end_mask = var_2422_end_mask_0, x = var_2390_cast)[name = tensor<string, []>("op_2422_cast")];
            tensor<int32, [4]> var_2423_begin_0 = const()[name = tensor<string, []>("op_2423_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_2423_end_0 = const()[name = tensor<string, []>("op_2423_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_2423_end_mask_0 = const()[name = tensor<string, []>("op_2423_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2423_cast = slice_by_index(begin = var_2423_begin_0, end = var_2423_end_0, end_mask = var_2423_end_mask_0, x = var_2390_cast)[name = tensor<string, []>("op_2423_cast")];
            tensor<int32, [4]> var_2424_begin_0 = const()[name = tensor<string, []>("op_2424_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_2424_end_0 = const()[name = tensor<string, []>("op_2424_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_2424_end_mask_0 = const()[name = tensor<string, []>("op_2424_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2424_cast = slice_by_index(begin = var_2424_begin_0, end = var_2424_end_0, end_mask = var_2424_end_mask_0, x = var_2390_cast)[name = tensor<string, []>("op_2424_cast")];
            tensor<int32, [4]> var_2425_begin_0 = const()[name = tensor<string, []>("op_2425_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_2425_end_0 = const()[name = tensor<string, []>("op_2425_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_2425_end_mask_0 = const()[name = tensor<string, []>("op_2425_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2425_cast = slice_by_index(begin = var_2425_begin_0, end = var_2425_end_0, end_mask = var_2425_end_mask_0, x = var_2390_cast)[name = tensor<string, []>("op_2425_cast")];
            tensor<int32, [4]> var_2426_begin_0 = const()[name = tensor<string, []>("op_2426_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_2426_end_0 = const()[name = tensor<string, []>("op_2426_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_2426_end_mask_0 = const()[name = tensor<string, []>("op_2426_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2426_cast = slice_by_index(begin = var_2426_begin_0, end = var_2426_end_0, end_mask = var_2426_end_mask_0, x = var_2390_cast)[name = tensor<string, []>("op_2426_cast")];
            tensor<int32, [4]> var_2427_begin_0 = const()[name = tensor<string, []>("op_2427_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_2427_end_0 = const()[name = tensor<string, []>("op_2427_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_2427_end_mask_0 = const()[name = tensor<string, []>("op_2427_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2427_cast = slice_by_index(begin = var_2427_begin_0, end = var_2427_end_0, end_mask = var_2427_end_mask_0, x = var_2390_cast)[name = tensor<string, []>("op_2427_cast")];
            tensor<int32, [4]> var_2428_begin_0 = const()[name = tensor<string, []>("op_2428_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_2428_end_0 = const()[name = tensor<string, []>("op_2428_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_2428_end_mask_0 = const()[name = tensor<string, []>("op_2428_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2428_cast = slice_by_index(begin = var_2428_begin_0, end = var_2428_end_0, end_mask = var_2428_end_mask_0, x = var_2390_cast)[name = tensor<string, []>("op_2428_cast")];
            tensor<int32, [4]> var_2429_begin_0 = const()[name = tensor<string, []>("op_2429_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2429_end_0 = const()[name = tensor<string, []>("op_2429_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_2429_end_mask_0 = const()[name = tensor<string, []>("op_2429_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2429_cast = slice_by_index(begin = var_2429_begin_0, end = var_2429_end_0, end_mask = var_2429_end_mask_0, x = var_2394_cast)[name = tensor<string, []>("op_2429_cast")];
            tensor<int32, [4]> var_2430_begin_0 = const()[name = tensor<string, []>("op_2430_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2430_end_0 = const()[name = tensor<string, []>("op_2430_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_2430_end_mask_0 = const()[name = tensor<string, []>("op_2430_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2430_cast = slice_by_index(begin = var_2430_begin_0, end = var_2430_end_0, end_mask = var_2430_end_mask_0, x = var_2394_cast)[name = tensor<string, []>("op_2430_cast")];
            tensor<int32, [4]> var_2431_begin_0 = const()[name = tensor<string, []>("op_2431_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_2431_end_0 = const()[name = tensor<string, []>("op_2431_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_2431_end_mask_0 = const()[name = tensor<string, []>("op_2431_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2431_cast = slice_by_index(begin = var_2431_begin_0, end = var_2431_end_0, end_mask = var_2431_end_mask_0, x = var_2394_cast)[name = tensor<string, []>("op_2431_cast")];
            tensor<int32, [4]> var_2432_begin_0 = const()[name = tensor<string, []>("op_2432_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_2432_end_0 = const()[name = tensor<string, []>("op_2432_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_2432_end_mask_0 = const()[name = tensor<string, []>("op_2432_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2432_cast = slice_by_index(begin = var_2432_begin_0, end = var_2432_end_0, end_mask = var_2432_end_mask_0, x = var_2394_cast)[name = tensor<string, []>("op_2432_cast")];
            tensor<int32, [4]> var_2433_begin_0 = const()[name = tensor<string, []>("op_2433_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_2433_end_0 = const()[name = tensor<string, []>("op_2433_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_2433_end_mask_0 = const()[name = tensor<string, []>("op_2433_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2433_cast = slice_by_index(begin = var_2433_begin_0, end = var_2433_end_0, end_mask = var_2433_end_mask_0, x = var_2394_cast)[name = tensor<string, []>("op_2433_cast")];
            tensor<int32, [4]> var_2434_begin_0 = const()[name = tensor<string, []>("op_2434_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_2434_end_0 = const()[name = tensor<string, []>("op_2434_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_2434_end_mask_0 = const()[name = tensor<string, []>("op_2434_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2434_cast = slice_by_index(begin = var_2434_begin_0, end = var_2434_end_0, end_mask = var_2434_end_mask_0, x = var_2394_cast)[name = tensor<string, []>("op_2434_cast")];
            tensor<int32, [4]> var_2435_begin_0 = const()[name = tensor<string, []>("op_2435_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_2435_end_0 = const()[name = tensor<string, []>("op_2435_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_2435_end_mask_0 = const()[name = tensor<string, []>("op_2435_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2435_cast = slice_by_index(begin = var_2435_begin_0, end = var_2435_end_0, end_mask = var_2435_end_mask_0, x = var_2394_cast)[name = tensor<string, []>("op_2435_cast")];
            tensor<int32, [4]> var_2436_begin_0 = const()[name = tensor<string, []>("op_2436_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_2436_end_0 = const()[name = tensor<string, []>("op_2436_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_2436_end_mask_0 = const()[name = tensor<string, []>("op_2436_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2436_cast = slice_by_index(begin = var_2436_begin_0, end = var_2436_end_0, end_mask = var_2436_end_mask_0, x = var_2394_cast)[name = tensor<string, []>("op_2436_cast")];
            tensor<int32, [4]> var_2437_begin_0 = const()[name = tensor<string, []>("op_2437_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2437_end_0 = const()[name = tensor<string, []>("op_2437_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_2437_end_mask_0 = const()[name = tensor<string, []>("op_2437_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2437_cast = slice_by_index(begin = var_2437_begin_0, end = var_2437_end_0, end_mask = var_2437_end_mask_0, x = var_2398_cast)[name = tensor<string, []>("op_2437_cast")];
            tensor<int32, [4]> var_2438_begin_0 = const()[name = tensor<string, []>("op_2438_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2438_end_0 = const()[name = tensor<string, []>("op_2438_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_2438_end_mask_0 = const()[name = tensor<string, []>("op_2438_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2438_cast = slice_by_index(begin = var_2438_begin_0, end = var_2438_end_0, end_mask = var_2438_end_mask_0, x = var_2398_cast)[name = tensor<string, []>("op_2438_cast")];
            tensor<int32, [4]> var_2439_begin_0 = const()[name = tensor<string, []>("op_2439_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_2439_end_0 = const()[name = tensor<string, []>("op_2439_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_2439_end_mask_0 = const()[name = tensor<string, []>("op_2439_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2439_cast = slice_by_index(begin = var_2439_begin_0, end = var_2439_end_0, end_mask = var_2439_end_mask_0, x = var_2398_cast)[name = tensor<string, []>("op_2439_cast")];
            tensor<int32, [4]> var_2440_begin_0 = const()[name = tensor<string, []>("op_2440_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_2440_end_0 = const()[name = tensor<string, []>("op_2440_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_2440_end_mask_0 = const()[name = tensor<string, []>("op_2440_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2440_cast = slice_by_index(begin = var_2440_begin_0, end = var_2440_end_0, end_mask = var_2440_end_mask_0, x = var_2398_cast)[name = tensor<string, []>("op_2440_cast")];
            tensor<int32, [4]> var_2441_begin_0 = const()[name = tensor<string, []>("op_2441_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_2441_end_0 = const()[name = tensor<string, []>("op_2441_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_2441_end_mask_0 = const()[name = tensor<string, []>("op_2441_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2441_cast = slice_by_index(begin = var_2441_begin_0, end = var_2441_end_0, end_mask = var_2441_end_mask_0, x = var_2398_cast)[name = tensor<string, []>("op_2441_cast")];
            tensor<int32, [4]> var_2442_begin_0 = const()[name = tensor<string, []>("op_2442_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_2442_end_0 = const()[name = tensor<string, []>("op_2442_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_2442_end_mask_0 = const()[name = tensor<string, []>("op_2442_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2442_cast = slice_by_index(begin = var_2442_begin_0, end = var_2442_end_0, end_mask = var_2442_end_mask_0, x = var_2398_cast)[name = tensor<string, []>("op_2442_cast")];
            tensor<int32, [4]> var_2443_begin_0 = const()[name = tensor<string, []>("op_2443_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_2443_end_0 = const()[name = tensor<string, []>("op_2443_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_2443_end_mask_0 = const()[name = tensor<string, []>("op_2443_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2443_cast = slice_by_index(begin = var_2443_begin_0, end = var_2443_end_0, end_mask = var_2443_end_mask_0, x = var_2398_cast)[name = tensor<string, []>("op_2443_cast")];
            tensor<int32, [4]> var_2444_begin_0 = const()[name = tensor<string, []>("op_2444_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_2444_end_0 = const()[name = tensor<string, []>("op_2444_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_2444_end_mask_0 = const()[name = tensor<string, []>("op_2444_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2444_cast = slice_by_index(begin = var_2444_begin_0, end = var_2444_end_0, end_mask = var_2444_end_mask_0, x = var_2398_cast)[name = tensor<string, []>("op_2444_cast")];
            tensor<int32, [4]> var_2445_begin_0 = const()[name = tensor<string, []>("op_2445_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2445_end_0 = const()[name = tensor<string, []>("op_2445_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_2445_end_mask_0 = const()[name = tensor<string, []>("op_2445_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2445_cast = slice_by_index(begin = var_2445_begin_0, end = var_2445_end_0, end_mask = var_2445_end_mask_0, x = var_2402_cast)[name = tensor<string, []>("op_2445_cast")];
            tensor<int32, [4]> var_2446_begin_0 = const()[name = tensor<string, []>("op_2446_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2446_end_0 = const()[name = tensor<string, []>("op_2446_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_2446_end_mask_0 = const()[name = tensor<string, []>("op_2446_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2446_cast = slice_by_index(begin = var_2446_begin_0, end = var_2446_end_0, end_mask = var_2446_end_mask_0, x = var_2402_cast)[name = tensor<string, []>("op_2446_cast")];
            tensor<int32, [4]> var_2447_begin_0 = const()[name = tensor<string, []>("op_2447_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_2447_end_0 = const()[name = tensor<string, []>("op_2447_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_2447_end_mask_0 = const()[name = tensor<string, []>("op_2447_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2447_cast = slice_by_index(begin = var_2447_begin_0, end = var_2447_end_0, end_mask = var_2447_end_mask_0, x = var_2402_cast)[name = tensor<string, []>("op_2447_cast")];
            tensor<int32, [4]> var_2448_begin_0 = const()[name = tensor<string, []>("op_2448_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_2448_end_0 = const()[name = tensor<string, []>("op_2448_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_2448_end_mask_0 = const()[name = tensor<string, []>("op_2448_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2448_cast = slice_by_index(begin = var_2448_begin_0, end = var_2448_end_0, end_mask = var_2448_end_mask_0, x = var_2402_cast)[name = tensor<string, []>("op_2448_cast")];
            tensor<int32, [4]> var_2449_begin_0 = const()[name = tensor<string, []>("op_2449_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_2449_end_0 = const()[name = tensor<string, []>("op_2449_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_2449_end_mask_0 = const()[name = tensor<string, []>("op_2449_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2449_cast = slice_by_index(begin = var_2449_begin_0, end = var_2449_end_0, end_mask = var_2449_end_mask_0, x = var_2402_cast)[name = tensor<string, []>("op_2449_cast")];
            tensor<int32, [4]> var_2450_begin_0 = const()[name = tensor<string, []>("op_2450_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_2450_end_0 = const()[name = tensor<string, []>("op_2450_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_2450_end_mask_0 = const()[name = tensor<string, []>("op_2450_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2450_cast = slice_by_index(begin = var_2450_begin_0, end = var_2450_end_0, end_mask = var_2450_end_mask_0, x = var_2402_cast)[name = tensor<string, []>("op_2450_cast")];
            tensor<int32, [4]> var_2451_begin_0 = const()[name = tensor<string, []>("op_2451_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_2451_end_0 = const()[name = tensor<string, []>("op_2451_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_2451_end_mask_0 = const()[name = tensor<string, []>("op_2451_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2451_cast = slice_by_index(begin = var_2451_begin_0, end = var_2451_end_0, end_mask = var_2451_end_mask_0, x = var_2402_cast)[name = tensor<string, []>("op_2451_cast")];
            tensor<int32, [4]> var_2452_begin_0 = const()[name = tensor<string, []>("op_2452_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_2452_end_0 = const()[name = tensor<string, []>("op_2452_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_2452_end_mask_0 = const()[name = tensor<string, []>("op_2452_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2452_cast = slice_by_index(begin = var_2452_begin_0, end = var_2452_end_0, end_mask = var_2452_end_mask_0, x = var_2402_cast)[name = tensor<string, []>("op_2452_cast")];
            tensor<int32, [4]> var_2453_begin_0 = const()[name = tensor<string, []>("op_2453_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2453_end_0 = const()[name = tensor<string, []>("op_2453_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_2453_end_mask_0 = const()[name = tensor<string, []>("op_2453_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2453_cast = slice_by_index(begin = var_2453_begin_0, end = var_2453_end_0, end_mask = var_2453_end_mask_0, x = var_2406_cast)[name = tensor<string, []>("op_2453_cast")];
            tensor<int32, [4]> var_2454_begin_0 = const()[name = tensor<string, []>("op_2454_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2454_end_0 = const()[name = tensor<string, []>("op_2454_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_2454_end_mask_0 = const()[name = tensor<string, []>("op_2454_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2454_cast = slice_by_index(begin = var_2454_begin_0, end = var_2454_end_0, end_mask = var_2454_end_mask_0, x = var_2406_cast)[name = tensor<string, []>("op_2454_cast")];
            tensor<int32, [4]> var_2455_begin_0 = const()[name = tensor<string, []>("op_2455_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_2455_end_0 = const()[name = tensor<string, []>("op_2455_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_2455_end_mask_0 = const()[name = tensor<string, []>("op_2455_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2455_cast = slice_by_index(begin = var_2455_begin_0, end = var_2455_end_0, end_mask = var_2455_end_mask_0, x = var_2406_cast)[name = tensor<string, []>("op_2455_cast")];
            tensor<int32, [4]> var_2456_begin_0 = const()[name = tensor<string, []>("op_2456_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_2456_end_0 = const()[name = tensor<string, []>("op_2456_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_2456_end_mask_0 = const()[name = tensor<string, []>("op_2456_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2456_cast = slice_by_index(begin = var_2456_begin_0, end = var_2456_end_0, end_mask = var_2456_end_mask_0, x = var_2406_cast)[name = tensor<string, []>("op_2456_cast")];
            tensor<int32, [4]> var_2457_begin_0 = const()[name = tensor<string, []>("op_2457_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_2457_end_0 = const()[name = tensor<string, []>("op_2457_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_2457_end_mask_0 = const()[name = tensor<string, []>("op_2457_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2457_cast = slice_by_index(begin = var_2457_begin_0, end = var_2457_end_0, end_mask = var_2457_end_mask_0, x = var_2406_cast)[name = tensor<string, []>("op_2457_cast")];
            tensor<int32, [4]> var_2458_begin_0 = const()[name = tensor<string, []>("op_2458_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_2458_end_0 = const()[name = tensor<string, []>("op_2458_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_2458_end_mask_0 = const()[name = tensor<string, []>("op_2458_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2458_cast = slice_by_index(begin = var_2458_begin_0, end = var_2458_end_0, end_mask = var_2458_end_mask_0, x = var_2406_cast)[name = tensor<string, []>("op_2458_cast")];
            tensor<int32, [4]> var_2459_begin_0 = const()[name = tensor<string, []>("op_2459_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_2459_end_0 = const()[name = tensor<string, []>("op_2459_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_2459_end_mask_0 = const()[name = tensor<string, []>("op_2459_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2459_cast = slice_by_index(begin = var_2459_begin_0, end = var_2459_end_0, end_mask = var_2459_end_mask_0, x = var_2406_cast)[name = tensor<string, []>("op_2459_cast")];
            tensor<int32, [4]> var_2460_begin_0 = const()[name = tensor<string, []>("op_2460_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_2460_end_0 = const()[name = tensor<string, []>("op_2460_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_2460_end_mask_0 = const()[name = tensor<string, []>("op_2460_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2460_cast = slice_by_index(begin = var_2460_begin_0, end = var_2460_end_0, end_mask = var_2460_end_mask_0, x = var_2406_cast)[name = tensor<string, []>("op_2460_cast")];
            tensor<int32, [4]> var_2461_begin_0 = const()[name = tensor<string, []>("op_2461_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2461_end_0 = const()[name = tensor<string, []>("op_2461_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_2461_end_mask_0 = const()[name = tensor<string, []>("op_2461_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2461_cast = slice_by_index(begin = var_2461_begin_0, end = var_2461_end_0, end_mask = var_2461_end_mask_0, x = var_2410_cast)[name = tensor<string, []>("op_2461_cast")];
            tensor<int32, [4]> var_2462_begin_0 = const()[name = tensor<string, []>("op_2462_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2462_end_0 = const()[name = tensor<string, []>("op_2462_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_2462_end_mask_0 = const()[name = tensor<string, []>("op_2462_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2462_cast = slice_by_index(begin = var_2462_begin_0, end = var_2462_end_0, end_mask = var_2462_end_mask_0, x = var_2410_cast)[name = tensor<string, []>("op_2462_cast")];
            tensor<int32, [4]> var_2463_begin_0 = const()[name = tensor<string, []>("op_2463_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_2463_end_0 = const()[name = tensor<string, []>("op_2463_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_2463_end_mask_0 = const()[name = tensor<string, []>("op_2463_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2463_cast = slice_by_index(begin = var_2463_begin_0, end = var_2463_end_0, end_mask = var_2463_end_mask_0, x = var_2410_cast)[name = tensor<string, []>("op_2463_cast")];
            tensor<int32, [4]> var_2464_begin_0 = const()[name = tensor<string, []>("op_2464_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_2464_end_0 = const()[name = tensor<string, []>("op_2464_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_2464_end_mask_0 = const()[name = tensor<string, []>("op_2464_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2464_cast = slice_by_index(begin = var_2464_begin_0, end = var_2464_end_0, end_mask = var_2464_end_mask_0, x = var_2410_cast)[name = tensor<string, []>("op_2464_cast")];
            tensor<int32, [4]> var_2465_begin_0 = const()[name = tensor<string, []>("op_2465_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_2465_end_0 = const()[name = tensor<string, []>("op_2465_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            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, 40, 1, 512]> var_2465_cast = slice_by_index(begin = var_2465_begin_0, end = var_2465_end_0, end_mask = var_2465_end_mask_0, x = var_2410_cast)[name = tensor<string, []>("op_2465_cast")];
            tensor<int32, [4]> var_2466_begin_0 = const()[name = tensor<string, []>("op_2466_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_2466_end_0 = const()[name = tensor<string, []>("op_2466_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_2466_end_mask_0 = const()[name = tensor<string, []>("op_2466_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2466_cast = slice_by_index(begin = var_2466_begin_0, end = var_2466_end_0, end_mask = var_2466_end_mask_0, x = var_2410_cast)[name = tensor<string, []>("op_2466_cast")];
            tensor<int32, [4]> var_2467_begin_0 = const()[name = tensor<string, []>("op_2467_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_2467_end_0 = const()[name = tensor<string, []>("op_2467_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_2467_end_mask_0 = const()[name = tensor<string, []>("op_2467_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2467_cast = slice_by_index(begin = var_2467_begin_0, end = var_2467_end_0, end_mask = var_2467_end_mask_0, x = var_2410_cast)[name = tensor<string, []>("op_2467_cast")];
            tensor<int32, [4]> var_2468_begin_0 = const()[name = tensor<string, []>("op_2468_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_2468_end_0 = const()[name = tensor<string, []>("op_2468_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_2468_end_mask_0 = const()[name = tensor<string, []>("op_2468_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2468_cast = slice_by_index(begin = var_2468_begin_0, end = var_2468_end_0, end_mask = var_2468_end_mask_0, x = var_2410_cast)[name = tensor<string, []>("op_2468_cast")];
            tensor<int32, [4]> var_2469_begin_0 = const()[name = tensor<string, []>("op_2469_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2469_end_0 = const()[name = tensor<string, []>("op_2469_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            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, 40, 1, 512]> var_2469_cast = slice_by_index(begin = var_2469_begin_0, end = var_2469_end_0, end_mask = var_2469_end_mask_0, x = var_2414_cast)[name = tensor<string, []>("op_2469_cast")];
            tensor<int32, [4]> var_2470_begin_0 = const()[name = tensor<string, []>("op_2470_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2470_end_0 = const()[name = tensor<string, []>("op_2470_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_2470_end_mask_0 = const()[name = tensor<string, []>("op_2470_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2470_cast = slice_by_index(begin = var_2470_begin_0, end = var_2470_end_0, end_mask = var_2470_end_mask_0, x = var_2414_cast)[name = tensor<string, []>("op_2470_cast")];
            tensor<int32, [4]> var_2471_begin_0 = const()[name = tensor<string, []>("op_2471_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_2471_end_0 = const()[name = tensor<string, []>("op_2471_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_2471_end_mask_0 = const()[name = tensor<string, []>("op_2471_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2471_cast = slice_by_index(begin = var_2471_begin_0, end = var_2471_end_0, end_mask = var_2471_end_mask_0, x = var_2414_cast)[name = tensor<string, []>("op_2471_cast")];
            tensor<int32, [4]> var_2472_begin_0 = const()[name = tensor<string, []>("op_2472_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_2472_end_0 = const()[name = tensor<string, []>("op_2472_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_2472_end_mask_0 = const()[name = tensor<string, []>("op_2472_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2472_cast = slice_by_index(begin = var_2472_begin_0, end = var_2472_end_0, end_mask = var_2472_end_mask_0, x = var_2414_cast)[name = tensor<string, []>("op_2472_cast")];
            tensor<int32, [4]> var_2473_begin_0 = const()[name = tensor<string, []>("op_2473_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_2473_end_0 = const()[name = tensor<string, []>("op_2473_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            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, 40, 1, 512]> var_2473_cast = slice_by_index(begin = var_2473_begin_0, end = var_2473_end_0, end_mask = var_2473_end_mask_0, x = var_2414_cast)[name = tensor<string, []>("op_2473_cast")];
            tensor<int32, [4]> var_2474_begin_0 = const()[name = tensor<string, []>("op_2474_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_2474_end_0 = const()[name = tensor<string, []>("op_2474_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_2474_end_mask_0 = const()[name = tensor<string, []>("op_2474_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2474_cast = slice_by_index(begin = var_2474_begin_0, end = var_2474_end_0, end_mask = var_2474_end_mask_0, x = var_2414_cast)[name = tensor<string, []>("op_2474_cast")];
            tensor<int32, [4]> var_2475_begin_0 = const()[name = tensor<string, []>("op_2475_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_2475_end_0 = const()[name = tensor<string, []>("op_2475_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_2475_end_mask_0 = const()[name = tensor<string, []>("op_2475_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2475_cast = slice_by_index(begin = var_2475_begin_0, end = var_2475_end_0, end_mask = var_2475_end_mask_0, x = var_2414_cast)[name = tensor<string, []>("op_2475_cast")];
            tensor<int32, [4]> var_2476_begin_0 = const()[name = tensor<string, []>("op_2476_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_2476_end_0 = const()[name = tensor<string, []>("op_2476_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_2476_end_mask_0 = const()[name = tensor<string, []>("op_2476_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2476_cast = slice_by_index(begin = var_2476_begin_0, end = var_2476_end_0, end_mask = var_2476_end_mask_0, x = var_2414_cast)[name = tensor<string, []>("op_2476_cast")];
            tensor<int32, [4]> var_2477_begin_0 = const()[name = tensor<string, []>("op_2477_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2477_end_0 = const()[name = tensor<string, []>("op_2477_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            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, 40, 1, 512]> var_2477_cast = slice_by_index(begin = var_2477_begin_0, end = var_2477_end_0, end_mask = var_2477_end_mask_0, x = var_2418_cast)[name = tensor<string, []>("op_2477_cast")];
            tensor<int32, [4]> var_2478_begin_0 = const()[name = tensor<string, []>("op_2478_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2478_end_0 = const()[name = tensor<string, []>("op_2478_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_2478_end_mask_0 = const()[name = tensor<string, []>("op_2478_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2478_cast = slice_by_index(begin = var_2478_begin_0, end = var_2478_end_0, end_mask = var_2478_end_mask_0, x = var_2418_cast)[name = tensor<string, []>("op_2478_cast")];
            tensor<int32, [4]> var_2479_begin_0 = const()[name = tensor<string, []>("op_2479_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_2479_end_0 = const()[name = tensor<string, []>("op_2479_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_2479_end_mask_0 = const()[name = tensor<string, []>("op_2479_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2479_cast = slice_by_index(begin = var_2479_begin_0, end = var_2479_end_0, end_mask = var_2479_end_mask_0, x = var_2418_cast)[name = tensor<string, []>("op_2479_cast")];
            tensor<int32, [4]> var_2480_begin_0 = const()[name = tensor<string, []>("op_2480_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_2480_end_0 = const()[name = tensor<string, []>("op_2480_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_2480_end_mask_0 = const()[name = tensor<string, []>("op_2480_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2480_cast = slice_by_index(begin = var_2480_begin_0, end = var_2480_end_0, end_mask = var_2480_end_mask_0, x = var_2418_cast)[name = tensor<string, []>("op_2480_cast")];
            tensor<int32, [4]> var_2481_begin_0 = const()[name = tensor<string, []>("op_2481_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_2481_end_0 = const()[name = tensor<string, []>("op_2481_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            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, 40, 1, 512]> var_2481_cast = slice_by_index(begin = var_2481_begin_0, end = var_2481_end_0, end_mask = var_2481_end_mask_0, x = var_2418_cast)[name = tensor<string, []>("op_2481_cast")];
            tensor<int32, [4]> var_2482_begin_0 = const()[name = tensor<string, []>("op_2482_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_2482_end_0 = const()[name = tensor<string, []>("op_2482_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_2482_end_mask_0 = const()[name = tensor<string, []>("op_2482_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2482_cast = slice_by_index(begin = var_2482_begin_0, end = var_2482_end_0, end_mask = var_2482_end_mask_0, x = var_2418_cast)[name = tensor<string, []>("op_2482_cast")];
            tensor<int32, [4]> var_2483_begin_0 = const()[name = tensor<string, []>("op_2483_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_2483_end_0 = const()[name = tensor<string, []>("op_2483_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_2483_end_mask_0 = const()[name = tensor<string, []>("op_2483_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2483_cast = slice_by_index(begin = var_2483_begin_0, end = var_2483_end_0, end_mask = var_2483_end_mask_0, x = var_2418_cast)[name = tensor<string, []>("op_2483_cast")];
            tensor<int32, [4]> var_2484_begin_0 = const()[name = tensor<string, []>("op_2484_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_2484_end_0 = const()[name = tensor<string, []>("op_2484_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_2484_end_mask_0 = const()[name = tensor<string, []>("op_2484_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_2484_cast = slice_by_index(begin = var_2484_begin_0, end = var_2484_end_0, end_mask = var_2484_end_mask_0, x = var_2418_cast)[name = tensor<string, []>("op_2484_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_2489_begin_0 = const()[name = tensor<string, []>("op_2489_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2489_end_0 = const()[name = tensor<string, []>("op_2489_end_0"), val = tensor<int32, [4]>([2, 77, 1, 40])];
            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, 77, 1, 320]> transpose_28 = transpose(perm = k_15_perm_0, x = k_13_cast)[name = tensor<string, []>("transpose_28")];
            tensor<fp16, [2, 77, 1, 40]> 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_28)[name = tensor<string, []>("op_2489_cast")];
            tensor<int32, [4]> var_2493_begin_0 = const()[name = tensor<string, []>("op_2493_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 40])];
            tensor<int32, [4]> var_2493_end_0 = const()[name = tensor<string, []>("op_2493_end_0"), val = tensor<int32, [4]>([2, 77, 1, 80])];
            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, 77, 1, 40]> 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_28)[name = tensor<string, []>("op_2493_cast")];
            tensor<int32, [4]> var_2497_begin_0 = const()[name = tensor<string, []>("op_2497_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 80])];
            tensor<int32, [4]> var_2497_end_0 = const()[name = tensor<string, []>("op_2497_end_0"), val = tensor<int32, [4]>([2, 77, 1, 120])];
            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, 77, 1, 40]> 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_28)[name = tensor<string, []>("op_2497_cast")];
            tensor<int32, [4]> var_2501_begin_0 = const()[name = tensor<string, []>("op_2501_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 120])];
            tensor<int32, [4]> var_2501_end_0 = const()[name = tensor<string, []>("op_2501_end_0"), val = tensor<int32, [4]>([2, 77, 1, 160])];
            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, 77, 1, 40]> 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_28)[name = tensor<string, []>("op_2501_cast")];
            tensor<int32, [4]> var_2505_begin_0 = const()[name = tensor<string, []>("op_2505_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_2505_end_0 = const()[name = tensor<string, []>("op_2505_end_0"), val = tensor<int32, [4]>([2, 77, 1, 200])];
            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, 77, 1, 40]> 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_28)[name = tensor<string, []>("op_2505_cast")];
            tensor<int32, [4]> var_2509_begin_0 = const()[name = tensor<string, []>("op_2509_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 200])];
            tensor<int32, [4]> var_2509_end_0 = const()[name = tensor<string, []>("op_2509_end_0"), val = tensor<int32, [4]>([2, 77, 1, 240])];
            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, 77, 1, 40]> 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_28)[name = tensor<string, []>("op_2509_cast")];
            tensor<int32, [4]> var_2513_begin_0 = const()[name = tensor<string, []>("op_2513_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 240])];
            tensor<int32, [4]> var_2513_end_0 = const()[name = tensor<string, []>("op_2513_end_0"), val = tensor<int32, [4]>([2, 77, 1, 280])];
            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, 77, 1, 40]> 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_28)[name = tensor<string, []>("op_2513_cast")];
            tensor<int32, [4]> var_2517_begin_0 = const()[name = tensor<string, []>("op_2517_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 280])];
            tensor<int32, [4]> var_2517_end_0 = const()[name = tensor<string, []>("op_2517_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            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, 77, 1, 40]> 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_28)[name = tensor<string, []>("op_2517_cast")];
            tensor<int32, [4]> var_2519_begin_0 = const()[name = tensor<string, []>("op_2519_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2519_end_0 = const()[name = tensor<string, []>("op_2519_end_0"), val = tensor<int32, [4]>([2, 40, 1, 77])];
            tensor<bool, [4]> var_2519_end_mask_0 = const()[name = tensor<string, []>("op_2519_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_2519_cast = slice_by_index(begin = var_2519_begin_0, end = var_2519_end_0, end_mask = var_2519_end_mask_0, x = v_7_cast)[name = tensor<string, []>("op_2519_cast")];
            tensor<int32, [4]> var_2523_begin_0 = const()[name = tensor<string, []>("op_2523_begin_0"), val = tensor<int32, [4]>([0, 40, 0, 0])];
            tensor<int32, [4]> var_2523_end_0 = const()[name = tensor<string, []>("op_2523_end_0"), val = tensor<int32, [4]>([2, 80, 1, 77])];
            tensor<bool, [4]> var_2523_end_mask_0 = const()[name = tensor<string, []>("op_2523_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_2523_cast = slice_by_index(begin = var_2523_begin_0, end = var_2523_end_0, end_mask = var_2523_end_mask_0, x = v_7_cast)[name = tensor<string, []>("op_2523_cast")];
            tensor<int32, [4]> var_2527_begin_0 = const()[name = tensor<string, []>("op_2527_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_2527_end_0 = const()[name = tensor<string, []>("op_2527_end_0"), val = tensor<int32, [4]>([2, 120, 1, 77])];
            tensor<bool, [4]> var_2527_end_mask_0 = const()[name = tensor<string, []>("op_2527_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_2527_cast = slice_by_index(begin = var_2527_begin_0, end = var_2527_end_0, end_mask = var_2527_end_mask_0, x = v_7_cast)[name = tensor<string, []>("op_2527_cast")];
            tensor<int32, [4]> var_2531_begin_0 = const()[name = tensor<string, []>("op_2531_begin_0"), val = tensor<int32, [4]>([0, 120, 0, 0])];
            tensor<int32, [4]> var_2531_end_0 = const()[name = tensor<string, []>("op_2531_end_0"), val = tensor<int32, [4]>([2, 160, 1, 77])];
            tensor<bool, [4]> var_2531_end_mask_0 = const()[name = tensor<string, []>("op_2531_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_2531_cast = slice_by_index(begin = var_2531_begin_0, end = var_2531_end_0, end_mask = var_2531_end_mask_0, x = v_7_cast)[name = tensor<string, []>("op_2531_cast")];
            tensor<int32, [4]> var_2535_begin_0 = const()[name = tensor<string, []>("op_2535_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_2535_end_0 = const()[name = tensor<string, []>("op_2535_end_0"), val = tensor<int32, [4]>([2, 200, 1, 77])];
            tensor<bool, [4]> var_2535_end_mask_0 = const()[name = tensor<string, []>("op_2535_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_2535_cast = slice_by_index(begin = var_2535_begin_0, end = var_2535_end_0, end_mask = var_2535_end_mask_0, x = v_7_cast)[name = tensor<string, []>("op_2535_cast")];
            tensor<int32, [4]> var_2539_begin_0 = const()[name = tensor<string, []>("op_2539_begin_0"), val = tensor<int32, [4]>([0, 200, 0, 0])];
            tensor<int32, [4]> var_2539_end_0 = const()[name = tensor<string, []>("op_2539_end_0"), val = tensor<int32, [4]>([2, 240, 1, 77])];
            tensor<bool, [4]> var_2539_end_mask_0 = const()[name = tensor<string, []>("op_2539_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_2539_cast = slice_by_index(begin = var_2539_begin_0, end = var_2539_end_0, end_mask = var_2539_end_mask_0, x = v_7_cast)[name = tensor<string, []>("op_2539_cast")];
            tensor<int32, [4]> var_2543_begin_0 = const()[name = tensor<string, []>("op_2543_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_2543_end_0 = const()[name = tensor<string, []>("op_2543_end_0"), val = tensor<int32, [4]>([2, 280, 1, 77])];
            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, 40, 1, 77]> 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_7_cast)[name = tensor<string, []>("op_2543_cast")];
            tensor<int32, [4]> var_2547_begin_0 = const()[name = tensor<string, []>("op_2547_begin_0"), val = tensor<int32, [4]>([0, 280, 0, 0])];
            tensor<int32, [4]> var_2547_end_0 = const()[name = tensor<string, []>("op_2547_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            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, 40, 1, 77]> 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_7_cast)[name = tensor<string, []>("op_2547_cast")];
            tensor<string, []> var_2551_equation_0 = const()[name = tensor<string, []>("op_2551_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2551_cast = einsum(equation = var_2551_equation_0, values = (var_2489_cast, var_2421_cast))[name = tensor<string, []>("op_2551_cast")];
            tensor<fp16, []> var_2552_to_fp16 = const()[name = tensor<string, []>("op_2552_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_385_cast = mul(x = var_2551_cast, y = var_2552_to_fp16)[name = tensor<string, []>("aw_chunk_385_cast")];
            tensor<string, []> var_2555_equation_0 = const()[name = tensor<string, []>("op_2555_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2555_cast = einsum(equation = var_2555_equation_0, values = (var_2489_cast, var_2422_cast))[name = tensor<string, []>("op_2555_cast")];
            tensor<fp16, []> var_2556_to_fp16 = const()[name = tensor<string, []>("op_2556_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_387_cast = mul(x = var_2555_cast, y = var_2556_to_fp16)[name = tensor<string, []>("aw_chunk_387_cast")];
            tensor<string, []> var_2559_equation_0 = const()[name = tensor<string, []>("op_2559_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2559_cast = einsum(equation = var_2559_equation_0, values = (var_2489_cast, var_2423_cast))[name = tensor<string, []>("op_2559_cast")];
            tensor<fp16, []> var_2560_to_fp16 = const()[name = tensor<string, []>("op_2560_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_389_cast = mul(x = var_2559_cast, y = var_2560_to_fp16)[name = tensor<string, []>("aw_chunk_389_cast")];
            tensor<string, []> var_2563_equation_0 = const()[name = tensor<string, []>("op_2563_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2563_cast = einsum(equation = var_2563_equation_0, values = (var_2489_cast, var_2424_cast))[name = tensor<string, []>("op_2563_cast")];
            tensor<fp16, []> var_2564_to_fp16 = const()[name = tensor<string, []>("op_2564_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_391_cast = mul(x = var_2563_cast, y = var_2564_to_fp16)[name = tensor<string, []>("aw_chunk_391_cast")];
            tensor<string, []> var_2567_equation_0 = const()[name = tensor<string, []>("op_2567_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2567_cast = einsum(equation = var_2567_equation_0, values = (var_2489_cast, var_2425_cast))[name = tensor<string, []>("op_2567_cast")];
            tensor<fp16, []> var_2568_to_fp16 = const()[name = tensor<string, []>("op_2568_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_393_cast = mul(x = var_2567_cast, y = var_2568_to_fp16)[name = tensor<string, []>("aw_chunk_393_cast")];
            tensor<string, []> var_2571_equation_0 = const()[name = tensor<string, []>("op_2571_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2571_cast = einsum(equation = var_2571_equation_0, values = (var_2489_cast, var_2426_cast))[name = tensor<string, []>("op_2571_cast")];
            tensor<fp16, []> var_2572_to_fp16 = const()[name = tensor<string, []>("op_2572_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_395_cast = mul(x = var_2571_cast, y = var_2572_to_fp16)[name = tensor<string, []>("aw_chunk_395_cast")];
            tensor<string, []> var_2575_equation_0 = const()[name = tensor<string, []>("op_2575_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2575_cast = einsum(equation = var_2575_equation_0, values = (var_2489_cast, var_2427_cast))[name = tensor<string, []>("op_2575_cast")];
            tensor<fp16, []> var_2576_to_fp16 = const()[name = tensor<string, []>("op_2576_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_397_cast = mul(x = var_2575_cast, y = var_2576_to_fp16)[name = tensor<string, []>("aw_chunk_397_cast")];
            tensor<string, []> var_2579_equation_0 = const()[name = tensor<string, []>("op_2579_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2579_cast = einsum(equation = var_2579_equation_0, values = (var_2489_cast, var_2428_cast))[name = tensor<string, []>("op_2579_cast")];
            tensor<fp16, []> var_2580_to_fp16 = const()[name = tensor<string, []>("op_2580_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_399_cast = mul(x = var_2579_cast, y = var_2580_to_fp16)[name = tensor<string, []>("aw_chunk_399_cast")];
            tensor<string, []> var_2583_equation_0 = const()[name = tensor<string, []>("op_2583_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2583_cast = einsum(equation = var_2583_equation_0, values = (var_2493_cast, var_2429_cast))[name = tensor<string, []>("op_2583_cast")];
            tensor<fp16, []> var_2584_to_fp16 = const()[name = tensor<string, []>("op_2584_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_401_cast = mul(x = var_2583_cast, y = var_2584_to_fp16)[name = tensor<string, []>("aw_chunk_401_cast")];
            tensor<string, []> var_2587_equation_0 = const()[name = tensor<string, []>("op_2587_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2587_cast = einsum(equation = var_2587_equation_0, values = (var_2493_cast, var_2430_cast))[name = tensor<string, []>("op_2587_cast")];
            tensor<fp16, []> var_2588_to_fp16 = const()[name = tensor<string, []>("op_2588_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_403_cast = mul(x = var_2587_cast, y = var_2588_to_fp16)[name = tensor<string, []>("aw_chunk_403_cast")];
            tensor<string, []> var_2591_equation_0 = const()[name = tensor<string, []>("op_2591_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2591_cast = einsum(equation = var_2591_equation_0, values = (var_2493_cast, var_2431_cast))[name = tensor<string, []>("op_2591_cast")];
            tensor<fp16, []> var_2592_to_fp16 = const()[name = tensor<string, []>("op_2592_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_405_cast = mul(x = var_2591_cast, y = var_2592_to_fp16)[name = tensor<string, []>("aw_chunk_405_cast")];
            tensor<string, []> var_2595_equation_0 = const()[name = tensor<string, []>("op_2595_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2595_cast = einsum(equation = var_2595_equation_0, values = (var_2493_cast, var_2432_cast))[name = tensor<string, []>("op_2595_cast")];
            tensor<fp16, []> var_2596_to_fp16 = const()[name = tensor<string, []>("op_2596_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_407_cast = mul(x = var_2595_cast, y = var_2596_to_fp16)[name = tensor<string, []>("aw_chunk_407_cast")];
            tensor<string, []> var_2599_equation_0 = const()[name = tensor<string, []>("op_2599_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2599_cast = einsum(equation = var_2599_equation_0, values = (var_2493_cast, var_2433_cast))[name = tensor<string, []>("op_2599_cast")];
            tensor<fp16, []> var_2600_to_fp16 = const()[name = tensor<string, []>("op_2600_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_409_cast = mul(x = var_2599_cast, y = var_2600_to_fp16)[name = tensor<string, []>("aw_chunk_409_cast")];
            tensor<string, []> var_2603_equation_0 = const()[name = tensor<string, []>("op_2603_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2603_cast = einsum(equation = var_2603_equation_0, values = (var_2493_cast, var_2434_cast))[name = tensor<string, []>("op_2603_cast")];
            tensor<fp16, []> var_2604_to_fp16 = const()[name = tensor<string, []>("op_2604_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_411_cast = mul(x = var_2603_cast, y = var_2604_to_fp16)[name = tensor<string, []>("aw_chunk_411_cast")];
            tensor<string, []> var_2607_equation_0 = const()[name = tensor<string, []>("op_2607_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2607_cast = einsum(equation = var_2607_equation_0, values = (var_2493_cast, var_2435_cast))[name = tensor<string, []>("op_2607_cast")];
            tensor<fp16, []> var_2608_to_fp16 = const()[name = tensor<string, []>("op_2608_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_413_cast = mul(x = var_2607_cast, y = var_2608_to_fp16)[name = tensor<string, []>("aw_chunk_413_cast")];
            tensor<string, []> var_2611_equation_0 = const()[name = tensor<string, []>("op_2611_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2611_cast = einsum(equation = var_2611_equation_0, values = (var_2493_cast, var_2436_cast))[name = tensor<string, []>("op_2611_cast")];
            tensor<fp16, []> var_2612_to_fp16 = const()[name = tensor<string, []>("op_2612_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_415_cast = mul(x = var_2611_cast, y = var_2612_to_fp16)[name = tensor<string, []>("aw_chunk_415_cast")];
            tensor<string, []> var_2615_equation_0 = const()[name = tensor<string, []>("op_2615_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2615_cast = einsum(equation = var_2615_equation_0, values = (var_2497_cast, var_2437_cast))[name = tensor<string, []>("op_2615_cast")];
            tensor<fp16, []> var_2616_to_fp16 = const()[name = tensor<string, []>("op_2616_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_417_cast = mul(x = var_2615_cast, y = var_2616_to_fp16)[name = tensor<string, []>("aw_chunk_417_cast")];
            tensor<string, []> var_2619_equation_0 = const()[name = tensor<string, []>("op_2619_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2619_cast = einsum(equation = var_2619_equation_0, values = (var_2497_cast, var_2438_cast))[name = tensor<string, []>("op_2619_cast")];
            tensor<fp16, []> var_2620_to_fp16 = const()[name = tensor<string, []>("op_2620_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_419_cast = mul(x = var_2619_cast, y = var_2620_to_fp16)[name = tensor<string, []>("aw_chunk_419_cast")];
            tensor<string, []> var_2623_equation_0 = const()[name = tensor<string, []>("op_2623_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2623_cast = einsum(equation = var_2623_equation_0, values = (var_2497_cast, var_2439_cast))[name = tensor<string, []>("op_2623_cast")];
            tensor<fp16, []> var_2624_to_fp16 = const()[name = tensor<string, []>("op_2624_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_421_cast = mul(x = var_2623_cast, y = var_2624_to_fp16)[name = tensor<string, []>("aw_chunk_421_cast")];
            tensor<string, []> var_2627_equation_0 = const()[name = tensor<string, []>("op_2627_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2627_cast = einsum(equation = var_2627_equation_0, values = (var_2497_cast, var_2440_cast))[name = tensor<string, []>("op_2627_cast")];
            tensor<fp16, []> var_2628_to_fp16 = const()[name = tensor<string, []>("op_2628_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_423_cast = mul(x = var_2627_cast, y = var_2628_to_fp16)[name = tensor<string, []>("aw_chunk_423_cast")];
            tensor<string, []> var_2631_equation_0 = const()[name = tensor<string, []>("op_2631_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2631_cast = einsum(equation = var_2631_equation_0, values = (var_2497_cast, var_2441_cast))[name = tensor<string, []>("op_2631_cast")];
            tensor<fp16, []> var_2632_to_fp16 = const()[name = tensor<string, []>("op_2632_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_425_cast = mul(x = var_2631_cast, y = var_2632_to_fp16)[name = tensor<string, []>("aw_chunk_425_cast")];
            tensor<string, []> var_2635_equation_0 = const()[name = tensor<string, []>("op_2635_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2635_cast = einsum(equation = var_2635_equation_0, values = (var_2497_cast, var_2442_cast))[name = tensor<string, []>("op_2635_cast")];
            tensor<fp16, []> var_2636_to_fp16 = const()[name = tensor<string, []>("op_2636_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_427_cast = mul(x = var_2635_cast, y = var_2636_to_fp16)[name = tensor<string, []>("aw_chunk_427_cast")];
            tensor<string, []> var_2639_equation_0 = const()[name = tensor<string, []>("op_2639_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2639_cast = einsum(equation = var_2639_equation_0, values = (var_2497_cast, var_2443_cast))[name = tensor<string, []>("op_2639_cast")];
            tensor<fp16, []> var_2640_to_fp16 = const()[name = tensor<string, []>("op_2640_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_429_cast = mul(x = var_2639_cast, y = var_2640_to_fp16)[name = tensor<string, []>("aw_chunk_429_cast")];
            tensor<string, []> var_2643_equation_0 = const()[name = tensor<string, []>("op_2643_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2643_cast = einsum(equation = var_2643_equation_0, values = (var_2497_cast, var_2444_cast))[name = tensor<string, []>("op_2643_cast")];
            tensor<fp16, []> var_2644_to_fp16 = const()[name = tensor<string, []>("op_2644_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_431_cast = mul(x = var_2643_cast, y = var_2644_to_fp16)[name = tensor<string, []>("aw_chunk_431_cast")];
            tensor<string, []> var_2647_equation_0 = const()[name = tensor<string, []>("op_2647_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2647_cast = einsum(equation = var_2647_equation_0, values = (var_2501_cast, var_2445_cast))[name = tensor<string, []>("op_2647_cast")];
            tensor<fp16, []> var_2648_to_fp16 = const()[name = tensor<string, []>("op_2648_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_433_cast = mul(x = var_2647_cast, y = var_2648_to_fp16)[name = tensor<string, []>("aw_chunk_433_cast")];
            tensor<string, []> var_2651_equation_0 = const()[name = tensor<string, []>("op_2651_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2651_cast = einsum(equation = var_2651_equation_0, values = (var_2501_cast, var_2446_cast))[name = tensor<string, []>("op_2651_cast")];
            tensor<fp16, []> var_2652_to_fp16 = const()[name = tensor<string, []>("op_2652_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_435_cast = mul(x = var_2651_cast, y = var_2652_to_fp16)[name = tensor<string, []>("aw_chunk_435_cast")];
            tensor<string, []> var_2655_equation_0 = const()[name = tensor<string, []>("op_2655_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2655_cast = einsum(equation = var_2655_equation_0, values = (var_2501_cast, var_2447_cast))[name = tensor<string, []>("op_2655_cast")];
            tensor<fp16, []> var_2656_to_fp16 = const()[name = tensor<string, []>("op_2656_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_437_cast = mul(x = var_2655_cast, y = var_2656_to_fp16)[name = tensor<string, []>("aw_chunk_437_cast")];
            tensor<string, []> var_2659_equation_0 = const()[name = tensor<string, []>("op_2659_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2659_cast = einsum(equation = var_2659_equation_0, values = (var_2501_cast, var_2448_cast))[name = tensor<string, []>("op_2659_cast")];
            tensor<fp16, []> var_2660_to_fp16 = const()[name = tensor<string, []>("op_2660_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_439_cast = mul(x = var_2659_cast, y = var_2660_to_fp16)[name = tensor<string, []>("aw_chunk_439_cast")];
            tensor<string, []> var_2663_equation_0 = const()[name = tensor<string, []>("op_2663_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2663_cast = einsum(equation = var_2663_equation_0, values = (var_2501_cast, var_2449_cast))[name = tensor<string, []>("op_2663_cast")];
            tensor<fp16, []> var_2664_to_fp16 = const()[name = tensor<string, []>("op_2664_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_441_cast = mul(x = var_2663_cast, y = var_2664_to_fp16)[name = tensor<string, []>("aw_chunk_441_cast")];
            tensor<string, []> var_2667_equation_0 = const()[name = tensor<string, []>("op_2667_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2667_cast = einsum(equation = var_2667_equation_0, values = (var_2501_cast, var_2450_cast))[name = tensor<string, []>("op_2667_cast")];
            tensor<fp16, []> var_2668_to_fp16 = const()[name = tensor<string, []>("op_2668_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_443_cast = mul(x = var_2667_cast, y = var_2668_to_fp16)[name = tensor<string, []>("aw_chunk_443_cast")];
            tensor<string, []> var_2671_equation_0 = const()[name = tensor<string, []>("op_2671_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2671_cast = einsum(equation = var_2671_equation_0, values = (var_2501_cast, var_2451_cast))[name = tensor<string, []>("op_2671_cast")];
            tensor<fp16, []> var_2672_to_fp16 = const()[name = tensor<string, []>("op_2672_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_445_cast = mul(x = var_2671_cast, y = var_2672_to_fp16)[name = tensor<string, []>("aw_chunk_445_cast")];
            tensor<string, []> var_2675_equation_0 = const()[name = tensor<string, []>("op_2675_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2675_cast = einsum(equation = var_2675_equation_0, values = (var_2501_cast, var_2452_cast))[name = tensor<string, []>("op_2675_cast")];
            tensor<fp16, []> var_2676_to_fp16 = const()[name = tensor<string, []>("op_2676_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_447_cast = mul(x = var_2675_cast, y = var_2676_to_fp16)[name = tensor<string, []>("aw_chunk_447_cast")];
            tensor<string, []> var_2679_equation_0 = const()[name = tensor<string, []>("op_2679_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2679_cast = einsum(equation = var_2679_equation_0, values = (var_2505_cast, var_2453_cast))[name = tensor<string, []>("op_2679_cast")];
            tensor<fp16, []> var_2680_to_fp16 = const()[name = tensor<string, []>("op_2680_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_449_cast = mul(x = var_2679_cast, y = var_2680_to_fp16)[name = tensor<string, []>("aw_chunk_449_cast")];
            tensor<string, []> var_2683_equation_0 = const()[name = tensor<string, []>("op_2683_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2683_cast = einsum(equation = var_2683_equation_0, values = (var_2505_cast, var_2454_cast))[name = tensor<string, []>("op_2683_cast")];
            tensor<fp16, []> var_2684_to_fp16 = const()[name = tensor<string, []>("op_2684_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_451_cast = mul(x = var_2683_cast, y = var_2684_to_fp16)[name = tensor<string, []>("aw_chunk_451_cast")];
            tensor<string, []> var_2687_equation_0 = const()[name = tensor<string, []>("op_2687_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2687_cast = einsum(equation = var_2687_equation_0, values = (var_2505_cast, var_2455_cast))[name = tensor<string, []>("op_2687_cast")];
            tensor<fp16, []> var_2688_to_fp16 = const()[name = tensor<string, []>("op_2688_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_453_cast = mul(x = var_2687_cast, y = var_2688_to_fp16)[name = tensor<string, []>("aw_chunk_453_cast")];
            tensor<string, []> var_2691_equation_0 = const()[name = tensor<string, []>("op_2691_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2691_cast = einsum(equation = var_2691_equation_0, values = (var_2505_cast, var_2456_cast))[name = tensor<string, []>("op_2691_cast")];
            tensor<fp16, []> var_2692_to_fp16 = const()[name = tensor<string, []>("op_2692_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_455_cast = mul(x = var_2691_cast, y = var_2692_to_fp16)[name = tensor<string, []>("aw_chunk_455_cast")];
            tensor<string, []> var_2695_equation_0 = const()[name = tensor<string, []>("op_2695_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2695_cast = einsum(equation = var_2695_equation_0, values = (var_2505_cast, var_2457_cast))[name = tensor<string, []>("op_2695_cast")];
            tensor<fp16, []> var_2696_to_fp16 = const()[name = tensor<string, []>("op_2696_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_457_cast = mul(x = var_2695_cast, y = var_2696_to_fp16)[name = tensor<string, []>("aw_chunk_457_cast")];
            tensor<string, []> var_2699_equation_0 = const()[name = tensor<string, []>("op_2699_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2699_cast = einsum(equation = var_2699_equation_0, values = (var_2505_cast, var_2458_cast))[name = tensor<string, []>("op_2699_cast")];
            tensor<fp16, []> var_2700_to_fp16 = const()[name = tensor<string, []>("op_2700_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_459_cast = mul(x = var_2699_cast, y = var_2700_to_fp16)[name = tensor<string, []>("aw_chunk_459_cast")];
            tensor<string, []> var_2703_equation_0 = const()[name = tensor<string, []>("op_2703_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2703_cast = einsum(equation = var_2703_equation_0, values = (var_2505_cast, var_2459_cast))[name = tensor<string, []>("op_2703_cast")];
            tensor<fp16, []> var_2704_to_fp16 = const()[name = tensor<string, []>("op_2704_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_461_cast = mul(x = var_2703_cast, y = var_2704_to_fp16)[name = tensor<string, []>("aw_chunk_461_cast")];
            tensor<string, []> var_2707_equation_0 = const()[name = tensor<string, []>("op_2707_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2707_cast = einsum(equation = var_2707_equation_0, values = (var_2505_cast, var_2460_cast))[name = tensor<string, []>("op_2707_cast")];
            tensor<fp16, []> var_2708_to_fp16 = const()[name = tensor<string, []>("op_2708_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_463_cast = mul(x = var_2707_cast, y = var_2708_to_fp16)[name = tensor<string, []>("aw_chunk_463_cast")];
            tensor<string, []> var_2711_equation_0 = const()[name = tensor<string, []>("op_2711_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2711_cast = einsum(equation = var_2711_equation_0, values = (var_2509_cast, var_2461_cast))[name = tensor<string, []>("op_2711_cast")];
            tensor<fp16, []> var_2712_to_fp16 = const()[name = tensor<string, []>("op_2712_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_465_cast = mul(x = var_2711_cast, y = var_2712_to_fp16)[name = tensor<string, []>("aw_chunk_465_cast")];
            tensor<string, []> var_2715_equation_0 = const()[name = tensor<string, []>("op_2715_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2715_cast = einsum(equation = var_2715_equation_0, values = (var_2509_cast, var_2462_cast))[name = tensor<string, []>("op_2715_cast")];
            tensor<fp16, []> var_2716_to_fp16 = const()[name = tensor<string, []>("op_2716_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_467_cast = mul(x = var_2715_cast, y = var_2716_to_fp16)[name = tensor<string, []>("aw_chunk_467_cast")];
            tensor<string, []> var_2719_equation_0 = const()[name = tensor<string, []>("op_2719_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2719_cast = einsum(equation = var_2719_equation_0, values = (var_2509_cast, var_2463_cast))[name = tensor<string, []>("op_2719_cast")];
            tensor<fp16, []> var_2720_to_fp16 = const()[name = tensor<string, []>("op_2720_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_469_cast = mul(x = var_2719_cast, y = var_2720_to_fp16)[name = tensor<string, []>("aw_chunk_469_cast")];
            tensor<string, []> var_2723_equation_0 = const()[name = tensor<string, []>("op_2723_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2723_cast = einsum(equation = var_2723_equation_0, values = (var_2509_cast, var_2464_cast))[name = tensor<string, []>("op_2723_cast")];
            tensor<fp16, []> var_2724_to_fp16 = const()[name = tensor<string, []>("op_2724_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_471_cast = mul(x = var_2723_cast, y = var_2724_to_fp16)[name = tensor<string, []>("aw_chunk_471_cast")];
            tensor<string, []> var_2727_equation_0 = const()[name = tensor<string, []>("op_2727_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2727_cast = einsum(equation = var_2727_equation_0, values = (var_2509_cast, var_2465_cast))[name = tensor<string, []>("op_2727_cast")];
            tensor<fp16, []> var_2728_to_fp16 = const()[name = tensor<string, []>("op_2728_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_473_cast = mul(x = var_2727_cast, y = var_2728_to_fp16)[name = tensor<string, []>("aw_chunk_473_cast")];
            tensor<string, []> var_2731_equation_0 = const()[name = tensor<string, []>("op_2731_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2731_cast = einsum(equation = var_2731_equation_0, values = (var_2509_cast, var_2466_cast))[name = tensor<string, []>("op_2731_cast")];
            tensor<fp16, []> var_2732_to_fp16 = const()[name = tensor<string, []>("op_2732_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_475_cast = mul(x = var_2731_cast, y = var_2732_to_fp16)[name = tensor<string, []>("aw_chunk_475_cast")];
            tensor<string, []> var_2735_equation_0 = const()[name = tensor<string, []>("op_2735_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2735_cast = einsum(equation = var_2735_equation_0, values = (var_2509_cast, var_2467_cast))[name = tensor<string, []>("op_2735_cast")];
            tensor<fp16, []> var_2736_to_fp16 = const()[name = tensor<string, []>("op_2736_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_477_cast = mul(x = var_2735_cast, y = var_2736_to_fp16)[name = tensor<string, []>("aw_chunk_477_cast")];
            tensor<string, []> var_2739_equation_0 = const()[name = tensor<string, []>("op_2739_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2739_cast = einsum(equation = var_2739_equation_0, values = (var_2509_cast, var_2468_cast))[name = tensor<string, []>("op_2739_cast")];
            tensor<fp16, []> var_2740_to_fp16 = const()[name = tensor<string, []>("op_2740_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_479_cast = mul(x = var_2739_cast, y = var_2740_to_fp16)[name = tensor<string, []>("aw_chunk_479_cast")];
            tensor<string, []> var_2743_equation_0 = const()[name = tensor<string, []>("op_2743_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2743_cast = einsum(equation = var_2743_equation_0, values = (var_2513_cast, var_2469_cast))[name = tensor<string, []>("op_2743_cast")];
            tensor<fp16, []> var_2744_to_fp16 = const()[name = tensor<string, []>("op_2744_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_481_cast = mul(x = var_2743_cast, y = var_2744_to_fp16)[name = tensor<string, []>("aw_chunk_481_cast")];
            tensor<string, []> var_2747_equation_0 = const()[name = tensor<string, []>("op_2747_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2747_cast = einsum(equation = var_2747_equation_0, values = (var_2513_cast, var_2470_cast))[name = tensor<string, []>("op_2747_cast")];
            tensor<fp16, []> var_2748_to_fp16 = const()[name = tensor<string, []>("op_2748_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_483_cast = mul(x = var_2747_cast, y = var_2748_to_fp16)[name = tensor<string, []>("aw_chunk_483_cast")];
            tensor<string, []> var_2751_equation_0 = const()[name = tensor<string, []>("op_2751_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2751_cast = einsum(equation = var_2751_equation_0, values = (var_2513_cast, var_2471_cast))[name = tensor<string, []>("op_2751_cast")];
            tensor<fp16, []> var_2752_to_fp16 = const()[name = tensor<string, []>("op_2752_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_485_cast = mul(x = var_2751_cast, y = var_2752_to_fp16)[name = tensor<string, []>("aw_chunk_485_cast")];
            tensor<string, []> var_2755_equation_0 = const()[name = tensor<string, []>("op_2755_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2755_cast = einsum(equation = var_2755_equation_0, values = (var_2513_cast, var_2472_cast))[name = tensor<string, []>("op_2755_cast")];
            tensor<fp16, []> var_2756_to_fp16 = const()[name = tensor<string, []>("op_2756_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_487_cast = mul(x = var_2755_cast, y = var_2756_to_fp16)[name = tensor<string, []>("aw_chunk_487_cast")];
            tensor<string, []> var_2759_equation_0 = const()[name = tensor<string, []>("op_2759_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2759_cast = einsum(equation = var_2759_equation_0, values = (var_2513_cast, var_2473_cast))[name = tensor<string, []>("op_2759_cast")];
            tensor<fp16, []> var_2760_to_fp16 = const()[name = tensor<string, []>("op_2760_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_489_cast = mul(x = var_2759_cast, y = var_2760_to_fp16)[name = tensor<string, []>("aw_chunk_489_cast")];
            tensor<string, []> var_2763_equation_0 = const()[name = tensor<string, []>("op_2763_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2763_cast = einsum(equation = var_2763_equation_0, values = (var_2513_cast, var_2474_cast))[name = tensor<string, []>("op_2763_cast")];
            tensor<fp16, []> var_2764_to_fp16 = const()[name = tensor<string, []>("op_2764_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_491_cast = mul(x = var_2763_cast, y = var_2764_to_fp16)[name = tensor<string, []>("aw_chunk_491_cast")];
            tensor<string, []> var_2767_equation_0 = const()[name = tensor<string, []>("op_2767_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2767_cast = einsum(equation = var_2767_equation_0, values = (var_2513_cast, var_2475_cast))[name = tensor<string, []>("op_2767_cast")];
            tensor<fp16, []> var_2768_to_fp16 = const()[name = tensor<string, []>("op_2768_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_493_cast = mul(x = var_2767_cast, y = var_2768_to_fp16)[name = tensor<string, []>("aw_chunk_493_cast")];
            tensor<string, []> var_2771_equation_0 = const()[name = tensor<string, []>("op_2771_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2771_cast = einsum(equation = var_2771_equation_0, values = (var_2513_cast, var_2476_cast))[name = tensor<string, []>("op_2771_cast")];
            tensor<fp16, []> var_2772_to_fp16 = const()[name = tensor<string, []>("op_2772_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_495_cast = mul(x = var_2771_cast, y = var_2772_to_fp16)[name = tensor<string, []>("aw_chunk_495_cast")];
            tensor<string, []> var_2775_equation_0 = const()[name = tensor<string, []>("op_2775_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2775_cast = einsum(equation = var_2775_equation_0, values = (var_2517_cast, var_2477_cast))[name = tensor<string, []>("op_2775_cast")];
            tensor<fp16, []> var_2776_to_fp16 = const()[name = tensor<string, []>("op_2776_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_497_cast = mul(x = var_2775_cast, y = var_2776_to_fp16)[name = tensor<string, []>("aw_chunk_497_cast")];
            tensor<string, []> var_2779_equation_0 = const()[name = tensor<string, []>("op_2779_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2779_cast = einsum(equation = var_2779_equation_0, values = (var_2517_cast, var_2478_cast))[name = tensor<string, []>("op_2779_cast")];
            tensor<fp16, []> var_2780_to_fp16 = const()[name = tensor<string, []>("op_2780_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_499_cast = mul(x = var_2779_cast, y = var_2780_to_fp16)[name = tensor<string, []>("aw_chunk_499_cast")];
            tensor<string, []> var_2783_equation_0 = const()[name = tensor<string, []>("op_2783_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2783_cast = einsum(equation = var_2783_equation_0, values = (var_2517_cast, var_2479_cast))[name = tensor<string, []>("op_2783_cast")];
            tensor<fp16, []> var_2784_to_fp16 = const()[name = tensor<string, []>("op_2784_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_501_cast = mul(x = var_2783_cast, y = var_2784_to_fp16)[name = tensor<string, []>("aw_chunk_501_cast")];
            tensor<string, []> var_2787_equation_0 = const()[name = tensor<string, []>("op_2787_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2787_cast = einsum(equation = var_2787_equation_0, values = (var_2517_cast, var_2480_cast))[name = tensor<string, []>("op_2787_cast")];
            tensor<fp16, []> var_2788_to_fp16 = const()[name = tensor<string, []>("op_2788_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_503_cast = mul(x = var_2787_cast, y = var_2788_to_fp16)[name = tensor<string, []>("aw_chunk_503_cast")];
            tensor<string, []> var_2791_equation_0 = const()[name = tensor<string, []>("op_2791_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2791_cast = einsum(equation = var_2791_equation_0, values = (var_2517_cast, var_2481_cast))[name = tensor<string, []>("op_2791_cast")];
            tensor<fp16, []> var_2792_to_fp16 = const()[name = tensor<string, []>("op_2792_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_505_cast = mul(x = var_2791_cast, y = var_2792_to_fp16)[name = tensor<string, []>("aw_chunk_505_cast")];
            tensor<string, []> var_2795_equation_0 = const()[name = tensor<string, []>("op_2795_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2795_cast = einsum(equation = var_2795_equation_0, values = (var_2517_cast, var_2482_cast))[name = tensor<string, []>("op_2795_cast")];
            tensor<fp16, []> var_2796_to_fp16 = const()[name = tensor<string, []>("op_2796_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_507_cast = mul(x = var_2795_cast, y = var_2796_to_fp16)[name = tensor<string, []>("aw_chunk_507_cast")];
            tensor<string, []> var_2799_equation_0 = const()[name = tensor<string, []>("op_2799_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2799_cast = einsum(equation = var_2799_equation_0, values = (var_2517_cast, var_2483_cast))[name = tensor<string, []>("op_2799_cast")];
            tensor<fp16, []> var_2800_to_fp16 = const()[name = tensor<string, []>("op_2800_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_509_cast = mul(x = var_2799_cast, y = var_2800_to_fp16)[name = tensor<string, []>("aw_chunk_509_cast")];
            tensor<string, []> var_2803_equation_0 = const()[name = tensor<string, []>("op_2803_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2803_cast = einsum(equation = var_2803_equation_0, values = (var_2517_cast, var_2484_cast))[name = tensor<string, []>("op_2803_cast")];
            tensor<fp16, []> var_2804_to_fp16 = const()[name = tensor<string, []>("op_2804_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_511_cast = mul(x = var_2803_cast, y = var_2804_to_fp16)[name = tensor<string, []>("aw_chunk_511_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2806_cast = softmax(axis = var_123, x = aw_chunk_385_cast)[name = tensor<string, []>("op_2806_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2807_cast = softmax(axis = var_123, x = aw_chunk_387_cast)[name = tensor<string, []>("op_2807_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2808_cast = softmax(axis = var_123, x = aw_chunk_389_cast)[name = tensor<string, []>("op_2808_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2809_cast = softmax(axis = var_123, x = aw_chunk_391_cast)[name = tensor<string, []>("op_2809_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2810_cast = softmax(axis = var_123, x = aw_chunk_393_cast)[name = tensor<string, []>("op_2810_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2811_cast = softmax(axis = var_123, x = aw_chunk_395_cast)[name = tensor<string, []>("op_2811_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2812_cast = softmax(axis = var_123, x = aw_chunk_397_cast)[name = tensor<string, []>("op_2812_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2813_cast = softmax(axis = var_123, x = aw_chunk_399_cast)[name = tensor<string, []>("op_2813_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2814_cast = softmax(axis = var_123, x = aw_chunk_401_cast)[name = tensor<string, []>("op_2814_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2815_cast = softmax(axis = var_123, x = aw_chunk_403_cast)[name = tensor<string, []>("op_2815_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2816_cast = softmax(axis = var_123, x = aw_chunk_405_cast)[name = tensor<string, []>("op_2816_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2817_cast = softmax(axis = var_123, x = aw_chunk_407_cast)[name = tensor<string, []>("op_2817_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2818_cast = softmax(axis = var_123, x = aw_chunk_409_cast)[name = tensor<string, []>("op_2818_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2819_cast = softmax(axis = var_123, x = aw_chunk_411_cast)[name = tensor<string, []>("op_2819_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2820_cast = softmax(axis = var_123, x = aw_chunk_413_cast)[name = tensor<string, []>("op_2820_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2821_cast = softmax(axis = var_123, x = aw_chunk_415_cast)[name = tensor<string, []>("op_2821_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2822_cast = softmax(axis = var_123, x = aw_chunk_417_cast)[name = tensor<string, []>("op_2822_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2823_cast = softmax(axis = var_123, x = aw_chunk_419_cast)[name = tensor<string, []>("op_2823_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2824_cast = softmax(axis = var_123, x = aw_chunk_421_cast)[name = tensor<string, []>("op_2824_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2825_cast = softmax(axis = var_123, x = aw_chunk_423_cast)[name = tensor<string, []>("op_2825_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2826_cast = softmax(axis = var_123, x = aw_chunk_425_cast)[name = tensor<string, []>("op_2826_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2827_cast = softmax(axis = var_123, x = aw_chunk_427_cast)[name = tensor<string, []>("op_2827_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2828_cast = softmax(axis = var_123, x = aw_chunk_429_cast)[name = tensor<string, []>("op_2828_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2829_cast = softmax(axis = var_123, x = aw_chunk_431_cast)[name = tensor<string, []>("op_2829_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2830_cast = softmax(axis = var_123, x = aw_chunk_433_cast)[name = tensor<string, []>("op_2830_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2831_cast = softmax(axis = var_123, x = aw_chunk_435_cast)[name = tensor<string, []>("op_2831_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2832_cast = softmax(axis = var_123, x = aw_chunk_437_cast)[name = tensor<string, []>("op_2832_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2833_cast = softmax(axis = var_123, x = aw_chunk_439_cast)[name = tensor<string, []>("op_2833_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2834_cast = softmax(axis = var_123, x = aw_chunk_441_cast)[name = tensor<string, []>("op_2834_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2835_cast = softmax(axis = var_123, x = aw_chunk_443_cast)[name = tensor<string, []>("op_2835_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2836_cast = softmax(axis = var_123, x = aw_chunk_445_cast)[name = tensor<string, []>("op_2836_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2837_cast = softmax(axis = var_123, x = aw_chunk_447_cast)[name = tensor<string, []>("op_2837_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2838_cast = softmax(axis = var_123, x = aw_chunk_449_cast)[name = tensor<string, []>("op_2838_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2839_cast = softmax(axis = var_123, x = aw_chunk_451_cast)[name = tensor<string, []>("op_2839_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2840_cast = softmax(axis = var_123, x = aw_chunk_453_cast)[name = tensor<string, []>("op_2840_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2841_cast = softmax(axis = var_123, x = aw_chunk_455_cast)[name = tensor<string, []>("op_2841_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2842_cast = softmax(axis = var_123, x = aw_chunk_457_cast)[name = tensor<string, []>("op_2842_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2843_cast = softmax(axis = var_123, x = aw_chunk_459_cast)[name = tensor<string, []>("op_2843_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2844_cast = softmax(axis = var_123, x = aw_chunk_461_cast)[name = tensor<string, []>("op_2844_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2845_cast = softmax(axis = var_123, x = aw_chunk_463_cast)[name = tensor<string, []>("op_2845_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2846_cast = softmax(axis = var_123, x = aw_chunk_465_cast)[name = tensor<string, []>("op_2846_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2847_cast = softmax(axis = var_123, x = aw_chunk_467_cast)[name = tensor<string, []>("op_2847_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2848_cast = softmax(axis = var_123, x = aw_chunk_469_cast)[name = tensor<string, []>("op_2848_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2849_cast = softmax(axis = var_123, x = aw_chunk_471_cast)[name = tensor<string, []>("op_2849_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2850_cast = softmax(axis = var_123, x = aw_chunk_473_cast)[name = tensor<string, []>("op_2850_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2851_cast = softmax(axis = var_123, x = aw_chunk_475_cast)[name = tensor<string, []>("op_2851_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2852_cast = softmax(axis = var_123, x = aw_chunk_477_cast)[name = tensor<string, []>("op_2852_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2853_cast = softmax(axis = var_123, x = aw_chunk_479_cast)[name = tensor<string, []>("op_2853_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2854_cast = softmax(axis = var_123, x = aw_chunk_481_cast)[name = tensor<string, []>("op_2854_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2855_cast = softmax(axis = var_123, x = aw_chunk_483_cast)[name = tensor<string, []>("op_2855_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2856_cast = softmax(axis = var_123, x = aw_chunk_485_cast)[name = tensor<string, []>("op_2856_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2857_cast = softmax(axis = var_123, x = aw_chunk_487_cast)[name = tensor<string, []>("op_2857_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2858_cast = softmax(axis = var_123, x = aw_chunk_489_cast)[name = tensor<string, []>("op_2858_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2859_cast = softmax(axis = var_123, x = aw_chunk_491_cast)[name = tensor<string, []>("op_2859_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2860_cast = softmax(axis = var_123, x = aw_chunk_493_cast)[name = tensor<string, []>("op_2860_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2861_cast = softmax(axis = var_123, x = aw_chunk_495_cast)[name = tensor<string, []>("op_2861_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2862_cast = softmax(axis = var_123, x = aw_chunk_497_cast)[name = tensor<string, []>("op_2862_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2863_cast = softmax(axis = var_123, x = aw_chunk_499_cast)[name = tensor<string, []>("op_2863_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2864_cast = softmax(axis = var_123, x = aw_chunk_501_cast)[name = tensor<string, []>("op_2864_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2865_cast = softmax(axis = var_123, x = aw_chunk_503_cast)[name = tensor<string, []>("op_2865_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2866_cast = softmax(axis = var_123, x = aw_chunk_505_cast)[name = tensor<string, []>("op_2866_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2867_cast = softmax(axis = var_123, x = aw_chunk_507_cast)[name = tensor<string, []>("op_2867_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2868_cast = softmax(axis = var_123, x = aw_chunk_509_cast)[name = tensor<string, []>("op_2868_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2869_cast = softmax(axis = var_123, x = aw_chunk_511_cast)[name = tensor<string, []>("op_2869_cast")];
            tensor<string, []> var_2871_equation_0 = const()[name = tensor<string, []>("op_2871_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2871_cast = einsum(equation = var_2871_equation_0, values = (var_2519_cast, var_2806_cast))[name = tensor<string, []>("op_2871_cast")];
            tensor<string, []> var_2873_equation_0 = const()[name = tensor<string, []>("op_2873_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2873_cast = einsum(equation = var_2873_equation_0, values = (var_2519_cast, var_2807_cast))[name = tensor<string, []>("op_2873_cast")];
            tensor<string, []> var_2875_equation_0 = const()[name = tensor<string, []>("op_2875_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2875_cast = einsum(equation = var_2875_equation_0, values = (var_2519_cast, var_2808_cast))[name = tensor<string, []>("op_2875_cast")];
            tensor<string, []> var_2877_equation_0 = const()[name = tensor<string, []>("op_2877_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2877_cast = einsum(equation = var_2877_equation_0, values = (var_2519_cast, var_2809_cast))[name = tensor<string, []>("op_2877_cast")];
            tensor<string, []> var_2879_equation_0 = const()[name = tensor<string, []>("op_2879_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2879_cast = einsum(equation = var_2879_equation_0, values = (var_2519_cast, var_2810_cast))[name = tensor<string, []>("op_2879_cast")];
            tensor<string, []> var_2881_equation_0 = const()[name = tensor<string, []>("op_2881_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2881_cast = einsum(equation = var_2881_equation_0, values = (var_2519_cast, var_2811_cast))[name = tensor<string, []>("op_2881_cast")];
            tensor<string, []> var_2883_equation_0 = const()[name = tensor<string, []>("op_2883_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2883_cast = einsum(equation = var_2883_equation_0, values = (var_2519_cast, var_2812_cast))[name = tensor<string, []>("op_2883_cast")];
            tensor<string, []> var_2885_equation_0 = const()[name = tensor<string, []>("op_2885_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2885_cast = einsum(equation = var_2885_equation_0, values = (var_2519_cast, var_2813_cast))[name = tensor<string, []>("op_2885_cast")];
            tensor<string, []> var_2887_equation_0 = const()[name = tensor<string, []>("op_2887_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2887_cast = einsum(equation = var_2887_equation_0, values = (var_2523_cast, var_2814_cast))[name = tensor<string, []>("op_2887_cast")];
            tensor<string, []> var_2889_equation_0 = const()[name = tensor<string, []>("op_2889_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2889_cast = einsum(equation = var_2889_equation_0, values = (var_2523_cast, var_2815_cast))[name = tensor<string, []>("op_2889_cast")];
            tensor<string, []> var_2891_equation_0 = const()[name = tensor<string, []>("op_2891_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2891_cast = einsum(equation = var_2891_equation_0, values = (var_2523_cast, var_2816_cast))[name = tensor<string, []>("op_2891_cast")];
            tensor<string, []> var_2893_equation_0 = const()[name = tensor<string, []>("op_2893_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2893_cast = einsum(equation = var_2893_equation_0, values = (var_2523_cast, var_2817_cast))[name = tensor<string, []>("op_2893_cast")];
            tensor<string, []> var_2895_equation_0 = const()[name = tensor<string, []>("op_2895_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2895_cast = einsum(equation = var_2895_equation_0, values = (var_2523_cast, var_2818_cast))[name = tensor<string, []>("op_2895_cast")];
            tensor<string, []> var_2897_equation_0 = const()[name = tensor<string, []>("op_2897_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2897_cast = einsum(equation = var_2897_equation_0, values = (var_2523_cast, var_2819_cast))[name = tensor<string, []>("op_2897_cast")];
            tensor<string, []> var_2899_equation_0 = const()[name = tensor<string, []>("op_2899_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2899_cast = einsum(equation = var_2899_equation_0, values = (var_2523_cast, var_2820_cast))[name = tensor<string, []>("op_2899_cast")];
            tensor<string, []> var_2901_equation_0 = const()[name = tensor<string, []>("op_2901_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2901_cast = einsum(equation = var_2901_equation_0, values = (var_2523_cast, var_2821_cast))[name = tensor<string, []>("op_2901_cast")];
            tensor<string, []> var_2903_equation_0 = const()[name = tensor<string, []>("op_2903_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2903_cast = einsum(equation = var_2903_equation_0, values = (var_2527_cast, var_2822_cast))[name = tensor<string, []>("op_2903_cast")];
            tensor<string, []> var_2905_equation_0 = const()[name = tensor<string, []>("op_2905_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2905_cast = einsum(equation = var_2905_equation_0, values = (var_2527_cast, var_2823_cast))[name = tensor<string, []>("op_2905_cast")];
            tensor<string, []> var_2907_equation_0 = const()[name = tensor<string, []>("op_2907_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2907_cast = einsum(equation = var_2907_equation_0, values = (var_2527_cast, var_2824_cast))[name = tensor<string, []>("op_2907_cast")];
            tensor<string, []> var_2909_equation_0 = const()[name = tensor<string, []>("op_2909_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2909_cast = einsum(equation = var_2909_equation_0, values = (var_2527_cast, var_2825_cast))[name = tensor<string, []>("op_2909_cast")];
            tensor<string, []> var_2911_equation_0 = const()[name = tensor<string, []>("op_2911_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2911_cast = einsum(equation = var_2911_equation_0, values = (var_2527_cast, var_2826_cast))[name = tensor<string, []>("op_2911_cast")];
            tensor<string, []> var_2913_equation_0 = const()[name = tensor<string, []>("op_2913_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2913_cast = einsum(equation = var_2913_equation_0, values = (var_2527_cast, var_2827_cast))[name = tensor<string, []>("op_2913_cast")];
            tensor<string, []> var_2915_equation_0 = const()[name = tensor<string, []>("op_2915_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2915_cast = einsum(equation = var_2915_equation_0, values = (var_2527_cast, var_2828_cast))[name = tensor<string, []>("op_2915_cast")];
            tensor<string, []> var_2917_equation_0 = const()[name = tensor<string, []>("op_2917_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2917_cast = einsum(equation = var_2917_equation_0, values = (var_2527_cast, var_2829_cast))[name = tensor<string, []>("op_2917_cast")];
            tensor<string, []> var_2919_equation_0 = const()[name = tensor<string, []>("op_2919_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2919_cast = einsum(equation = var_2919_equation_0, values = (var_2531_cast, var_2830_cast))[name = tensor<string, []>("op_2919_cast")];
            tensor<string, []> var_2921_equation_0 = const()[name = tensor<string, []>("op_2921_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2921_cast = einsum(equation = var_2921_equation_0, values = (var_2531_cast, var_2831_cast))[name = tensor<string, []>("op_2921_cast")];
            tensor<string, []> var_2923_equation_0 = const()[name = tensor<string, []>("op_2923_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2923_cast = einsum(equation = var_2923_equation_0, values = (var_2531_cast, var_2832_cast))[name = tensor<string, []>("op_2923_cast")];
            tensor<string, []> var_2925_equation_0 = const()[name = tensor<string, []>("op_2925_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2925_cast = einsum(equation = var_2925_equation_0, values = (var_2531_cast, var_2833_cast))[name = tensor<string, []>("op_2925_cast")];
            tensor<string, []> var_2927_equation_0 = const()[name = tensor<string, []>("op_2927_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2927_cast = einsum(equation = var_2927_equation_0, values = (var_2531_cast, var_2834_cast))[name = tensor<string, []>("op_2927_cast")];
            tensor<string, []> var_2929_equation_0 = const()[name = tensor<string, []>("op_2929_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2929_cast = einsum(equation = var_2929_equation_0, values = (var_2531_cast, var_2835_cast))[name = tensor<string, []>("op_2929_cast")];
            tensor<string, []> var_2931_equation_0 = const()[name = tensor<string, []>("op_2931_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2931_cast = einsum(equation = var_2931_equation_0, values = (var_2531_cast, var_2836_cast))[name = tensor<string, []>("op_2931_cast")];
            tensor<string, []> var_2933_equation_0 = const()[name = tensor<string, []>("op_2933_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2933_cast = einsum(equation = var_2933_equation_0, values = (var_2531_cast, var_2837_cast))[name = tensor<string, []>("op_2933_cast")];
            tensor<string, []> var_2935_equation_0 = const()[name = tensor<string, []>("op_2935_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2935_cast = einsum(equation = var_2935_equation_0, values = (var_2535_cast, var_2838_cast))[name = tensor<string, []>("op_2935_cast")];
            tensor<string, []> var_2937_equation_0 = const()[name = tensor<string, []>("op_2937_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2937_cast = einsum(equation = var_2937_equation_0, values = (var_2535_cast, var_2839_cast))[name = tensor<string, []>("op_2937_cast")];
            tensor<string, []> var_2939_equation_0 = const()[name = tensor<string, []>("op_2939_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2939_cast = einsum(equation = var_2939_equation_0, values = (var_2535_cast, var_2840_cast))[name = tensor<string, []>("op_2939_cast")];
            tensor<string, []> var_2941_equation_0 = const()[name = tensor<string, []>("op_2941_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2941_cast = einsum(equation = var_2941_equation_0, values = (var_2535_cast, var_2841_cast))[name = tensor<string, []>("op_2941_cast")];
            tensor<string, []> var_2943_equation_0 = const()[name = tensor<string, []>("op_2943_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2943_cast = einsum(equation = var_2943_equation_0, values = (var_2535_cast, var_2842_cast))[name = tensor<string, []>("op_2943_cast")];
            tensor<string, []> var_2945_equation_0 = const()[name = tensor<string, []>("op_2945_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2945_cast = einsum(equation = var_2945_equation_0, values = (var_2535_cast, var_2843_cast))[name = tensor<string, []>("op_2945_cast")];
            tensor<string, []> var_2947_equation_0 = const()[name = tensor<string, []>("op_2947_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2947_cast = einsum(equation = var_2947_equation_0, values = (var_2535_cast, var_2844_cast))[name = tensor<string, []>("op_2947_cast")];
            tensor<string, []> var_2949_equation_0 = const()[name = tensor<string, []>("op_2949_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2949_cast = einsum(equation = var_2949_equation_0, values = (var_2535_cast, var_2845_cast))[name = tensor<string, []>("op_2949_cast")];
            tensor<string, []> var_2951_equation_0 = const()[name = tensor<string, []>("op_2951_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2951_cast = einsum(equation = var_2951_equation_0, values = (var_2539_cast, var_2846_cast))[name = tensor<string, []>("op_2951_cast")];
            tensor<string, []> var_2953_equation_0 = const()[name = tensor<string, []>("op_2953_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2953_cast = einsum(equation = var_2953_equation_0, values = (var_2539_cast, var_2847_cast))[name = tensor<string, []>("op_2953_cast")];
            tensor<string, []> var_2955_equation_0 = const()[name = tensor<string, []>("op_2955_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2955_cast = einsum(equation = var_2955_equation_0, values = (var_2539_cast, var_2848_cast))[name = tensor<string, []>("op_2955_cast")];
            tensor<string, []> var_2957_equation_0 = const()[name = tensor<string, []>("op_2957_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2957_cast = einsum(equation = var_2957_equation_0, values = (var_2539_cast, var_2849_cast))[name = tensor<string, []>("op_2957_cast")];
            tensor<string, []> var_2959_equation_0 = const()[name = tensor<string, []>("op_2959_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2959_cast = einsum(equation = var_2959_equation_0, values = (var_2539_cast, var_2850_cast))[name = tensor<string, []>("op_2959_cast")];
            tensor<string, []> var_2961_equation_0 = const()[name = tensor<string, []>("op_2961_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2961_cast = einsum(equation = var_2961_equation_0, values = (var_2539_cast, var_2851_cast))[name = tensor<string, []>("op_2961_cast")];
            tensor<string, []> var_2963_equation_0 = const()[name = tensor<string, []>("op_2963_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2963_cast = einsum(equation = var_2963_equation_0, values = (var_2539_cast, var_2852_cast))[name = tensor<string, []>("op_2963_cast")];
            tensor<string, []> var_2965_equation_0 = const()[name = tensor<string, []>("op_2965_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2965_cast = einsum(equation = var_2965_equation_0, values = (var_2539_cast, var_2853_cast))[name = tensor<string, []>("op_2965_cast")];
            tensor<string, []> var_2967_equation_0 = const()[name = tensor<string, []>("op_2967_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2967_cast = einsum(equation = var_2967_equation_0, values = (var_2543_cast, var_2854_cast))[name = tensor<string, []>("op_2967_cast")];
            tensor<string, []> var_2969_equation_0 = const()[name = tensor<string, []>("op_2969_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2969_cast = einsum(equation = var_2969_equation_0, values = (var_2543_cast, var_2855_cast))[name = tensor<string, []>("op_2969_cast")];
            tensor<string, []> var_2971_equation_0 = const()[name = tensor<string, []>("op_2971_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2971_cast = einsum(equation = var_2971_equation_0, values = (var_2543_cast, var_2856_cast))[name = tensor<string, []>("op_2971_cast")];
            tensor<string, []> var_2973_equation_0 = const()[name = tensor<string, []>("op_2973_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2973_cast = einsum(equation = var_2973_equation_0, values = (var_2543_cast, var_2857_cast))[name = tensor<string, []>("op_2973_cast")];
            tensor<string, []> var_2975_equation_0 = const()[name = tensor<string, []>("op_2975_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2975_cast = einsum(equation = var_2975_equation_0, values = (var_2543_cast, var_2858_cast))[name = tensor<string, []>("op_2975_cast")];
            tensor<string, []> var_2977_equation_0 = const()[name = tensor<string, []>("op_2977_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2977_cast = einsum(equation = var_2977_equation_0, values = (var_2543_cast, var_2859_cast))[name = tensor<string, []>("op_2977_cast")];
            tensor<string, []> var_2979_equation_0 = const()[name = tensor<string, []>("op_2979_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2979_cast = einsum(equation = var_2979_equation_0, values = (var_2543_cast, var_2860_cast))[name = tensor<string, []>("op_2979_cast")];
            tensor<string, []> var_2981_equation_0 = const()[name = tensor<string, []>("op_2981_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2981_cast = einsum(equation = var_2981_equation_0, values = (var_2543_cast, var_2861_cast))[name = tensor<string, []>("op_2981_cast")];
            tensor<string, []> var_2983_equation_0 = const()[name = tensor<string, []>("op_2983_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2983_cast = einsum(equation = var_2983_equation_0, values = (var_2547_cast, var_2862_cast))[name = tensor<string, []>("op_2983_cast")];
            tensor<string, []> var_2985_equation_0 = const()[name = tensor<string, []>("op_2985_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2985_cast = einsum(equation = var_2985_equation_0, values = (var_2547_cast, var_2863_cast))[name = tensor<string, []>("op_2985_cast")];
            tensor<string, []> var_2987_equation_0 = const()[name = tensor<string, []>("op_2987_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2987_cast = einsum(equation = var_2987_equation_0, values = (var_2547_cast, var_2864_cast))[name = tensor<string, []>("op_2987_cast")];
            tensor<string, []> var_2989_equation_0 = const()[name = tensor<string, []>("op_2989_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2989_cast = einsum(equation = var_2989_equation_0, values = (var_2547_cast, var_2865_cast))[name = tensor<string, []>("op_2989_cast")];
            tensor<string, []> var_2991_equation_0 = const()[name = tensor<string, []>("op_2991_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2991_cast = einsum(equation = var_2991_equation_0, values = (var_2547_cast, var_2866_cast))[name = tensor<string, []>("op_2991_cast")];
            tensor<string, []> var_2993_equation_0 = const()[name = tensor<string, []>("op_2993_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2993_cast = einsum(equation = var_2993_equation_0, values = (var_2547_cast, var_2867_cast))[name = tensor<string, []>("op_2993_cast")];
            tensor<string, []> var_2995_equation_0 = const()[name = tensor<string, []>("op_2995_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2995_cast = einsum(equation = var_2995_equation_0, values = (var_2547_cast, var_2868_cast))[name = tensor<string, []>("op_2995_cast")];
            tensor<string, []> var_2997_equation_0 = const()[name = tensor<string, []>("op_2997_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_2997_cast = einsum(equation = var_2997_equation_0, values = (var_2547_cast, var_2869_cast))[name = tensor<string, []>("op_2997_cast")];
            tensor<bool, []> var_2999_interleave_0 = const()[name = tensor<string, []>("op_2999_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_2999_cast = concat(axis = var_95, interleave = var_2999_interleave_0, values = (var_2871_cast, var_2873_cast, var_2875_cast, var_2877_cast, var_2879_cast, var_2881_cast, var_2883_cast, var_2885_cast))[name = tensor<string, []>("op_2999_cast")];
            tensor<bool, []> var_3001_interleave_0 = const()[name = tensor<string, []>("op_3001_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_3001_cast = concat(axis = var_95, interleave = var_3001_interleave_0, values = (var_2887_cast, var_2889_cast, var_2891_cast, var_2893_cast, var_2895_cast, var_2897_cast, var_2899_cast, var_2901_cast))[name = tensor<string, []>("op_3001_cast")];
            tensor<bool, []> var_3003_interleave_0 = const()[name = tensor<string, []>("op_3003_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_3003_cast = concat(axis = var_95, interleave = var_3003_interleave_0, values = (var_2903_cast, var_2905_cast, var_2907_cast, var_2909_cast, var_2911_cast, var_2913_cast, var_2915_cast, var_2917_cast))[name = tensor<string, []>("op_3003_cast")];
            tensor<bool, []> var_3005_interleave_0 = const()[name = tensor<string, []>("op_3005_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_3005_cast = concat(axis = var_95, interleave = var_3005_interleave_0, values = (var_2919_cast, var_2921_cast, var_2923_cast, var_2925_cast, var_2927_cast, var_2929_cast, var_2931_cast, var_2933_cast))[name = tensor<string, []>("op_3005_cast")];
            tensor<bool, []> var_3007_interleave_0 = const()[name = tensor<string, []>("op_3007_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_3007_cast = concat(axis = var_95, interleave = var_3007_interleave_0, values = (var_2935_cast, var_2937_cast, var_2939_cast, var_2941_cast, var_2943_cast, var_2945_cast, var_2947_cast, var_2949_cast))[name = tensor<string, []>("op_3007_cast")];
            tensor<bool, []> var_3009_interleave_0 = const()[name = tensor<string, []>("op_3009_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_3009_cast = concat(axis = var_95, interleave = var_3009_interleave_0, values = (var_2951_cast, var_2953_cast, var_2955_cast, var_2957_cast, var_2959_cast, var_2961_cast, var_2963_cast, var_2965_cast))[name = tensor<string, []>("op_3009_cast")];
            tensor<bool, []> var_3011_interleave_0 = const()[name = tensor<string, []>("op_3011_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_3011_cast = concat(axis = var_95, interleave = var_3011_interleave_0, values = (var_2967_cast, var_2969_cast, var_2971_cast, var_2973_cast, var_2975_cast, var_2977_cast, var_2979_cast, var_2981_cast))[name = tensor<string, []>("op_3011_cast")];
            tensor<bool, []> var_3013_interleave_0 = const()[name = tensor<string, []>("op_3013_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_3013_cast = concat(axis = var_95, interleave = var_3013_interleave_0, values = (var_2983_cast, var_2985_cast, var_2987_cast, var_2989_cast, var_2991_cast, var_2993_cast, var_2995_cast, var_2997_cast))[name = tensor<string, []>("op_3013_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_123, interleave = input_53_interleave_0, values = (var_2999_cast, var_3001_cast, var_3003_cast, var_3005_cast, var_3007_cast, var_3009_cast, var_3011_cast, var_3013_cast))[name = tensor<string, []>("input_53_cast")];
            tensor<int32, [2]> var_3019 = const()[name = tensor<string, []>("op_3019"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3021 = const()[name = tensor<string, []>("op_3021"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_3023_pad_type_0 = const()[name = tensor<string, []>("op_3023_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_3023_pad_0 = const()[name = tensor<string, []>("op_3023_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7699968))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7776832))), name = tensor<string, []>("down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            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, []>(7777024)))];
            tensor<fp16, [2, 320, 1, 4096]> var_3023_cast = conv(bias = down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_3021, groups = var_123, pad = var_3023_pad_0, pad_type = var_3023_pad_type_0, strides = var_3019, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_53_cast)[name = tensor<string, []>("op_3023_cast")];
            tensor<fp16, [2, 320, 1, 4096]> inputs_11_cast = add(x = var_3023_cast, y = inputs_9_cast)[name = tensor<string, []>("inputs_11_cast")];
            tensor<int32, [1]> var_3027 = const()[name = tensor<string, []>("op_3027"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_11_cast = reduce_mean(axes = var_3027, keep_dims = var_118, x = inputs_11_cast)[name = tensor<string, []>("channels_mean_11_cast")];
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_11_cast = sub(x = inputs_11_cast, y = channels_mean_11_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_11_cast")];
            tensor<int32, [1]> var_3031 = const()[name = tensor<string, []>("op_3031"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_3032_cast = reduce_mean(axes = var_3031, keep_dims = var_118, x = zero_mean_sq_11_cast)[name = tensor<string, []>("op_3032_cast")];
            tensor<fp16, []> var_3033_to_fp16 = const()[name = tensor<string, []>("op_3033_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_3034_cast = add(x = var_3032_cast, y = var_3033_to_fp16)[name = tensor<string, []>("op_3034_cast")];
            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_3034_cast)[name = tensor<string, []>("denom_11_cast")];
            tensor<fp16, [2, 320, 1, 4096]> out_11_cast = mul(x = zero_mean_11_cast, y = denom_11_cast)[name = tensor<string, []>("out_11_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_3038_to_fp16 = const()[name = tensor<string, []>("op_3038_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7777728)))];
            tensor<fp16, [2, 320, 1, 4096]> var_3039_cast = add(x = out_11_cast, y = var_3038_to_fp16)[name = tensor<string, []>("op_3039_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_3041_to_fp16 = const()[name = tensor<string, []>("op_3041_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7778432)))];
            tensor<fp16, [2, 320, 1, 4096]> input_55_cast = mul(x = var_3039_cast, y = var_3041_to_fp16)[name = tensor<string, []>("input_55_cast")];
            tensor<int32, [2]> var_3049 = const()[name = tensor<string, []>("op_3049"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3051 = const()[name = tensor<string, []>("op_3051"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_3053_pad_type_0 = const()[name = tensor<string, []>("op_3053_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_3053_pad_0 = const()[name = tensor<string, []>("op_3053_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [614400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7779136))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8393600))), name = tensor<string, []>("down_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([2560, 320, 1, 1])];
            tensor<fp16, [2560]> down_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8393792))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8395776))), name = tensor<string, []>("down_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([2560])];
            tensor<fp16, [2, 2560, 1, 4096]> var_3053_cast = conv(bias = down_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_3051, groups = var_123, pad = var_3053_pad_0, pad_type = var_3053_pad_type_0, strides = var_3049, weight = down_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_55_cast)[name = tensor<string, []>("op_3053_cast")];
            tensor<int32, [2]> var_3054_split_sizes_0 = const()[name = tensor<string, []>("op_3054_split_sizes_0"), val = tensor<int32, [2]>([1280, 1280])];
            tensor<int32, []> var_3054_axis_0 = const()[name = tensor<string, []>("op_3054_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 1280, 1, 4096]> var_3054_cast_0, tensor<fp16, [2, 1280, 1, 4096]> var_3054_cast_1 = split(axis = var_3054_axis_0, split_sizes = var_3054_split_sizes_0, x = var_3053_cast)[name = tensor<string, []>("op_3054_cast")];
            tensor<string, []> var_3056_mode_0 = const()[name = tensor<string, []>("op_3056_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 1280, 1, 4096]> var_3056_cast = gelu(mode = var_3056_mode_0, x = var_3054_cast_1)[name = tensor<string, []>("op_3056_cast")];
            tensor<fp16, [2, 1280, 1, 4096]> input_57_cast = mul(x = var_3054_cast_0, y = var_3056_cast)[name = tensor<string, []>("input_57_cast")];
            tensor<int32, [2]> var_3060 = const()[name = tensor<string, []>("op_3060"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3062 = const()[name = tensor<string, []>("op_3062"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_3064_pad_type_0 = const()[name = tensor<string, []>("op_3064_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_3064_pad_0 = const()[name = tensor<string, []>("op_3064_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8395968))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8703232))), name = tensor<string, []>("down_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 1280, 1, 1])];
            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, []>(8703424)))];
            tensor<fp16, [2, 320, 1, 4096]> var_3064_cast = conv(bias = down_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_3062, groups = var_123, pad = var_3064_pad_0, pad_type = var_3064_pad_type_0, strides = var_3060, weight = down_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_57_cast)[name = tensor<string, []>("op_3064_cast")];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_33_cast = add(x = var_3064_cast, y = inputs_11_cast)[name = tensor<string, []>("hidden_states_33_cast")];
            tensor<int32, [4]> var_3066 = const()[name = tensor<string, []>("op_3066"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> input_59_cast = reshape(shape = var_3066, x = hidden_states_33_cast)[name = tensor<string, []>("input_59_cast")];
            tensor<int32, [2]> var_3070 = const()[name = tensor<string, []>("op_3070"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3072 = const()[name = tensor<string, []>("op_3072"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8704128))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8780992))), name = tensor<string, []>("down_blocks_0_attentions_1_proj_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            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, []>(8781184)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_35_cast = conv(bias = down_blocks_0_attentions_1_proj_out_bias_to_fp16, dilations = var_3072, groups = var_123, pad = hidden_states_35_pad_0, pad_type = hidden_states_35_pad_type_0, strides = var_3070, weight = down_blocks_0_attentions_1_proj_out_weight_to_fp16_palettized, x = input_59_cast)[name = tensor<string, []>("hidden_states_35_cast")];
            tensor<fp16, [2, 320, 64, 64]> input_61_cast = add(x = hidden_states_35_cast, y = hidden_states_23_cast)[name = tensor<string, []>("input_61_cast")];
            tensor<int32, [2]> var_3079 = const()[name = tensor<string, []>("op_3079"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [2]> var_3081 = const()[name = tensor<string, []>("op_3081"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [691200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8781888))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9473152))), name = tensor<string, []>("down_blocks_0_downsamplers_0_conv_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 3, 3])];
            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, []>(9473344)))];
            tensor<fp16, [2, 320, 32, 32]> input_63_cast = conv(bias = down_blocks_0_downsamplers_0_conv_bias_to_fp16, dilations = var_3081, groups = var_123, pad = input_63_pad_0, pad_type = input_63_pad_type_0, strides = var_3079, weight = down_blocks_0_downsamplers_0_conv_weight_to_fp16_palettized, x = input_61_cast)[name = tensor<string, []>("input_63_cast")];
            tensor<int32, []> var_3089 = const()[name = tensor<string, []>("op_3089"), val = tensor<int32, []>(3)];
            tensor<bool, []> var_3106 = const()[name = tensor<string, []>("op_3106"), val = tensor<bool, []>(true)];
            tensor<int32, []> var_3111 = const()[name = tensor<string, []>("op_3111"), 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)[name = tensor<string, []>("reshape_24_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)[name = tensor<string, []>("reduce_mean_18_cast")];
            tensor<fp16, [2, 32, 10, 32, 32]> sub_12_cast = sub(x = reshape_24_cast, y = reduce_mean_18_cast)[name = tensor<string, []>("sub_12_cast")];
            tensor<fp16, [2, 32, 10, 32, 32]> square_6_cast = square(x = sub_12_cast)[name = tensor<string, []>("square_6_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)[name = tensor<string, []>("reduce_mean_20_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)[name = tensor<string, []>("add_12_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_6_cast = sqrt(x = add_12_cast)[name = tensor<string, []>("sqrt_6_cast")];
            tensor<fp16, [2, 32, 10, 32, 32]> real_div_6_cast = real_div(x = sub_12_cast, y = sqrt_6_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_25_cast")];
            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, []>(9474048)))];
            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, []>(9474752)))];
            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_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_25_cast)[name = tensor<string, []>("add_13_cast")];
            tensor<fp16, [2, 320, 32, 32]> input_67_cast = silu(x = add_13_cast)[name = tensor<string, []>("input_67_cast")];
            tensor<int32, [2]> var_3134 = const()[name = tensor<string, []>("op_3134"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3136 = const()[name = tensor<string, []>("op_3136"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1382400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9475456))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10857920))), name = tensor<string, []>("down_blocks_1_resnets_0_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 320, 3, 3])];
            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, []>(10858112)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_37_cast = conv(bias = down_blocks_1_resnets_0_conv1_bias_to_fp16, dilations = var_3136, groups = var_3111, pad = hidden_states_37_pad_0, pad_type = hidden_states_37_pad_type_0, strides = var_3134, weight = down_blocks_1_resnets_0_conv1_weight_to_fp16_palettized, x = input_67_cast)[name = tensor<string, []>("hidden_states_37_cast")];
            tensor<int32, [2]> var_3142 = const()[name = tensor<string, []>("op_3142"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3144 = const()[name = tensor<string, []>("op_3144"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [614400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10859456))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11473920))), name = tensor<string, []>("down_blocks_1_resnets_0_time_emb_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 1280, 1, 1])];
            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, []>(11474112)))];
            tensor<fp16, [2, 640, 1, 1]> temb_5_cast = conv(bias = down_blocks_1_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_3144, groups = var_3111, pad = temb_5_pad_0, pad_type = temb_5_pad_type_0, strides = var_3142, weight = down_blocks_1_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor<string, []>("temb_5_cast")];
            tensor<fp16, [2, 640, 32, 32]> input_71_cast = add(x = hidden_states_37_cast, y = temb_5_cast)[name = tensor<string, []>("input_71_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)[name = tensor<string, []>("reshape_28_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)[name = tensor<string, []>("reduce_mean_21_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> sub_14_cast = sub(x = reshape_28_cast, y = reduce_mean_21_cast)[name = tensor<string, []>("sub_14_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> square_7_cast = square(x = sub_14_cast)[name = tensor<string, []>("square_7_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)[name = tensor<string, []>("reduce_mean_23_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)[name = tensor<string, []>("add_14_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_7_cast = sqrt(x = add_14_cast)[name = tensor<string, []>("sqrt_7_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> real_div_7_cast = real_div(x = sub_14_cast, y = sqrt_7_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_29_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, []>(11475456)))];
            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, []>(11476800)))];
            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, []>(11478144)))];
            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, []>(11479488)))];
            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)[name = tensor<string, []>("add_15_cast")];
            tensor<fp16, [2, 640, 32, 32]> input_75_cast = silu(x = add_15_cast)[name = tensor<string, []>("input_75_cast")];
            tensor<int32, [2]> var_3154 = const()[name = tensor<string, []>("op_3154"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3156 = const()[name = tensor<string, []>("op_3156"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2764800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11480832))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14245696))), name = tensor<string, []>("down_blocks_1_resnets_0_conv2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 3, 3])];
            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, []>(14245888)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_39_cast = conv(bias = down_blocks_1_resnets_0_conv2_bias_to_fp16, dilations = var_3156, groups = var_3111, pad = hidden_states_39_pad_0, pad_type = hidden_states_39_pad_type_0, strides = var_3154, weight = down_blocks_1_resnets_0_conv2_weight_to_fp16_palettized, x = input_75_cast)[name = tensor<string, []>("hidden_states_39_cast")];
            tensor<int32, [2]> var_3161 = const()[name = tensor<string, []>("op_3161"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3163 = const()[name = tensor<string, []>("op_3163"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [153600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14247232))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14400896))), name = tensor<string, []>("down_blocks_1_resnets_0_conv_shortcut_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 320, 1, 1])];
            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, []>(14401088)))];
            tensor<fp16, [2, 640, 32, 32]> x_1_cast = conv(bias = down_blocks_1_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_3163, groups = var_3111, pad = x_1_pad_0, pad_type = x_1_pad_type_0, strides = var_3161, weight = down_blocks_1_resnets_0_conv_shortcut_weight_to_fp16_palettized, x = input_63_cast)[name = tensor<string, []>("x_1_cast")];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_41_cast = add(x = x_1_cast, y = hidden_states_39_cast)[name = tensor<string, []>("hidden_states_41_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)[name = tensor<string, []>("reshape_32_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)[name = tensor<string, []>("reduce_mean_24_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> sub_16_cast = sub(x = reshape_32_cast, y = reduce_mean_24_cast)[name = tensor<string, []>("sub_16_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> square_8_cast = square(x = sub_16_cast)[name = tensor<string, []>("square_8_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)[name = tensor<string, []>("reduce_mean_26_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)[name = tensor<string, []>("add_16_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_8_cast = sqrt(x = add_16_cast)[name = tensor<string, []>("sqrt_8_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> real_div_8_cast = real_div(x = sub_16_cast, y = sqrt_8_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_33_cast")];
            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, []>(14402432)))];
            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, []>(14403776)))];
            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_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_33_cast)[name = tensor<string, []>("add_17_cast")];
            tensor<int32, [2]> var_3183 = const()[name = tensor<string, []>("op_3183"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3185 = const()[name = tensor<string, []>("op_3185"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14405120))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14712384))), name = tensor<string, []>("down_blocks_1_attentions_0_proj_in_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            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, []>(14712576)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_43_cast = conv(bias = down_blocks_1_attentions_0_proj_in_bias_to_fp16, dilations = var_3185, groups = var_3111, pad = hidden_states_43_pad_0, pad_type = hidden_states_43_pad_type_0, strides = var_3183, weight = down_blocks_1_attentions_0_proj_in_weight_to_fp16_palettized, x = add_17_cast)[name = tensor<string, []>("hidden_states_43_cast")];
            tensor<int32, [4]> var_3190 = const()[name = tensor<string, []>("op_3190"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<fp16, [2, 640, 1, 1024]> inputs_13_cast = reshape(shape = var_3190, x = hidden_states_43_cast)[name = tensor<string, []>("inputs_13_cast")];
            tensor<int32, [1]> var_3200 = const()[name = tensor<string, []>("op_3200"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_13_cast = reduce_mean(axes = var_3200, keep_dims = var_3106, x = inputs_13_cast)[name = tensor<string, []>("channels_mean_13_cast")];
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_13_cast = sub(x = inputs_13_cast, y = channels_mean_13_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_13_cast")];
            tensor<int32, [1]> var_3204 = const()[name = tensor<string, []>("op_3204"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_3205_cast = reduce_mean(axes = var_3204, keep_dims = var_3106, x = zero_mean_sq_13_cast)[name = tensor<string, []>("op_3205_cast")];
            tensor<fp16, []> var_3206_to_fp16 = const()[name = tensor<string, []>("op_3206_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_3207_cast = add(x = var_3205_cast, y = var_3206_to_fp16)[name = tensor<string, []>("op_3207_cast")];
            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_3207_cast)[name = tensor<string, []>("denom_13_cast")];
            tensor<fp16, [2, 640, 1, 1024]> out_13_cast = mul(x = zero_mean_13_cast, y = denom_13_cast)[name = tensor<string, []>("out_13_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_3211_to_fp16 = const()[name = tensor<string, []>("op_3211_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14713920)))];
            tensor<fp16, [2, 640, 1, 1024]> var_3212_cast = add(x = out_13_cast, y = var_3211_to_fp16)[name = tensor<string, []>("op_3212_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_3214_to_fp16 = const()[name = tensor<string, []>("op_3214_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14715264)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_45_cast = mul(x = var_3212_cast, y = var_3214_to_fp16)[name = tensor<string, []>("hidden_states_45_cast")];
            tensor<int32, [2]> var_3221 = const()[name = tensor<string, []>("op_3221"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3223 = const()[name = tensor<string, []>("op_3223"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14716608))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15023872))), name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 1024]> q_9_cast = conv(dilations = var_3223, groups = var_3111, pad = q_9_pad_0, pad_type = q_9_pad_type_0, strides = var_3221, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_45_cast)[name = tensor<string, []>("q_9_cast")];
            tensor<int32, [2]> var_3227 = const()[name = tensor<string, []>("op_3227"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3229 = const()[name = tensor<string, []>("op_3229"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15024064))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15331328))), name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 1024]> k_17_cast = conv(dilations = var_3229, groups = var_3111, pad = k_17_pad_0, pad_type = k_17_pad_type_0, strides = var_3227, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_45_cast)[name = tensor<string, []>("k_17_cast")];
            tensor<int32, [2]> var_3233 = const()[name = tensor<string, []>("op_3233"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3235 = const()[name = tensor<string, []>("op_3235"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15331520))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15638784))), name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 1024]> v_9_cast = conv(dilations = var_3235, groups = var_3111, pad = v_9_pad_0, pad_type = v_9_pad_type_0, strides = var_3233, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_45_cast)[name = tensor<string, []>("v_9_cast")];
            tensor<int32, [4]> var_3239_begin_0 = const()[name = tensor<string, []>("op_3239_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3239_end_0 = const()[name = tensor<string, []>("op_3239_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_3239_end_mask_0 = const()[name = tensor<string, []>("op_3239_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_3239_cast = slice_by_index(begin = var_3239_begin_0, end = var_3239_end_0, end_mask = var_3239_end_mask_0, x = q_9_cast)[name = tensor<string, []>("op_3239_cast")];
            tensor<int32, [4]> var_3243_begin_0 = const()[name = tensor<string, []>("op_3243_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_3243_end_0 = const()[name = tensor<string, []>("op_3243_end_0"), val = tensor<int32, [4]>([2, 160, 1, 1024])];
            tensor<bool, [4]> var_3243_end_mask_0 = const()[name = tensor<string, []>("op_3243_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_3243_cast = slice_by_index(begin = var_3243_begin_0, end = var_3243_end_0, end_mask = var_3243_end_mask_0, x = q_9_cast)[name = tensor<string, []>("op_3243_cast")];
            tensor<int32, [4]> var_3247_begin_0 = const()[name = tensor<string, []>("op_3247_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_3247_end_0 = const()[name = tensor<string, []>("op_3247_end_0"), val = tensor<int32, [4]>([2, 240, 1, 1024])];
            tensor<bool, [4]> var_3247_end_mask_0 = const()[name = tensor<string, []>("op_3247_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_3247_cast = slice_by_index(begin = var_3247_begin_0, end = var_3247_end_0, end_mask = var_3247_end_mask_0, x = q_9_cast)[name = tensor<string, []>("op_3247_cast")];
            tensor<int32, [4]> var_3251_begin_0 = const()[name = tensor<string, []>("op_3251_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_3251_end_0 = const()[name = tensor<string, []>("op_3251_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_3251_end_mask_0 = const()[name = tensor<string, []>("op_3251_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_3251_cast = slice_by_index(begin = var_3251_begin_0, end = var_3251_end_0, end_mask = var_3251_end_mask_0, x = q_9_cast)[name = tensor<string, []>("op_3251_cast")];
            tensor<int32, [4]> var_3255_begin_0 = const()[name = tensor<string, []>("op_3255_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_3255_end_0 = const()[name = tensor<string, []>("op_3255_end_0"), val = tensor<int32, [4]>([2, 400, 1, 1024])];
            tensor<bool, [4]> var_3255_end_mask_0 = const()[name = tensor<string, []>("op_3255_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_3255_cast = slice_by_index(begin = var_3255_begin_0, end = var_3255_end_0, end_mask = var_3255_end_mask_0, x = q_9_cast)[name = tensor<string, []>("op_3255_cast")];
            tensor<int32, [4]> var_3259_begin_0 = const()[name = tensor<string, []>("op_3259_begin_0"), val = tensor<int32, [4]>([0, 400, 0, 0])];
            tensor<int32, [4]> var_3259_end_0 = const()[name = tensor<string, []>("op_3259_end_0"), val = tensor<int32, [4]>([2, 480, 1, 1024])];
            tensor<bool, [4]> var_3259_end_mask_0 = const()[name = tensor<string, []>("op_3259_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_3259_cast = slice_by_index(begin = var_3259_begin_0, end = var_3259_end_0, end_mask = var_3259_end_mask_0, x = q_9_cast)[name = tensor<string, []>("op_3259_cast")];
            tensor<int32, [4]> var_3263_begin_0 = const()[name = tensor<string, []>("op_3263_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_3263_end_0 = const()[name = tensor<string, []>("op_3263_end_0"), val = tensor<int32, [4]>([2, 560, 1, 1024])];
            tensor<bool, [4]> var_3263_end_mask_0 = const()[name = tensor<string, []>("op_3263_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_3263_cast = slice_by_index(begin = var_3263_begin_0, end = var_3263_end_0, end_mask = var_3263_end_mask_0, x = q_9_cast)[name = tensor<string, []>("op_3263_cast")];
            tensor<int32, [4]> var_3267_begin_0 = const()[name = tensor<string, []>("op_3267_begin_0"), val = tensor<int32, [4]>([0, 560, 0, 0])];
            tensor<int32, [4]> var_3267_end_0 = const()[name = tensor<string, []>("op_3267_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_3267_end_mask_0 = const()[name = tensor<string, []>("op_3267_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_3267_cast = slice_by_index(begin = var_3267_begin_0, end = var_3267_end_0, end_mask = var_3267_end_mask_0, x = q_9_cast)[name = tensor<string, []>("op_3267_cast")];
            tensor<int32, [4]> var_3270_begin_0 = const()[name = tensor<string, []>("op_3270_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3270_end_0 = const()[name = tensor<string, []>("op_3270_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_3270_end_mask_0 = const()[name = tensor<string, []>("op_3270_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3270_cast = slice_by_index(begin = var_3270_begin_0, end = var_3270_end_0, end_mask = var_3270_end_mask_0, x = var_3239_cast)[name = tensor<string, []>("op_3270_cast")];
            tensor<int32, [4]> var_3271_begin_0 = const()[name = tensor<string, []>("op_3271_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3271_end_0 = const()[name = tensor<string, []>("op_3271_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_3271_end_mask_0 = const()[name = tensor<string, []>("op_3271_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3271_cast = slice_by_index(begin = var_3271_begin_0, end = var_3271_end_0, end_mask = var_3271_end_mask_0, x = var_3239_cast)[name = tensor<string, []>("op_3271_cast")];
            tensor<int32, [4]> var_3272_begin_0 = const()[name = tensor<string, []>("op_3272_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3272_end_0 = const()[name = tensor<string, []>("op_3272_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_3272_end_mask_0 = const()[name = tensor<string, []>("op_3272_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3272_cast = slice_by_index(begin = var_3272_begin_0, end = var_3272_end_0, end_mask = var_3272_end_mask_0, x = var_3243_cast)[name = tensor<string, []>("op_3272_cast")];
            tensor<int32, [4]> var_3273_begin_0 = const()[name = tensor<string, []>("op_3273_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3273_end_0 = const()[name = tensor<string, []>("op_3273_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_3273_end_mask_0 = const()[name = tensor<string, []>("op_3273_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3273_cast = slice_by_index(begin = var_3273_begin_0, end = var_3273_end_0, end_mask = var_3273_end_mask_0, x = var_3243_cast)[name = tensor<string, []>("op_3273_cast")];
            tensor<int32, [4]> var_3274_begin_0 = const()[name = tensor<string, []>("op_3274_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3274_end_0 = const()[name = tensor<string, []>("op_3274_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_3274_end_mask_0 = const()[name = tensor<string, []>("op_3274_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3274_cast = slice_by_index(begin = var_3274_begin_0, end = var_3274_end_0, end_mask = var_3274_end_mask_0, x = var_3247_cast)[name = tensor<string, []>("op_3274_cast")];
            tensor<int32, [4]> var_3275_begin_0 = const()[name = tensor<string, []>("op_3275_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3275_end_0 = const()[name = tensor<string, []>("op_3275_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_3275_end_mask_0 = const()[name = tensor<string, []>("op_3275_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3275_cast = slice_by_index(begin = var_3275_begin_0, end = var_3275_end_0, end_mask = var_3275_end_mask_0, x = var_3247_cast)[name = tensor<string, []>("op_3275_cast")];
            tensor<int32, [4]> var_3276_begin_0 = const()[name = tensor<string, []>("op_3276_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3276_end_0 = const()[name = tensor<string, []>("op_3276_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_3276_end_mask_0 = const()[name = tensor<string, []>("op_3276_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3276_cast = slice_by_index(begin = var_3276_begin_0, end = var_3276_end_0, end_mask = var_3276_end_mask_0, x = var_3251_cast)[name = tensor<string, []>("op_3276_cast")];
            tensor<int32, [4]> var_3277_begin_0 = const()[name = tensor<string, []>("op_3277_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3277_end_0 = const()[name = tensor<string, []>("op_3277_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_3277_end_mask_0 = const()[name = tensor<string, []>("op_3277_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3277_cast = slice_by_index(begin = var_3277_begin_0, end = var_3277_end_0, end_mask = var_3277_end_mask_0, x = var_3251_cast)[name = tensor<string, []>("op_3277_cast")];
            tensor<int32, [4]> var_3278_begin_0 = const()[name = tensor<string, []>("op_3278_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3278_end_0 = const()[name = tensor<string, []>("op_3278_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_3278_end_mask_0 = const()[name = tensor<string, []>("op_3278_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3278_cast = slice_by_index(begin = var_3278_begin_0, end = var_3278_end_0, end_mask = var_3278_end_mask_0, x = var_3255_cast)[name = tensor<string, []>("op_3278_cast")];
            tensor<int32, [4]> var_3279_begin_0 = const()[name = tensor<string, []>("op_3279_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3279_end_0 = const()[name = tensor<string, []>("op_3279_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_3279_end_mask_0 = const()[name = tensor<string, []>("op_3279_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3279_cast = slice_by_index(begin = var_3279_begin_0, end = var_3279_end_0, end_mask = var_3279_end_mask_0, x = var_3255_cast)[name = tensor<string, []>("op_3279_cast")];
            tensor<int32, [4]> var_3280_begin_0 = const()[name = tensor<string, []>("op_3280_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3280_end_0 = const()[name = tensor<string, []>("op_3280_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_3280_end_mask_0 = const()[name = tensor<string, []>("op_3280_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3280_cast = slice_by_index(begin = var_3280_begin_0, end = var_3280_end_0, end_mask = var_3280_end_mask_0, x = var_3259_cast)[name = tensor<string, []>("op_3280_cast")];
            tensor<int32, [4]> var_3281_begin_0 = const()[name = tensor<string, []>("op_3281_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3281_end_0 = const()[name = tensor<string, []>("op_3281_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_3281_end_mask_0 = const()[name = tensor<string, []>("op_3281_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3281_cast = slice_by_index(begin = var_3281_begin_0, end = var_3281_end_0, end_mask = var_3281_end_mask_0, x = var_3259_cast)[name = tensor<string, []>("op_3281_cast")];
            tensor<int32, [4]> var_3282_begin_0 = const()[name = tensor<string, []>("op_3282_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3282_end_0 = const()[name = tensor<string, []>("op_3282_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_3282_end_mask_0 = const()[name = tensor<string, []>("op_3282_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3282_cast = slice_by_index(begin = var_3282_begin_0, end = var_3282_end_0, end_mask = var_3282_end_mask_0, x = var_3263_cast)[name = tensor<string, []>("op_3282_cast")];
            tensor<int32, [4]> var_3283_begin_0 = const()[name = tensor<string, []>("op_3283_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3283_end_0 = const()[name = tensor<string, []>("op_3283_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_3283_end_mask_0 = const()[name = tensor<string, []>("op_3283_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3283_cast = slice_by_index(begin = var_3283_begin_0, end = var_3283_end_0, end_mask = var_3283_end_mask_0, x = var_3263_cast)[name = tensor<string, []>("op_3283_cast")];
            tensor<int32, [4]> var_3284_begin_0 = const()[name = tensor<string, []>("op_3284_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3284_end_0 = const()[name = tensor<string, []>("op_3284_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_3284_end_mask_0 = const()[name = tensor<string, []>("op_3284_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3284_cast = slice_by_index(begin = var_3284_begin_0, end = var_3284_end_0, end_mask = var_3284_end_mask_0, x = var_3267_cast)[name = tensor<string, []>("op_3284_cast")];
            tensor<int32, [4]> var_3285_begin_0 = const()[name = tensor<string, []>("op_3285_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3285_end_0 = const()[name = tensor<string, []>("op_3285_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_3285_end_mask_0 = const()[name = tensor<string, []>("op_3285_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3285_cast = slice_by_index(begin = var_3285_begin_0, end = var_3285_end_0, end_mask = var_3285_end_mask_0, x = var_3267_cast)[name = tensor<string, []>("op_3285_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_3290_begin_0 = const()[name = tensor<string, []>("op_3290_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3290_end_0 = const()[name = tensor<string, []>("op_3290_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 80])];
            tensor<bool, [4]> var_3290_end_mask_0 = const()[name = tensor<string, []>("op_3290_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)[name = tensor<string, []>("transpose_27")];
            tensor<fp16, [2, 1024, 1, 80]> var_3290_cast = slice_by_index(begin = var_3290_begin_0, end = var_3290_end_0, end_mask = var_3290_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_3290_cast")];
            tensor<int32, [4]> var_3294_begin_0 = const()[name = tensor<string, []>("op_3294_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 80])];
            tensor<int32, [4]> var_3294_end_0 = const()[name = tensor<string, []>("op_3294_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 160])];
            tensor<bool, [4]> var_3294_end_mask_0 = const()[name = tensor<string, []>("op_3294_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_3294_cast = slice_by_index(begin = var_3294_begin_0, end = var_3294_end_0, end_mask = var_3294_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_3294_cast")];
            tensor<int32, [4]> var_3298_begin_0 = const()[name = tensor<string, []>("op_3298_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_3298_end_0 = const()[name = tensor<string, []>("op_3298_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 240])];
            tensor<bool, [4]> var_3298_end_mask_0 = const()[name = tensor<string, []>("op_3298_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_3298_cast = slice_by_index(begin = var_3298_begin_0, end = var_3298_end_0, end_mask = var_3298_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_3298_cast")];
            tensor<int32, [4]> var_3302_begin_0 = const()[name = tensor<string, []>("op_3302_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 240])];
            tensor<int32, [4]> var_3302_end_0 = const()[name = tensor<string, []>("op_3302_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_3302_end_mask_0 = const()[name = tensor<string, []>("op_3302_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_3302_cast = slice_by_index(begin = var_3302_begin_0, end = var_3302_end_0, end_mask = var_3302_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_3302_cast")];
            tensor<int32, [4]> var_3306_begin_0 = const()[name = tensor<string, []>("op_3306_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_3306_end_0 = const()[name = tensor<string, []>("op_3306_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 400])];
            tensor<bool, [4]> var_3306_end_mask_0 = const()[name = tensor<string, []>("op_3306_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_3306_cast = slice_by_index(begin = var_3306_begin_0, end = var_3306_end_0, end_mask = var_3306_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_3306_cast")];
            tensor<int32, [4]> var_3310_begin_0 = const()[name = tensor<string, []>("op_3310_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 400])];
            tensor<int32, [4]> var_3310_end_0 = const()[name = tensor<string, []>("op_3310_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 480])];
            tensor<bool, [4]> var_3310_end_mask_0 = const()[name = tensor<string, []>("op_3310_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_3310_cast = slice_by_index(begin = var_3310_begin_0, end = var_3310_end_0, end_mask = var_3310_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_3310_cast")];
            tensor<int32, [4]> var_3314_begin_0 = const()[name = tensor<string, []>("op_3314_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 480])];
            tensor<int32, [4]> var_3314_end_0 = const()[name = tensor<string, []>("op_3314_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 560])];
            tensor<bool, [4]> var_3314_end_mask_0 = const()[name = tensor<string, []>("op_3314_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_3314_cast = slice_by_index(begin = var_3314_begin_0, end = var_3314_end_0, end_mask = var_3314_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_3314_cast")];
            tensor<int32, [4]> var_3318_begin_0 = const()[name = tensor<string, []>("op_3318_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 560])];
            tensor<int32, [4]> var_3318_end_0 = const()[name = tensor<string, []>("op_3318_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_3318_end_mask_0 = const()[name = tensor<string, []>("op_3318_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_3318_cast = slice_by_index(begin = var_3318_begin_0, end = var_3318_end_0, end_mask = var_3318_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_3318_cast")];
            tensor<int32, [4]> var_3320_begin_0 = const()[name = tensor<string, []>("op_3320_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3320_end_0 = const()[name = tensor<string, []>("op_3320_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_3320_end_mask_0 = const()[name = tensor<string, []>("op_3320_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_3320_cast = slice_by_index(begin = var_3320_begin_0, end = var_3320_end_0, end_mask = var_3320_end_mask_0, x = v_9_cast)[name = tensor<string, []>("op_3320_cast")];
            tensor<int32, [4]> var_3324_begin_0 = const()[name = tensor<string, []>("op_3324_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_3324_end_0 = const()[name = tensor<string, []>("op_3324_end_0"), val = tensor<int32, [4]>([2, 160, 1, 1024])];
            tensor<bool, [4]> var_3324_end_mask_0 = const()[name = tensor<string, []>("op_3324_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_3324_cast = slice_by_index(begin = var_3324_begin_0, end = var_3324_end_0, end_mask = var_3324_end_mask_0, x = v_9_cast)[name = tensor<string, []>("op_3324_cast")];
            tensor<int32, [4]> var_3328_begin_0 = const()[name = tensor<string, []>("op_3328_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_3328_end_0 = const()[name = tensor<string, []>("op_3328_end_0"), val = tensor<int32, [4]>([2, 240, 1, 1024])];
            tensor<bool, [4]> var_3328_end_mask_0 = const()[name = tensor<string, []>("op_3328_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_3328_cast = slice_by_index(begin = var_3328_begin_0, end = var_3328_end_0, end_mask = var_3328_end_mask_0, x = v_9_cast)[name = tensor<string, []>("op_3328_cast")];
            tensor<int32, [4]> var_3332_begin_0 = const()[name = tensor<string, []>("op_3332_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_3332_end_0 = const()[name = tensor<string, []>("op_3332_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_3332_end_mask_0 = const()[name = tensor<string, []>("op_3332_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_3332_cast = slice_by_index(begin = var_3332_begin_0, end = var_3332_end_0, end_mask = var_3332_end_mask_0, x = v_9_cast)[name = tensor<string, []>("op_3332_cast")];
            tensor<int32, [4]> var_3336_begin_0 = const()[name = tensor<string, []>("op_3336_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_3336_end_0 = const()[name = tensor<string, []>("op_3336_end_0"), val = tensor<int32, [4]>([2, 400, 1, 1024])];
            tensor<bool, [4]> var_3336_end_mask_0 = const()[name = tensor<string, []>("op_3336_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_3336_cast = slice_by_index(begin = var_3336_begin_0, end = var_3336_end_0, end_mask = var_3336_end_mask_0, x = v_9_cast)[name = tensor<string, []>("op_3336_cast")];
            tensor<int32, [4]> var_3340_begin_0 = const()[name = tensor<string, []>("op_3340_begin_0"), val = tensor<int32, [4]>([0, 400, 0, 0])];
            tensor<int32, [4]> var_3340_end_0 = const()[name = tensor<string, []>("op_3340_end_0"), val = tensor<int32, [4]>([2, 480, 1, 1024])];
            tensor<bool, [4]> var_3340_end_mask_0 = const()[name = tensor<string, []>("op_3340_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_3340_cast = slice_by_index(begin = var_3340_begin_0, end = var_3340_end_0, end_mask = var_3340_end_mask_0, x = v_9_cast)[name = tensor<string, []>("op_3340_cast")];
            tensor<int32, [4]> var_3344_begin_0 = const()[name = tensor<string, []>("op_3344_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_3344_end_0 = const()[name = tensor<string, []>("op_3344_end_0"), val = tensor<int32, [4]>([2, 560, 1, 1024])];
            tensor<bool, [4]> var_3344_end_mask_0 = const()[name = tensor<string, []>("op_3344_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_3344_cast = slice_by_index(begin = var_3344_begin_0, end = var_3344_end_0, end_mask = var_3344_end_mask_0, x = v_9_cast)[name = tensor<string, []>("op_3344_cast")];
            tensor<int32, [4]> var_3348_begin_0 = const()[name = tensor<string, []>("op_3348_begin_0"), val = tensor<int32, [4]>([0, 560, 0, 0])];
            tensor<int32, [4]> var_3348_end_0 = const()[name = tensor<string, []>("op_3348_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_3348_end_mask_0 = const()[name = tensor<string, []>("op_3348_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_3348_cast = slice_by_index(begin = var_3348_begin_0, end = var_3348_end_0, end_mask = var_3348_end_mask_0, x = v_9_cast)[name = tensor<string, []>("op_3348_cast")];
            tensor<string, []> var_3352_equation_0 = const()[name = tensor<string, []>("op_3352_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3352_cast = einsum(equation = var_3352_equation_0, values = (var_3290_cast, var_3270_cast))[name = tensor<string, []>("op_3352_cast")];
            tensor<fp16, []> var_3353_to_fp16 = const()[name = tensor<string, []>("op_3353_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_513_cast = mul(x = var_3352_cast, y = var_3353_to_fp16)[name = tensor<string, []>("aw_chunk_513_cast")];
            tensor<string, []> var_3356_equation_0 = const()[name = tensor<string, []>("op_3356_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3356_cast = einsum(equation = var_3356_equation_0, values = (var_3290_cast, var_3271_cast))[name = tensor<string, []>("op_3356_cast")];
            tensor<fp16, []> var_3357_to_fp16 = const()[name = tensor<string, []>("op_3357_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_515_cast = mul(x = var_3356_cast, y = var_3357_to_fp16)[name = tensor<string, []>("aw_chunk_515_cast")];
            tensor<string, []> var_3360_equation_0 = const()[name = tensor<string, []>("op_3360_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3360_cast = einsum(equation = var_3360_equation_0, values = (var_3294_cast, var_3272_cast))[name = tensor<string, []>("op_3360_cast")];
            tensor<fp16, []> var_3361_to_fp16 = const()[name = tensor<string, []>("op_3361_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_517_cast = mul(x = var_3360_cast, y = var_3361_to_fp16)[name = tensor<string, []>("aw_chunk_517_cast")];
            tensor<string, []> var_3364_equation_0 = const()[name = tensor<string, []>("op_3364_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3364_cast = einsum(equation = var_3364_equation_0, values = (var_3294_cast, var_3273_cast))[name = tensor<string, []>("op_3364_cast")];
            tensor<fp16, []> var_3365_to_fp16 = const()[name = tensor<string, []>("op_3365_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_519_cast = mul(x = var_3364_cast, y = var_3365_to_fp16)[name = tensor<string, []>("aw_chunk_519_cast")];
            tensor<string, []> var_3368_equation_0 = const()[name = tensor<string, []>("op_3368_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3368_cast = einsum(equation = var_3368_equation_0, values = (var_3298_cast, var_3274_cast))[name = tensor<string, []>("op_3368_cast")];
            tensor<fp16, []> var_3369_to_fp16 = const()[name = tensor<string, []>("op_3369_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_521_cast = mul(x = var_3368_cast, y = var_3369_to_fp16)[name = tensor<string, []>("aw_chunk_521_cast")];
            tensor<string, []> var_3372_equation_0 = const()[name = tensor<string, []>("op_3372_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3372_cast = einsum(equation = var_3372_equation_0, values = (var_3298_cast, var_3275_cast))[name = tensor<string, []>("op_3372_cast")];
            tensor<fp16, []> var_3373_to_fp16 = const()[name = tensor<string, []>("op_3373_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_523_cast = mul(x = var_3372_cast, y = var_3373_to_fp16)[name = tensor<string, []>("aw_chunk_523_cast")];
            tensor<string, []> var_3376_equation_0 = const()[name = tensor<string, []>("op_3376_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3376_cast = einsum(equation = var_3376_equation_0, values = (var_3302_cast, var_3276_cast))[name = tensor<string, []>("op_3376_cast")];
            tensor<fp16, []> var_3377_to_fp16 = const()[name = tensor<string, []>("op_3377_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_525_cast = mul(x = var_3376_cast, y = var_3377_to_fp16)[name = tensor<string, []>("aw_chunk_525_cast")];
            tensor<string, []> var_3380_equation_0 = const()[name = tensor<string, []>("op_3380_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3380_cast = einsum(equation = var_3380_equation_0, values = (var_3302_cast, var_3277_cast))[name = tensor<string, []>("op_3380_cast")];
            tensor<fp16, []> var_3381_to_fp16 = const()[name = tensor<string, []>("op_3381_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_527_cast = mul(x = var_3380_cast, y = var_3381_to_fp16)[name = tensor<string, []>("aw_chunk_527_cast")];
            tensor<string, []> var_3384_equation_0 = const()[name = tensor<string, []>("op_3384_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3384_cast = einsum(equation = var_3384_equation_0, values = (var_3306_cast, var_3278_cast))[name = tensor<string, []>("op_3384_cast")];
            tensor<fp16, []> var_3385_to_fp16 = const()[name = tensor<string, []>("op_3385_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_529_cast = mul(x = var_3384_cast, y = var_3385_to_fp16)[name = tensor<string, []>("aw_chunk_529_cast")];
            tensor<string, []> var_3388_equation_0 = const()[name = tensor<string, []>("op_3388_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3388_cast = einsum(equation = var_3388_equation_0, values = (var_3306_cast, var_3279_cast))[name = tensor<string, []>("op_3388_cast")];
            tensor<fp16, []> var_3389_to_fp16 = const()[name = tensor<string, []>("op_3389_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_531_cast = mul(x = var_3388_cast, y = var_3389_to_fp16)[name = tensor<string, []>("aw_chunk_531_cast")];
            tensor<string, []> var_3392_equation_0 = const()[name = tensor<string, []>("op_3392_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3392_cast = einsum(equation = var_3392_equation_0, values = (var_3310_cast, var_3280_cast))[name = tensor<string, []>("op_3392_cast")];
            tensor<fp16, []> var_3393_to_fp16 = const()[name = tensor<string, []>("op_3393_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_533_cast = mul(x = var_3392_cast, y = var_3393_to_fp16)[name = tensor<string, []>("aw_chunk_533_cast")];
            tensor<string, []> var_3396_equation_0 = const()[name = tensor<string, []>("op_3396_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3396_cast = einsum(equation = var_3396_equation_0, values = (var_3310_cast, var_3281_cast))[name = tensor<string, []>("op_3396_cast")];
            tensor<fp16, []> var_3397_to_fp16 = const()[name = tensor<string, []>("op_3397_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_535_cast = mul(x = var_3396_cast, y = var_3397_to_fp16)[name = tensor<string, []>("aw_chunk_535_cast")];
            tensor<string, []> var_3400_equation_0 = const()[name = tensor<string, []>("op_3400_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3400_cast = einsum(equation = var_3400_equation_0, values = (var_3314_cast, var_3282_cast))[name = tensor<string, []>("op_3400_cast")];
            tensor<fp16, []> var_3401_to_fp16 = const()[name = tensor<string, []>("op_3401_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_537_cast = mul(x = var_3400_cast, y = var_3401_to_fp16)[name = tensor<string, []>("aw_chunk_537_cast")];
            tensor<string, []> var_3404_equation_0 = const()[name = tensor<string, []>("op_3404_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3404_cast = einsum(equation = var_3404_equation_0, values = (var_3314_cast, var_3283_cast))[name = tensor<string, []>("op_3404_cast")];
            tensor<fp16, []> var_3405_to_fp16 = const()[name = tensor<string, []>("op_3405_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_539_cast = mul(x = var_3404_cast, y = var_3405_to_fp16)[name = tensor<string, []>("aw_chunk_539_cast")];
            tensor<string, []> var_3408_equation_0 = const()[name = tensor<string, []>("op_3408_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3408_cast = einsum(equation = var_3408_equation_0, values = (var_3318_cast, var_3284_cast))[name = tensor<string, []>("op_3408_cast")];
            tensor<fp16, []> var_3409_to_fp16 = const()[name = tensor<string, []>("op_3409_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_541_cast = mul(x = var_3408_cast, y = var_3409_to_fp16)[name = tensor<string, []>("aw_chunk_541_cast")];
            tensor<string, []> var_3412_equation_0 = const()[name = tensor<string, []>("op_3412_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3412_cast = einsum(equation = var_3412_equation_0, values = (var_3318_cast, var_3285_cast))[name = tensor<string, []>("op_3412_cast")];
            tensor<fp16, []> var_3413_to_fp16 = const()[name = tensor<string, []>("op_3413_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_543_cast = mul(x = var_3412_cast, y = var_3413_to_fp16)[name = tensor<string, []>("aw_chunk_543_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3415_cast = softmax(axis = var_3111, x = aw_chunk_513_cast)[name = tensor<string, []>("op_3415_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3416_cast = softmax(axis = var_3111, x = aw_chunk_515_cast)[name = tensor<string, []>("op_3416_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3417_cast = softmax(axis = var_3111, x = aw_chunk_517_cast)[name = tensor<string, []>("op_3417_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3418_cast = softmax(axis = var_3111, x = aw_chunk_519_cast)[name = tensor<string, []>("op_3418_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3419_cast = softmax(axis = var_3111, x = aw_chunk_521_cast)[name = tensor<string, []>("op_3419_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3420_cast = softmax(axis = var_3111, x = aw_chunk_523_cast)[name = tensor<string, []>("op_3420_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3421_cast = softmax(axis = var_3111, x = aw_chunk_525_cast)[name = tensor<string, []>("op_3421_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3422_cast = softmax(axis = var_3111, x = aw_chunk_527_cast)[name = tensor<string, []>("op_3422_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3423_cast = softmax(axis = var_3111, x = aw_chunk_529_cast)[name = tensor<string, []>("op_3423_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3424_cast = softmax(axis = var_3111, x = aw_chunk_531_cast)[name = tensor<string, []>("op_3424_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3425_cast = softmax(axis = var_3111, x = aw_chunk_533_cast)[name = tensor<string, []>("op_3425_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3426_cast = softmax(axis = var_3111, x = aw_chunk_535_cast)[name = tensor<string, []>("op_3426_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3427_cast = softmax(axis = var_3111, x = aw_chunk_537_cast)[name = tensor<string, []>("op_3427_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3428_cast = softmax(axis = var_3111, x = aw_chunk_539_cast)[name = tensor<string, []>("op_3428_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3429_cast = softmax(axis = var_3111, x = aw_chunk_541_cast)[name = tensor<string, []>("op_3429_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3430_cast = softmax(axis = var_3111, x = aw_chunk_543_cast)[name = tensor<string, []>("op_3430_cast")];
            tensor<string, []> var_3432_equation_0 = const()[name = tensor<string, []>("op_3432_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3432_cast = einsum(equation = var_3432_equation_0, values = (var_3320_cast, var_3415_cast))[name = tensor<string, []>("op_3432_cast")];
            tensor<string, []> var_3434_equation_0 = const()[name = tensor<string, []>("op_3434_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3434_cast = einsum(equation = var_3434_equation_0, values = (var_3320_cast, var_3416_cast))[name = tensor<string, []>("op_3434_cast")];
            tensor<string, []> var_3436_equation_0 = const()[name = tensor<string, []>("op_3436_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3436_cast = einsum(equation = var_3436_equation_0, values = (var_3324_cast, var_3417_cast))[name = tensor<string, []>("op_3436_cast")];
            tensor<string, []> var_3438_equation_0 = const()[name = tensor<string, []>("op_3438_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3438_cast = einsum(equation = var_3438_equation_0, values = (var_3324_cast, var_3418_cast))[name = tensor<string, []>("op_3438_cast")];
            tensor<string, []> var_3440_equation_0 = const()[name = tensor<string, []>("op_3440_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3440_cast = einsum(equation = var_3440_equation_0, values = (var_3328_cast, var_3419_cast))[name = tensor<string, []>("op_3440_cast")];
            tensor<string, []> var_3442_equation_0 = const()[name = tensor<string, []>("op_3442_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3442_cast = einsum(equation = var_3442_equation_0, values = (var_3328_cast, var_3420_cast))[name = tensor<string, []>("op_3442_cast")];
            tensor<string, []> var_3444_equation_0 = const()[name = tensor<string, []>("op_3444_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3444_cast = einsum(equation = var_3444_equation_0, values = (var_3332_cast, var_3421_cast))[name = tensor<string, []>("op_3444_cast")];
            tensor<string, []> var_3446_equation_0 = const()[name = tensor<string, []>("op_3446_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3446_cast = einsum(equation = var_3446_equation_0, values = (var_3332_cast, var_3422_cast))[name = tensor<string, []>("op_3446_cast")];
            tensor<string, []> var_3448_equation_0 = const()[name = tensor<string, []>("op_3448_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3448_cast = einsum(equation = var_3448_equation_0, values = (var_3336_cast, var_3423_cast))[name = tensor<string, []>("op_3448_cast")];
            tensor<string, []> var_3450_equation_0 = const()[name = tensor<string, []>("op_3450_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3450_cast = einsum(equation = var_3450_equation_0, values = (var_3336_cast, var_3424_cast))[name = tensor<string, []>("op_3450_cast")];
            tensor<string, []> var_3452_equation_0 = const()[name = tensor<string, []>("op_3452_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3452_cast = einsum(equation = var_3452_equation_0, values = (var_3340_cast, var_3425_cast))[name = tensor<string, []>("op_3452_cast")];
            tensor<string, []> var_3454_equation_0 = const()[name = tensor<string, []>("op_3454_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3454_cast = einsum(equation = var_3454_equation_0, values = (var_3340_cast, var_3426_cast))[name = tensor<string, []>("op_3454_cast")];
            tensor<string, []> var_3456_equation_0 = const()[name = tensor<string, []>("op_3456_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3456_cast = einsum(equation = var_3456_equation_0, values = (var_3344_cast, var_3427_cast))[name = tensor<string, []>("op_3456_cast")];
            tensor<string, []> var_3458_equation_0 = const()[name = tensor<string, []>("op_3458_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3458_cast = einsum(equation = var_3458_equation_0, values = (var_3344_cast, var_3428_cast))[name = tensor<string, []>("op_3458_cast")];
            tensor<string, []> var_3460_equation_0 = const()[name = tensor<string, []>("op_3460_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3460_cast = einsum(equation = var_3460_equation_0, values = (var_3348_cast, var_3429_cast))[name = tensor<string, []>("op_3460_cast")];
            tensor<string, []> var_3462_equation_0 = const()[name = tensor<string, []>("op_3462_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3462_cast = einsum(equation = var_3462_equation_0, values = (var_3348_cast, var_3430_cast))[name = tensor<string, []>("op_3462_cast")];
            tensor<bool, []> var_3464_interleave_0 = const()[name = tensor<string, []>("op_3464_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_3464_cast = concat(axis = var_3089, interleave = var_3464_interleave_0, values = (var_3432_cast, var_3434_cast))[name = tensor<string, []>("op_3464_cast")];
            tensor<bool, []> var_3466_interleave_0 = const()[name = tensor<string, []>("op_3466_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_3466_cast = concat(axis = var_3089, interleave = var_3466_interleave_0, values = (var_3436_cast, var_3438_cast))[name = tensor<string, []>("op_3466_cast")];
            tensor<bool, []> var_3468_interleave_0 = const()[name = tensor<string, []>("op_3468_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_3468_cast = concat(axis = var_3089, interleave = var_3468_interleave_0, values = (var_3440_cast, var_3442_cast))[name = tensor<string, []>("op_3468_cast")];
            tensor<bool, []> var_3470_interleave_0 = const()[name = tensor<string, []>("op_3470_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_3470_cast = concat(axis = var_3089, interleave = var_3470_interleave_0, values = (var_3444_cast, var_3446_cast))[name = tensor<string, []>("op_3470_cast")];
            tensor<bool, []> var_3472_interleave_0 = const()[name = tensor<string, []>("op_3472_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_3472_cast = concat(axis = var_3089, interleave = var_3472_interleave_0, values = (var_3448_cast, var_3450_cast))[name = tensor<string, []>("op_3472_cast")];
            tensor<bool, []> var_3474_interleave_0 = const()[name = tensor<string, []>("op_3474_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_3474_cast = concat(axis = var_3089, interleave = var_3474_interleave_0, values = (var_3452_cast, var_3454_cast))[name = tensor<string, []>("op_3474_cast")];
            tensor<bool, []> var_3476_interleave_0 = const()[name = tensor<string, []>("op_3476_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_3476_cast = concat(axis = var_3089, interleave = var_3476_interleave_0, values = (var_3456_cast, var_3458_cast))[name = tensor<string, []>("op_3476_cast")];
            tensor<bool, []> var_3478_interleave_0 = const()[name = tensor<string, []>("op_3478_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_3478_cast = concat(axis = var_3089, interleave = var_3478_interleave_0, values = (var_3460_cast, var_3462_cast))[name = tensor<string, []>("op_3478_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_3111, interleave = input_79_interleave_0, values = (var_3464_cast, var_3466_cast, var_3468_cast, var_3470_cast, var_3472_cast, var_3474_cast, var_3476_cast, var_3478_cast))[name = tensor<string, []>("input_79_cast")];
            tensor<int32, [2]> var_3484 = const()[name = tensor<string, []>("op_3484"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3486 = const()[name = tensor<string, []>("op_3486"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_3488_pad_type_0 = const()[name = tensor<string, []>("op_3488_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_3488_pad_0 = const()[name = tensor<string, []>("op_3488_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15638976))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15946240))), name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            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, []>(15946432)))];
            tensor<fp16, [2, 640, 1, 1024]> var_3488_cast = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_3486, groups = var_3111, pad = var_3488_pad_0, pad_type = var_3488_pad_type_0, strides = var_3484, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_79_cast)[name = tensor<string, []>("op_3488_cast")];
            tensor<fp16, [2, 640, 1, 1024]> inputs_15_cast = add(x = var_3488_cast, y = inputs_13_cast)[name = tensor<string, []>("inputs_15_cast")];
            tensor<int32, [1]> var_3492 = const()[name = tensor<string, []>("op_3492"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_15_cast = reduce_mean(axes = var_3492, keep_dims = var_3106, x = inputs_15_cast)[name = tensor<string, []>("channels_mean_15_cast")];
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_15_cast = sub(x = inputs_15_cast, y = channels_mean_15_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_15_cast")];
            tensor<int32, [1]> var_3496 = const()[name = tensor<string, []>("op_3496"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_3497_cast = reduce_mean(axes = var_3496, keep_dims = var_3106, x = zero_mean_sq_15_cast)[name = tensor<string, []>("op_3497_cast")];
            tensor<fp16, []> var_3498_to_fp16 = const()[name = tensor<string, []>("op_3498_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_3499_cast = add(x = var_3497_cast, y = var_3498_to_fp16)[name = tensor<string, []>("op_3499_cast")];
            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_3499_cast)[name = tensor<string, []>("denom_15_cast")];
            tensor<fp16, [2, 640, 1, 1024]> out_15_cast = mul(x = zero_mean_15_cast, y = denom_15_cast)[name = tensor<string, []>("out_15_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_3503_to_fp16 = const()[name = tensor<string, []>("op_3503_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15947776)))];
            tensor<fp16, [2, 640, 1, 1024]> var_3504_cast = add(x = out_15_cast, y = var_3503_to_fp16)[name = tensor<string, []>("op_3504_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_3506_to_fp16 = const()[name = tensor<string, []>("op_3506_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15949120)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_47_cast = mul(x = var_3504_cast, y = var_3506_to_fp16)[name = tensor<string, []>("hidden_states_47_cast")];
            tensor<int32, [2]> var_3513 = const()[name = tensor<string, []>("op_3513"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3515 = const()[name = tensor<string, []>("op_3515"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15950464))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16257728))), name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 1024]> q_11_cast = conv(dilations = var_3515, groups = var_3111, pad = q_11_pad_0, pad_type = q_11_pad_type_0, strides = var_3513, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_47_cast)[name = tensor<string, []>("q_11_cast")];
            tensor<int32, [2]> var_3519 = const()[name = tensor<string, []>("op_3519"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3521 = const()[name = tensor<string, []>("op_3521"), 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, 768, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [368640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16257920))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16626624))), name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 768, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> k_21_cast = conv(dilations = var_3521, groups = var_3111, pad = k_21_pad_0, pad_type = k_21_pad_type_0, strides = var_3519, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_21_cast")];
            tensor<int32, [2]> var_3525 = const()[name = tensor<string, []>("op_3525"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3527 = const()[name = tensor<string, []>("op_3527"), 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, 768, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [368640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16626816))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16995520))), name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 768, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> v_11_cast = conv(dilations = var_3527, groups = var_3111, pad = v_11_pad_0, pad_type = v_11_pad_type_0, strides = var_3525, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_11_cast")];
            tensor<int32, [4]> var_3531_begin_0 = const()[name = tensor<string, []>("op_3531_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3531_end_0 = const()[name = tensor<string, []>("op_3531_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_3531_end_mask_0 = const()[name = tensor<string, []>("op_3531_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_3531_cast = slice_by_index(begin = var_3531_begin_0, end = var_3531_end_0, end_mask = var_3531_end_mask_0, x = q_11_cast)[name = tensor<string, []>("op_3531_cast")];
            tensor<int32, [4]> var_3535_begin_0 = const()[name = tensor<string, []>("op_3535_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_3535_end_0 = const()[name = tensor<string, []>("op_3535_end_0"), val = tensor<int32, [4]>([2, 160, 1, 1024])];
            tensor<bool, [4]> var_3535_end_mask_0 = const()[name = tensor<string, []>("op_3535_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_3535_cast = slice_by_index(begin = var_3535_begin_0, end = var_3535_end_0, end_mask = var_3535_end_mask_0, x = q_11_cast)[name = tensor<string, []>("op_3535_cast")];
            tensor<int32, [4]> var_3539_begin_0 = const()[name = tensor<string, []>("op_3539_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_3539_end_0 = const()[name = tensor<string, []>("op_3539_end_0"), val = tensor<int32, [4]>([2, 240, 1, 1024])];
            tensor<bool, [4]> var_3539_end_mask_0 = const()[name = tensor<string, []>("op_3539_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_3539_cast = slice_by_index(begin = var_3539_begin_0, end = var_3539_end_0, end_mask = var_3539_end_mask_0, x = q_11_cast)[name = tensor<string, []>("op_3539_cast")];
            tensor<int32, [4]> var_3543_begin_0 = const()[name = tensor<string, []>("op_3543_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_3543_end_0 = const()[name = tensor<string, []>("op_3543_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_3543_end_mask_0 = const()[name = tensor<string, []>("op_3543_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_3543_cast = slice_by_index(begin = var_3543_begin_0, end = var_3543_end_0, end_mask = var_3543_end_mask_0, x = q_11_cast)[name = tensor<string, []>("op_3543_cast")];
            tensor<int32, [4]> var_3547_begin_0 = const()[name = tensor<string, []>("op_3547_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_3547_end_0 = const()[name = tensor<string, []>("op_3547_end_0"), val = tensor<int32, [4]>([2, 400, 1, 1024])];
            tensor<bool, [4]> var_3547_end_mask_0 = const()[name = tensor<string, []>("op_3547_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_3547_cast = slice_by_index(begin = var_3547_begin_0, end = var_3547_end_0, end_mask = var_3547_end_mask_0, x = q_11_cast)[name = tensor<string, []>("op_3547_cast")];
            tensor<int32, [4]> var_3551_begin_0 = const()[name = tensor<string, []>("op_3551_begin_0"), val = tensor<int32, [4]>([0, 400, 0, 0])];
            tensor<int32, [4]> var_3551_end_0 = const()[name = tensor<string, []>("op_3551_end_0"), val = tensor<int32, [4]>([2, 480, 1, 1024])];
            tensor<bool, [4]> var_3551_end_mask_0 = const()[name = tensor<string, []>("op_3551_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_3551_cast = slice_by_index(begin = var_3551_begin_0, end = var_3551_end_0, end_mask = var_3551_end_mask_0, x = q_11_cast)[name = tensor<string, []>("op_3551_cast")];
            tensor<int32, [4]> var_3555_begin_0 = const()[name = tensor<string, []>("op_3555_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_3555_end_0 = const()[name = tensor<string, []>("op_3555_end_0"), val = tensor<int32, [4]>([2, 560, 1, 1024])];
            tensor<bool, [4]> var_3555_end_mask_0 = const()[name = tensor<string, []>("op_3555_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_3555_cast = slice_by_index(begin = var_3555_begin_0, end = var_3555_end_0, end_mask = var_3555_end_mask_0, x = q_11_cast)[name = tensor<string, []>("op_3555_cast")];
            tensor<int32, [4]> var_3559_begin_0 = const()[name = tensor<string, []>("op_3559_begin_0"), val = tensor<int32, [4]>([0, 560, 0, 0])];
            tensor<int32, [4]> var_3559_end_0 = const()[name = tensor<string, []>("op_3559_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_3559_end_mask_0 = const()[name = tensor<string, []>("op_3559_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_3559_cast = slice_by_index(begin = var_3559_begin_0, end = var_3559_end_0, end_mask = var_3559_end_mask_0, x = q_11_cast)[name = tensor<string, []>("op_3559_cast")];
            tensor<int32, [4]> var_3562_begin_0 = const()[name = tensor<string, []>("op_3562_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3562_end_0 = const()[name = tensor<string, []>("op_3562_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_3562_end_mask_0 = const()[name = tensor<string, []>("op_3562_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3562_cast = slice_by_index(begin = var_3562_begin_0, end = var_3562_end_0, end_mask = var_3562_end_mask_0, x = var_3531_cast)[name = tensor<string, []>("op_3562_cast")];
            tensor<int32, [4]> var_3563_begin_0 = const()[name = tensor<string, []>("op_3563_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3563_end_0 = const()[name = tensor<string, []>("op_3563_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_3563_end_mask_0 = const()[name = tensor<string, []>("op_3563_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3563_cast = slice_by_index(begin = var_3563_begin_0, end = var_3563_end_0, end_mask = var_3563_end_mask_0, x = var_3531_cast)[name = tensor<string, []>("op_3563_cast")];
            tensor<int32, [4]> var_3564_begin_0 = const()[name = tensor<string, []>("op_3564_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3564_end_0 = const()[name = tensor<string, []>("op_3564_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_3564_end_mask_0 = const()[name = tensor<string, []>("op_3564_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3564_cast = slice_by_index(begin = var_3564_begin_0, end = var_3564_end_0, end_mask = var_3564_end_mask_0, x = var_3535_cast)[name = tensor<string, []>("op_3564_cast")];
            tensor<int32, [4]> var_3565_begin_0 = const()[name = tensor<string, []>("op_3565_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3565_end_0 = const()[name = tensor<string, []>("op_3565_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_3565_end_mask_0 = const()[name = tensor<string, []>("op_3565_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3565_cast = slice_by_index(begin = var_3565_begin_0, end = var_3565_end_0, end_mask = var_3565_end_mask_0, x = var_3535_cast)[name = tensor<string, []>("op_3565_cast")];
            tensor<int32, [4]> var_3566_begin_0 = const()[name = tensor<string, []>("op_3566_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3566_end_0 = const()[name = tensor<string, []>("op_3566_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_3566_end_mask_0 = const()[name = tensor<string, []>("op_3566_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3566_cast = slice_by_index(begin = var_3566_begin_0, end = var_3566_end_0, end_mask = var_3566_end_mask_0, x = var_3539_cast)[name = tensor<string, []>("op_3566_cast")];
            tensor<int32, [4]> var_3567_begin_0 = const()[name = tensor<string, []>("op_3567_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3567_end_0 = const()[name = tensor<string, []>("op_3567_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_3567_end_mask_0 = const()[name = tensor<string, []>("op_3567_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3567_cast = slice_by_index(begin = var_3567_begin_0, end = var_3567_end_0, end_mask = var_3567_end_mask_0, x = var_3539_cast)[name = tensor<string, []>("op_3567_cast")];
            tensor<int32, [4]> var_3568_begin_0 = const()[name = tensor<string, []>("op_3568_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3568_end_0 = const()[name = tensor<string, []>("op_3568_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_3568_end_mask_0 = const()[name = tensor<string, []>("op_3568_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3568_cast = slice_by_index(begin = var_3568_begin_0, end = var_3568_end_0, end_mask = var_3568_end_mask_0, x = var_3543_cast)[name = tensor<string, []>("op_3568_cast")];
            tensor<int32, [4]> var_3569_begin_0 = const()[name = tensor<string, []>("op_3569_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3569_end_0 = const()[name = tensor<string, []>("op_3569_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_3569_end_mask_0 = const()[name = tensor<string, []>("op_3569_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3569_cast = slice_by_index(begin = var_3569_begin_0, end = var_3569_end_0, end_mask = var_3569_end_mask_0, x = var_3543_cast)[name = tensor<string, []>("op_3569_cast")];
            tensor<int32, [4]> var_3570_begin_0 = const()[name = tensor<string, []>("op_3570_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3570_end_0 = const()[name = tensor<string, []>("op_3570_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_3570_end_mask_0 = const()[name = tensor<string, []>("op_3570_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3570_cast = slice_by_index(begin = var_3570_begin_0, end = var_3570_end_0, end_mask = var_3570_end_mask_0, x = var_3547_cast)[name = tensor<string, []>("op_3570_cast")];
            tensor<int32, [4]> var_3571_begin_0 = const()[name = tensor<string, []>("op_3571_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3571_end_0 = const()[name = tensor<string, []>("op_3571_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_3571_end_mask_0 = const()[name = tensor<string, []>("op_3571_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3571_cast = slice_by_index(begin = var_3571_begin_0, end = var_3571_end_0, end_mask = var_3571_end_mask_0, x = var_3547_cast)[name = tensor<string, []>("op_3571_cast")];
            tensor<int32, [4]> var_3572_begin_0 = const()[name = tensor<string, []>("op_3572_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3572_end_0 = const()[name = tensor<string, []>("op_3572_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_3572_end_mask_0 = const()[name = tensor<string, []>("op_3572_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3572_cast = slice_by_index(begin = var_3572_begin_0, end = var_3572_end_0, end_mask = var_3572_end_mask_0, x = var_3551_cast)[name = tensor<string, []>("op_3572_cast")];
            tensor<int32, [4]> var_3573_begin_0 = const()[name = tensor<string, []>("op_3573_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3573_end_0 = const()[name = tensor<string, []>("op_3573_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_3573_end_mask_0 = const()[name = tensor<string, []>("op_3573_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3573_cast = slice_by_index(begin = var_3573_begin_0, end = var_3573_end_0, end_mask = var_3573_end_mask_0, x = var_3551_cast)[name = tensor<string, []>("op_3573_cast")];
            tensor<int32, [4]> var_3574_begin_0 = const()[name = tensor<string, []>("op_3574_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3574_end_0 = const()[name = tensor<string, []>("op_3574_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_3574_end_mask_0 = const()[name = tensor<string, []>("op_3574_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3574_cast = slice_by_index(begin = var_3574_begin_0, end = var_3574_end_0, end_mask = var_3574_end_mask_0, x = var_3555_cast)[name = tensor<string, []>("op_3574_cast")];
            tensor<int32, [4]> var_3575_begin_0 = const()[name = tensor<string, []>("op_3575_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3575_end_0 = const()[name = tensor<string, []>("op_3575_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_3575_end_mask_0 = const()[name = tensor<string, []>("op_3575_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3575_cast = slice_by_index(begin = var_3575_begin_0, end = var_3575_end_0, end_mask = var_3575_end_mask_0, x = var_3555_cast)[name = tensor<string, []>("op_3575_cast")];
            tensor<int32, [4]> var_3576_begin_0 = const()[name = tensor<string, []>("op_3576_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3576_end_0 = const()[name = tensor<string, []>("op_3576_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_3576_end_mask_0 = const()[name = tensor<string, []>("op_3576_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3576_cast = slice_by_index(begin = var_3576_begin_0, end = var_3576_end_0, end_mask = var_3576_end_mask_0, x = var_3559_cast)[name = tensor<string, []>("op_3576_cast")];
            tensor<int32, [4]> var_3577_begin_0 = const()[name = tensor<string, []>("op_3577_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3577_end_0 = const()[name = tensor<string, []>("op_3577_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_3577_end_mask_0 = const()[name = tensor<string, []>("op_3577_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3577_cast = slice_by_index(begin = var_3577_begin_0, end = var_3577_end_0, end_mask = var_3577_end_mask_0, x = var_3559_cast)[name = tensor<string, []>("op_3577_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_3582_begin_0 = const()[name = tensor<string, []>("op_3582_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3582_end_0 = const()[name = tensor<string, []>("op_3582_end_0"), val = tensor<int32, [4]>([2, 77, 1, 80])];
            tensor<bool, [4]> var_3582_end_mask_0 = const()[name = tensor<string, []>("op_3582_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)[name = tensor<string, []>("transpose_26")];
            tensor<fp16, [2, 77, 1, 80]> var_3582_cast = slice_by_index(begin = var_3582_begin_0, end = var_3582_end_0, end_mask = var_3582_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_3582_cast")];
            tensor<int32, [4]> var_3586_begin_0 = const()[name = tensor<string, []>("op_3586_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 80])];
            tensor<int32, [4]> var_3586_end_0 = const()[name = tensor<string, []>("op_3586_end_0"), val = tensor<int32, [4]>([2, 77, 1, 160])];
            tensor<bool, [4]> var_3586_end_mask_0 = const()[name = tensor<string, []>("op_3586_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_3586_cast = slice_by_index(begin = var_3586_begin_0, end = var_3586_end_0, end_mask = var_3586_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_3586_cast")];
            tensor<int32, [4]> var_3590_begin_0 = const()[name = tensor<string, []>("op_3590_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_3590_end_0 = const()[name = tensor<string, []>("op_3590_end_0"), val = tensor<int32, [4]>([2, 77, 1, 240])];
            tensor<bool, [4]> var_3590_end_mask_0 = const()[name = tensor<string, []>("op_3590_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_3590_cast = slice_by_index(begin = var_3590_begin_0, end = var_3590_end_0, end_mask = var_3590_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_3590_cast")];
            tensor<int32, [4]> var_3594_begin_0 = const()[name = tensor<string, []>("op_3594_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 240])];
            tensor<int32, [4]> var_3594_end_0 = const()[name = tensor<string, []>("op_3594_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_3594_end_mask_0 = const()[name = tensor<string, []>("op_3594_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_3594_cast = slice_by_index(begin = var_3594_begin_0, end = var_3594_end_0, end_mask = var_3594_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_3594_cast")];
            tensor<int32, [4]> var_3598_begin_0 = const()[name = tensor<string, []>("op_3598_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_3598_end_0 = const()[name = tensor<string, []>("op_3598_end_0"), val = tensor<int32, [4]>([2, 77, 1, 400])];
            tensor<bool, [4]> var_3598_end_mask_0 = const()[name = tensor<string, []>("op_3598_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_3598_cast = slice_by_index(begin = var_3598_begin_0, end = var_3598_end_0, end_mask = var_3598_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_3598_cast")];
            tensor<int32, [4]> var_3602_begin_0 = const()[name = tensor<string, []>("op_3602_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 400])];
            tensor<int32, [4]> var_3602_end_0 = const()[name = tensor<string, []>("op_3602_end_0"), val = tensor<int32, [4]>([2, 77, 1, 480])];
            tensor<bool, [4]> var_3602_end_mask_0 = const()[name = tensor<string, []>("op_3602_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_3602_cast = slice_by_index(begin = var_3602_begin_0, end = var_3602_end_0, end_mask = var_3602_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_3602_cast")];
            tensor<int32, [4]> var_3606_begin_0 = const()[name = tensor<string, []>("op_3606_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 480])];
            tensor<int32, [4]> var_3606_end_0 = const()[name = tensor<string, []>("op_3606_end_0"), val = tensor<int32, [4]>([2, 77, 1, 560])];
            tensor<bool, [4]> var_3606_end_mask_0 = const()[name = tensor<string, []>("op_3606_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_3606_cast = slice_by_index(begin = var_3606_begin_0, end = var_3606_end_0, end_mask = var_3606_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_3606_cast")];
            tensor<int32, [4]> var_3610_begin_0 = const()[name = tensor<string, []>("op_3610_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 560])];
            tensor<int32, [4]> var_3610_end_0 = const()[name = tensor<string, []>("op_3610_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_3610_end_mask_0 = const()[name = tensor<string, []>("op_3610_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_3610_cast = slice_by_index(begin = var_3610_begin_0, end = var_3610_end_0, end_mask = var_3610_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_3610_cast")];
            tensor<int32, [4]> var_3612_begin_0 = const()[name = tensor<string, []>("op_3612_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3612_end_0 = const()[name = tensor<string, []>("op_3612_end_0"), val = tensor<int32, [4]>([2, 80, 1, 77])];
            tensor<bool, [4]> var_3612_end_mask_0 = const()[name = tensor<string, []>("op_3612_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_3612_cast = slice_by_index(begin = var_3612_begin_0, end = var_3612_end_0, end_mask = var_3612_end_mask_0, x = v_11_cast)[name = tensor<string, []>("op_3612_cast")];
            tensor<int32, [4]> var_3616_begin_0 = const()[name = tensor<string, []>("op_3616_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_3616_end_0 = const()[name = tensor<string, []>("op_3616_end_0"), val = tensor<int32, [4]>([2, 160, 1, 77])];
            tensor<bool, [4]> var_3616_end_mask_0 = const()[name = tensor<string, []>("op_3616_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_3616_cast = slice_by_index(begin = var_3616_begin_0, end = var_3616_end_0, end_mask = var_3616_end_mask_0, x = v_11_cast)[name = tensor<string, []>("op_3616_cast")];
            tensor<int32, [4]> var_3620_begin_0 = const()[name = tensor<string, []>("op_3620_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_3620_end_0 = const()[name = tensor<string, []>("op_3620_end_0"), val = tensor<int32, [4]>([2, 240, 1, 77])];
            tensor<bool, [4]> var_3620_end_mask_0 = const()[name = tensor<string, []>("op_3620_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_3620_cast = slice_by_index(begin = var_3620_begin_0, end = var_3620_end_0, end_mask = var_3620_end_mask_0, x = v_11_cast)[name = tensor<string, []>("op_3620_cast")];
            tensor<int32, [4]> var_3624_begin_0 = const()[name = tensor<string, []>("op_3624_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_3624_end_0 = const()[name = tensor<string, []>("op_3624_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_3624_end_mask_0 = const()[name = tensor<string, []>("op_3624_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_3624_cast = slice_by_index(begin = var_3624_begin_0, end = var_3624_end_0, end_mask = var_3624_end_mask_0, x = v_11_cast)[name = tensor<string, []>("op_3624_cast")];
            tensor<int32, [4]> var_3628_begin_0 = const()[name = tensor<string, []>("op_3628_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_3628_end_0 = const()[name = tensor<string, []>("op_3628_end_0"), val = tensor<int32, [4]>([2, 400, 1, 77])];
            tensor<bool, [4]> var_3628_end_mask_0 = const()[name = tensor<string, []>("op_3628_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_3628_cast = slice_by_index(begin = var_3628_begin_0, end = var_3628_end_0, end_mask = var_3628_end_mask_0, x = v_11_cast)[name = tensor<string, []>("op_3628_cast")];
            tensor<int32, [4]> var_3632_begin_0 = const()[name = tensor<string, []>("op_3632_begin_0"), val = tensor<int32, [4]>([0, 400, 0, 0])];
            tensor<int32, [4]> var_3632_end_0 = const()[name = tensor<string, []>("op_3632_end_0"), val = tensor<int32, [4]>([2, 480, 1, 77])];
            tensor<bool, [4]> var_3632_end_mask_0 = const()[name = tensor<string, []>("op_3632_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_3632_cast = slice_by_index(begin = var_3632_begin_0, end = var_3632_end_0, end_mask = var_3632_end_mask_0, x = v_11_cast)[name = tensor<string, []>("op_3632_cast")];
            tensor<int32, [4]> var_3636_begin_0 = const()[name = tensor<string, []>("op_3636_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_3636_end_0 = const()[name = tensor<string, []>("op_3636_end_0"), val = tensor<int32, [4]>([2, 560, 1, 77])];
            tensor<bool, [4]> var_3636_end_mask_0 = const()[name = tensor<string, []>("op_3636_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_3636_cast = slice_by_index(begin = var_3636_begin_0, end = var_3636_end_0, end_mask = var_3636_end_mask_0, x = v_11_cast)[name = tensor<string, []>("op_3636_cast")];
            tensor<int32, [4]> var_3640_begin_0 = const()[name = tensor<string, []>("op_3640_begin_0"), val = tensor<int32, [4]>([0, 560, 0, 0])];
            tensor<int32, [4]> var_3640_end_0 = const()[name = tensor<string, []>("op_3640_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_3640_end_mask_0 = const()[name = tensor<string, []>("op_3640_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_3640_cast = slice_by_index(begin = var_3640_begin_0, end = var_3640_end_0, end_mask = var_3640_end_mask_0, x = v_11_cast)[name = tensor<string, []>("op_3640_cast")];
            tensor<string, []> var_3644_equation_0 = const()[name = tensor<string, []>("op_3644_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3644_cast = einsum(equation = var_3644_equation_0, values = (var_3582_cast, var_3562_cast))[name = tensor<string, []>("op_3644_cast")];
            tensor<fp16, []> var_3645_to_fp16 = const()[name = tensor<string, []>("op_3645_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_545_cast = mul(x = var_3644_cast, y = var_3645_to_fp16)[name = tensor<string, []>("aw_chunk_545_cast")];
            tensor<string, []> var_3648_equation_0 = const()[name = tensor<string, []>("op_3648_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3648_cast = einsum(equation = var_3648_equation_0, values = (var_3582_cast, var_3563_cast))[name = tensor<string, []>("op_3648_cast")];
            tensor<fp16, []> var_3649_to_fp16 = const()[name = tensor<string, []>("op_3649_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_547_cast = mul(x = var_3648_cast, y = var_3649_to_fp16)[name = tensor<string, []>("aw_chunk_547_cast")];
            tensor<string, []> var_3652_equation_0 = const()[name = tensor<string, []>("op_3652_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3652_cast = einsum(equation = var_3652_equation_0, values = (var_3586_cast, var_3564_cast))[name = tensor<string, []>("op_3652_cast")];
            tensor<fp16, []> var_3653_to_fp16 = const()[name = tensor<string, []>("op_3653_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_549_cast = mul(x = var_3652_cast, y = var_3653_to_fp16)[name = tensor<string, []>("aw_chunk_549_cast")];
            tensor<string, []> var_3656_equation_0 = const()[name = tensor<string, []>("op_3656_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3656_cast = einsum(equation = var_3656_equation_0, values = (var_3586_cast, var_3565_cast))[name = tensor<string, []>("op_3656_cast")];
            tensor<fp16, []> var_3657_to_fp16 = const()[name = tensor<string, []>("op_3657_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_551_cast = mul(x = var_3656_cast, y = var_3657_to_fp16)[name = tensor<string, []>("aw_chunk_551_cast")];
            tensor<string, []> var_3660_equation_0 = const()[name = tensor<string, []>("op_3660_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3660_cast = einsum(equation = var_3660_equation_0, values = (var_3590_cast, var_3566_cast))[name = tensor<string, []>("op_3660_cast")];
            tensor<fp16, []> var_3661_to_fp16 = const()[name = tensor<string, []>("op_3661_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_553_cast = mul(x = var_3660_cast, y = var_3661_to_fp16)[name = tensor<string, []>("aw_chunk_553_cast")];
            tensor<string, []> var_3664_equation_0 = const()[name = tensor<string, []>("op_3664_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3664_cast = einsum(equation = var_3664_equation_0, values = (var_3590_cast, var_3567_cast))[name = tensor<string, []>("op_3664_cast")];
            tensor<fp16, []> var_3665_to_fp16 = const()[name = tensor<string, []>("op_3665_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_555_cast = mul(x = var_3664_cast, y = var_3665_to_fp16)[name = tensor<string, []>("aw_chunk_555_cast")];
            tensor<string, []> var_3668_equation_0 = const()[name = tensor<string, []>("op_3668_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3668_cast = einsum(equation = var_3668_equation_0, values = (var_3594_cast, var_3568_cast))[name = tensor<string, []>("op_3668_cast")];
            tensor<fp16, []> var_3669_to_fp16 = const()[name = tensor<string, []>("op_3669_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_557_cast = mul(x = var_3668_cast, y = var_3669_to_fp16)[name = tensor<string, []>("aw_chunk_557_cast")];
            tensor<string, []> var_3672_equation_0 = const()[name = tensor<string, []>("op_3672_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3672_cast = einsum(equation = var_3672_equation_0, values = (var_3594_cast, var_3569_cast))[name = tensor<string, []>("op_3672_cast")];
            tensor<fp16, []> var_3673_to_fp16 = const()[name = tensor<string, []>("op_3673_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_559_cast = mul(x = var_3672_cast, y = var_3673_to_fp16)[name = tensor<string, []>("aw_chunk_559_cast")];
            tensor<string, []> var_3676_equation_0 = const()[name = tensor<string, []>("op_3676_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3676_cast = einsum(equation = var_3676_equation_0, values = (var_3598_cast, var_3570_cast))[name = tensor<string, []>("op_3676_cast")];
            tensor<fp16, []> var_3677_to_fp16 = const()[name = tensor<string, []>("op_3677_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_561_cast = mul(x = var_3676_cast, y = var_3677_to_fp16)[name = tensor<string, []>("aw_chunk_561_cast")];
            tensor<string, []> var_3680_equation_0 = const()[name = tensor<string, []>("op_3680_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3680_cast = einsum(equation = var_3680_equation_0, values = (var_3598_cast, var_3571_cast))[name = tensor<string, []>("op_3680_cast")];
            tensor<fp16, []> var_3681_to_fp16 = const()[name = tensor<string, []>("op_3681_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_563_cast = mul(x = var_3680_cast, y = var_3681_to_fp16)[name = tensor<string, []>("aw_chunk_563_cast")];
            tensor<string, []> var_3684_equation_0 = const()[name = tensor<string, []>("op_3684_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3684_cast = einsum(equation = var_3684_equation_0, values = (var_3602_cast, var_3572_cast))[name = tensor<string, []>("op_3684_cast")];
            tensor<fp16, []> var_3685_to_fp16 = const()[name = tensor<string, []>("op_3685_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_565_cast = mul(x = var_3684_cast, y = var_3685_to_fp16)[name = tensor<string, []>("aw_chunk_565_cast")];
            tensor<string, []> var_3688_equation_0 = const()[name = tensor<string, []>("op_3688_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3688_cast = einsum(equation = var_3688_equation_0, values = (var_3602_cast, var_3573_cast))[name = tensor<string, []>("op_3688_cast")];
            tensor<fp16, []> var_3689_to_fp16 = const()[name = tensor<string, []>("op_3689_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_567_cast = mul(x = var_3688_cast, y = var_3689_to_fp16)[name = tensor<string, []>("aw_chunk_567_cast")];
            tensor<string, []> var_3692_equation_0 = const()[name = tensor<string, []>("op_3692_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3692_cast = einsum(equation = var_3692_equation_0, values = (var_3606_cast, var_3574_cast))[name = tensor<string, []>("op_3692_cast")];
            tensor<fp16, []> var_3693_to_fp16 = const()[name = tensor<string, []>("op_3693_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_569_cast = mul(x = var_3692_cast, y = var_3693_to_fp16)[name = tensor<string, []>("aw_chunk_569_cast")];
            tensor<string, []> var_3696_equation_0 = const()[name = tensor<string, []>("op_3696_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3696_cast = einsum(equation = var_3696_equation_0, values = (var_3606_cast, var_3575_cast))[name = tensor<string, []>("op_3696_cast")];
            tensor<fp16, []> var_3697_to_fp16 = const()[name = tensor<string, []>("op_3697_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_571_cast = mul(x = var_3696_cast, y = var_3697_to_fp16)[name = tensor<string, []>("aw_chunk_571_cast")];
            tensor<string, []> var_3700_equation_0 = const()[name = tensor<string, []>("op_3700_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3700_cast = einsum(equation = var_3700_equation_0, values = (var_3610_cast, var_3576_cast))[name = tensor<string, []>("op_3700_cast")];
            tensor<fp16, []> var_3701_to_fp16 = const()[name = tensor<string, []>("op_3701_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_573_cast = mul(x = var_3700_cast, y = var_3701_to_fp16)[name = tensor<string, []>("aw_chunk_573_cast")];
            tensor<string, []> var_3704_equation_0 = const()[name = tensor<string, []>("op_3704_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3704_cast = einsum(equation = var_3704_equation_0, values = (var_3610_cast, var_3577_cast))[name = tensor<string, []>("op_3704_cast")];
            tensor<fp16, []> var_3705_to_fp16 = const()[name = tensor<string, []>("op_3705_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_575_cast = mul(x = var_3704_cast, y = var_3705_to_fp16)[name = tensor<string, []>("aw_chunk_575_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3707_cast = softmax(axis = var_3111, x = aw_chunk_545_cast)[name = tensor<string, []>("op_3707_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3708_cast = softmax(axis = var_3111, x = aw_chunk_547_cast)[name = tensor<string, []>("op_3708_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3709_cast = softmax(axis = var_3111, x = aw_chunk_549_cast)[name = tensor<string, []>("op_3709_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3710_cast = softmax(axis = var_3111, x = aw_chunk_551_cast)[name = tensor<string, []>("op_3710_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3711_cast = softmax(axis = var_3111, x = aw_chunk_553_cast)[name = tensor<string, []>("op_3711_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3712_cast = softmax(axis = var_3111, x = aw_chunk_555_cast)[name = tensor<string, []>("op_3712_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3713_cast = softmax(axis = var_3111, x = aw_chunk_557_cast)[name = tensor<string, []>("op_3713_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3714_cast = softmax(axis = var_3111, x = aw_chunk_559_cast)[name = tensor<string, []>("op_3714_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3715_cast = softmax(axis = var_3111, x = aw_chunk_561_cast)[name = tensor<string, []>("op_3715_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3716_cast = softmax(axis = var_3111, x = aw_chunk_563_cast)[name = tensor<string, []>("op_3716_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3717_cast = softmax(axis = var_3111, x = aw_chunk_565_cast)[name = tensor<string, []>("op_3717_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3718_cast = softmax(axis = var_3111, x = aw_chunk_567_cast)[name = tensor<string, []>("op_3718_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3719_cast = softmax(axis = var_3111, x = aw_chunk_569_cast)[name = tensor<string, []>("op_3719_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3720_cast = softmax(axis = var_3111, x = aw_chunk_571_cast)[name = tensor<string, []>("op_3720_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3721_cast = softmax(axis = var_3111, x = aw_chunk_573_cast)[name = tensor<string, []>("op_3721_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3722_cast = softmax(axis = var_3111, x = aw_chunk_575_cast)[name = tensor<string, []>("op_3722_cast")];
            tensor<string, []> var_3724_equation_0 = const()[name = tensor<string, []>("op_3724_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3724_cast = einsum(equation = var_3724_equation_0, values = (var_3612_cast, var_3707_cast))[name = tensor<string, []>("op_3724_cast")];
            tensor<string, []> var_3726_equation_0 = const()[name = tensor<string, []>("op_3726_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3726_cast = einsum(equation = var_3726_equation_0, values = (var_3612_cast, var_3708_cast))[name = tensor<string, []>("op_3726_cast")];
            tensor<string, []> var_3728_equation_0 = const()[name = tensor<string, []>("op_3728_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3728_cast = einsum(equation = var_3728_equation_0, values = (var_3616_cast, var_3709_cast))[name = tensor<string, []>("op_3728_cast")];
            tensor<string, []> var_3730_equation_0 = const()[name = tensor<string, []>("op_3730_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3730_cast = einsum(equation = var_3730_equation_0, values = (var_3616_cast, var_3710_cast))[name = tensor<string, []>("op_3730_cast")];
            tensor<string, []> var_3732_equation_0 = const()[name = tensor<string, []>("op_3732_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3732_cast = einsum(equation = var_3732_equation_0, values = (var_3620_cast, var_3711_cast))[name = tensor<string, []>("op_3732_cast")];
            tensor<string, []> var_3734_equation_0 = const()[name = tensor<string, []>("op_3734_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3734_cast = einsum(equation = var_3734_equation_0, values = (var_3620_cast, var_3712_cast))[name = tensor<string, []>("op_3734_cast")];
            tensor<string, []> var_3736_equation_0 = const()[name = tensor<string, []>("op_3736_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3736_cast = einsum(equation = var_3736_equation_0, values = (var_3624_cast, var_3713_cast))[name = tensor<string, []>("op_3736_cast")];
            tensor<string, []> var_3738_equation_0 = const()[name = tensor<string, []>("op_3738_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3738_cast = einsum(equation = var_3738_equation_0, values = (var_3624_cast, var_3714_cast))[name = tensor<string, []>("op_3738_cast")];
            tensor<string, []> var_3740_equation_0 = const()[name = tensor<string, []>("op_3740_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3740_cast = einsum(equation = var_3740_equation_0, values = (var_3628_cast, var_3715_cast))[name = tensor<string, []>("op_3740_cast")];
            tensor<string, []> var_3742_equation_0 = const()[name = tensor<string, []>("op_3742_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3742_cast = einsum(equation = var_3742_equation_0, values = (var_3628_cast, var_3716_cast))[name = tensor<string, []>("op_3742_cast")];
            tensor<string, []> var_3744_equation_0 = const()[name = tensor<string, []>("op_3744_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3744_cast = einsum(equation = var_3744_equation_0, values = (var_3632_cast, var_3717_cast))[name = tensor<string, []>("op_3744_cast")];
            tensor<string, []> var_3746_equation_0 = const()[name = tensor<string, []>("op_3746_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3746_cast = einsum(equation = var_3746_equation_0, values = (var_3632_cast, var_3718_cast))[name = tensor<string, []>("op_3746_cast")];
            tensor<string, []> var_3748_equation_0 = const()[name = tensor<string, []>("op_3748_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3748_cast = einsum(equation = var_3748_equation_0, values = (var_3636_cast, var_3719_cast))[name = tensor<string, []>("op_3748_cast")];
            tensor<string, []> var_3750_equation_0 = const()[name = tensor<string, []>("op_3750_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3750_cast = einsum(equation = var_3750_equation_0, values = (var_3636_cast, var_3720_cast))[name = tensor<string, []>("op_3750_cast")];
            tensor<string, []> var_3752_equation_0 = const()[name = tensor<string, []>("op_3752_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3752_cast = einsum(equation = var_3752_equation_0, values = (var_3640_cast, var_3721_cast))[name = tensor<string, []>("op_3752_cast")];
            tensor<string, []> var_3754_equation_0 = const()[name = tensor<string, []>("op_3754_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_3754_cast = einsum(equation = var_3754_equation_0, values = (var_3640_cast, var_3722_cast))[name = tensor<string, []>("op_3754_cast")];
            tensor<bool, []> var_3756_interleave_0 = const()[name = tensor<string, []>("op_3756_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_3756_cast = concat(axis = var_3089, interleave = var_3756_interleave_0, values = (var_3724_cast, var_3726_cast))[name = tensor<string, []>("op_3756_cast")];
            tensor<bool, []> var_3758_interleave_0 = const()[name = tensor<string, []>("op_3758_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_3758_cast = concat(axis = var_3089, interleave = var_3758_interleave_0, values = (var_3728_cast, var_3730_cast))[name = tensor<string, []>("op_3758_cast")];
            tensor<bool, []> var_3760_interleave_0 = const()[name = tensor<string, []>("op_3760_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_3760_cast = concat(axis = var_3089, interleave = var_3760_interleave_0, values = (var_3732_cast, var_3734_cast))[name = tensor<string, []>("op_3760_cast")];
            tensor<bool, []> var_3762_interleave_0 = const()[name = tensor<string, []>("op_3762_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_3762_cast = concat(axis = var_3089, interleave = var_3762_interleave_0, values = (var_3736_cast, var_3738_cast))[name = tensor<string, []>("op_3762_cast")];
            tensor<bool, []> var_3764_interleave_0 = const()[name = tensor<string, []>("op_3764_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_3764_cast = concat(axis = var_3089, interleave = var_3764_interleave_0, values = (var_3740_cast, var_3742_cast))[name = tensor<string, []>("op_3764_cast")];
            tensor<bool, []> var_3766_interleave_0 = const()[name = tensor<string, []>("op_3766_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_3766_cast = concat(axis = var_3089, interleave = var_3766_interleave_0, values = (var_3744_cast, var_3746_cast))[name = tensor<string, []>("op_3766_cast")];
            tensor<bool, []> var_3768_interleave_0 = const()[name = tensor<string, []>("op_3768_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_3768_cast = concat(axis = var_3089, interleave = var_3768_interleave_0, values = (var_3748_cast, var_3750_cast))[name = tensor<string, []>("op_3768_cast")];
            tensor<bool, []> var_3770_interleave_0 = const()[name = tensor<string, []>("op_3770_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_3770_cast = concat(axis = var_3089, interleave = var_3770_interleave_0, values = (var_3752_cast, var_3754_cast))[name = tensor<string, []>("op_3770_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_3111, interleave = input_81_interleave_0, values = (var_3756_cast, var_3758_cast, var_3760_cast, var_3762_cast, var_3764_cast, var_3766_cast, var_3768_cast, var_3770_cast))[name = tensor<string, []>("input_81_cast")];
            tensor<int32, [2]> var_3776 = const()[name = tensor<string, []>("op_3776"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3778 = const()[name = tensor<string, []>("op_3778"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_3780_pad_type_0 = const()[name = tensor<string, []>("op_3780_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_3780_pad_0 = const()[name = tensor<string, []>("op_3780_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16995712))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17302976))), name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            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, []>(17303168)))];
            tensor<fp16, [2, 640, 1, 1024]> var_3780_cast = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_3778, groups = var_3111, pad = var_3780_pad_0, pad_type = var_3780_pad_type_0, strides = var_3776, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_81_cast)[name = tensor<string, []>("op_3780_cast")];
            tensor<fp16, [2, 640, 1, 1024]> inputs_17_cast = add(x = var_3780_cast, y = inputs_15_cast)[name = tensor<string, []>("inputs_17_cast")];
            tensor<int32, [1]> var_3784 = const()[name = tensor<string, []>("op_3784"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_17_cast = reduce_mean(axes = var_3784, keep_dims = var_3106, x = inputs_17_cast)[name = tensor<string, []>("channels_mean_17_cast")];
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_17_cast = sub(x = inputs_17_cast, y = channels_mean_17_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_17_cast")];
            tensor<int32, [1]> var_3788 = const()[name = tensor<string, []>("op_3788"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_3789_cast = reduce_mean(axes = var_3788, keep_dims = var_3106, x = zero_mean_sq_17_cast)[name = tensor<string, []>("op_3789_cast")];
            tensor<fp16, []> var_3790_to_fp16 = const()[name = tensor<string, []>("op_3790_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_3791_cast = add(x = var_3789_cast, y = var_3790_to_fp16)[name = tensor<string, []>("op_3791_cast")];
            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_3791_cast)[name = tensor<string, []>("denom_17_cast")];
            tensor<fp16, [2, 640, 1, 1024]> out_17_cast = mul(x = zero_mean_17_cast, y = denom_17_cast)[name = tensor<string, []>("out_17_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_3795_to_fp16 = const()[name = tensor<string, []>("op_3795_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17304512)))];
            tensor<fp16, [2, 640, 1, 1024]> var_3796_cast = add(x = out_17_cast, y = var_3795_to_fp16)[name = tensor<string, []>("op_3796_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_3798_to_fp16 = const()[name = tensor<string, []>("op_3798_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17305856)))];
            tensor<fp16, [2, 640, 1, 1024]> input_83_cast = mul(x = var_3796_cast, y = var_3798_to_fp16)[name = tensor<string, []>("input_83_cast")];
            tensor<int32, [2]> var_3806 = const()[name = tensor<string, []>("op_3806"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3808 = const()[name = tensor<string, []>("op_3808"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_3810_pad_type_0 = const()[name = tensor<string, []>("op_3810_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_3810_pad_0 = const()[name = tensor<string, []>("op_3810_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2457600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17307200))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19764864))), name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([5120, 640, 1, 1])];
            tensor<fp16, [5120]> down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3840]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19765056))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19768960))), name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([5120])];
            tensor<fp16, [2, 5120, 1, 1024]> var_3810_cast = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_3808, groups = var_3111, pad = var_3810_pad_0, pad_type = var_3810_pad_type_0, strides = var_3806, weight = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_83_cast)[name = tensor<string, []>("op_3810_cast")];
            tensor<int32, [2]> var_3811_split_sizes_0 = const()[name = tensor<string, []>("op_3811_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_3811_axis_0 = const()[name = tensor<string, []>("op_3811_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 1024]> var_3811_cast_0, tensor<fp16, [2, 2560, 1, 1024]> var_3811_cast_1 = split(axis = var_3811_axis_0, split_sizes = var_3811_split_sizes_0, x = var_3810_cast)[name = tensor<string, []>("op_3811_cast")];
            tensor<string, []> var_3813_mode_0 = const()[name = tensor<string, []>("op_3813_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 1024]> var_3813_cast = gelu(mode = var_3813_mode_0, x = var_3811_cast_1)[name = tensor<string, []>("op_3813_cast")];
            tensor<fp16, [2, 2560, 1, 1024]> input_85_cast = mul(x = var_3811_cast_0, y = var_3813_cast)[name = tensor<string, []>("input_85_cast")];
            tensor<int32, [2]> var_3817 = const()[name = tensor<string, []>("op_3817"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3819 = const()[name = tensor<string, []>("op_3819"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_3821_pad_type_0 = const()[name = tensor<string, []>("op_3821_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_3821_pad_0 = const()[name = tensor<string, []>("op_3821_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19769152))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20998016))), name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 2560, 1, 1])];
            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, []>(20998208)))];
            tensor<fp16, [2, 640, 1, 1024]> var_3821_cast = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_3819, groups = var_3111, pad = var_3821_pad_0, pad_type = var_3821_pad_type_0, strides = var_3817, weight = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_85_cast)[name = tensor<string, []>("op_3821_cast")];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_51_cast = add(x = var_3821_cast, y = inputs_17_cast)[name = tensor<string, []>("hidden_states_51_cast")];
            tensor<int32, [4]> var_3823 = const()[name = tensor<string, []>("op_3823"), val = tensor<int32, [4]>([2, 640, 32, 32])];
            tensor<fp16, [2, 640, 32, 32]> input_87_cast = reshape(shape = var_3823, x = hidden_states_51_cast)[name = tensor<string, []>("input_87_cast")];
            tensor<int32, [2]> var_3827 = const()[name = tensor<string, []>("op_3827"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3829 = const()[name = tensor<string, []>("op_3829"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20999552))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21306816))), name = tensor<string, []>("down_blocks_1_attentions_0_proj_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            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, []>(21307008)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_53_cast = conv(bias = down_blocks_1_attentions_0_proj_out_bias_to_fp16, dilations = var_3829, groups = var_3111, pad = hidden_states_53_pad_0, pad_type = hidden_states_53_pad_type_0, strides = var_3827, weight = down_blocks_1_attentions_0_proj_out_weight_to_fp16_palettized, x = input_87_cast)[name = tensor<string, []>("hidden_states_53_cast")];
            tensor<fp16, [2, 640, 32, 32]> input_89_cast = add(x = hidden_states_53_cast, y = hidden_states_41_cast)[name = tensor<string, []>("input_89_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)[name = tensor<string, []>("reshape_36_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)[name = tensor<string, []>("reduce_mean_27_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> sub_18_cast = sub(x = reshape_36_cast, y = reduce_mean_27_cast)[name = tensor<string, []>("sub_18_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> square_9_cast = square(x = sub_18_cast)[name = tensor<string, []>("square_9_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)[name = tensor<string, []>("reduce_mean_29_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)[name = tensor<string, []>("add_18_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_9_cast = sqrt(x = add_18_cast)[name = tensor<string, []>("sqrt_9_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> real_div_9_cast = real_div(x = sub_18_cast, y = sqrt_9_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_37_cast")];
            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, []>(21308352)))];
            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, []>(21309696)))];
            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_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_37_cast)[name = tensor<string, []>("add_19_cast")];
            tensor<fp16, [2, 640, 32, 32]> input_93_cast = silu(x = add_19_cast)[name = tensor<string, []>("input_93_cast")];
            tensor<int32, [2]> var_3844 = const()[name = tensor<string, []>("op_3844"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3846 = const()[name = tensor<string, []>("op_3846"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2764800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21311040))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24075904))), name = tensor<string, []>("down_blocks_1_resnets_1_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 3, 3])];
            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, []>(24076096)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_55_cast = conv(bias = down_blocks_1_resnets_1_conv1_bias_to_fp16, dilations = var_3846, groups = var_3111, pad = hidden_states_55_pad_0, pad_type = hidden_states_55_pad_type_0, strides = var_3844, weight = down_blocks_1_resnets_1_conv1_weight_to_fp16_palettized, x = input_93_cast)[name = tensor<string, []>("hidden_states_55_cast")];
            tensor<int32, [2]> var_3852 = const()[name = tensor<string, []>("op_3852"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3854 = const()[name = tensor<string, []>("op_3854"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [614400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24077440))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24691904))), name = tensor<string, []>("down_blocks_1_resnets_1_time_emb_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 1280, 1, 1])];
            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, []>(24692096)))];
            tensor<fp16, [2, 640, 1, 1]> temb_7_cast = conv(bias = down_blocks_1_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_3854, groups = var_3111, pad = temb_7_pad_0, pad_type = temb_7_pad_type_0, strides = var_3852, weight = down_blocks_1_resnets_1_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor<string, []>("temb_7_cast")];
            tensor<fp16, [2, 640, 32, 32]> input_97_cast = add(x = hidden_states_55_cast, y = temb_7_cast)[name = tensor<string, []>("input_97_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)[name = tensor<string, []>("reshape_40_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)[name = tensor<string, []>("reduce_mean_30_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> sub_20_cast = sub(x = reshape_40_cast, y = reduce_mean_30_cast)[name = tensor<string, []>("sub_20_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> square_10_cast = square(x = sub_20_cast)[name = tensor<string, []>("square_10_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)[name = tensor<string, []>("reduce_mean_32_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)[name = tensor<string, []>("add_20_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_10_cast = sqrt(x = add_20_cast)[name = tensor<string, []>("sqrt_10_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> real_div_10_cast = real_div(x = sub_20_cast, y = sqrt_10_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_41_cast")];
            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, []>(24693440)))];
            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, []>(24694784)))];
            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_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_41_cast)[name = tensor<string, []>("add_21_cast")];
            tensor<fp16, [2, 640, 32, 32]> input_101_cast = silu(x = add_21_cast)[name = tensor<string, []>("input_101_cast")];
            tensor<int32, [2]> var_3864 = const()[name = tensor<string, []>("op_3864"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3866 = const()[name = tensor<string, []>("op_3866"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2764800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24696128))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27460992))), name = tensor<string, []>("down_blocks_1_resnets_1_conv2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 3, 3])];
            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, []>(27461184)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_57_cast = conv(bias = down_blocks_1_resnets_1_conv2_bias_to_fp16, dilations = var_3866, groups = var_3111, pad = hidden_states_57_pad_0, pad_type = hidden_states_57_pad_type_0, strides = var_3864, weight = down_blocks_1_resnets_1_conv2_weight_to_fp16_palettized, x = input_101_cast)[name = tensor<string, []>("hidden_states_57_cast")];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_59_cast = add(x = input_89_cast, y = hidden_states_57_cast)[name = tensor<string, []>("hidden_states_59_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)[name = tensor<string, []>("reshape_44_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)[name = tensor<string, []>("reduce_mean_33_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> sub_22_cast = sub(x = reshape_44_cast, y = reduce_mean_33_cast)[name = tensor<string, []>("sub_22_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> square_11_cast = square(x = sub_22_cast)[name = tensor<string, []>("square_11_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)[name = tensor<string, []>("reduce_mean_35_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)[name = tensor<string, []>("add_22_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_11_cast = sqrt(x = add_22_cast)[name = tensor<string, []>("sqrt_11_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> real_div_11_cast = real_div(x = sub_22_cast, y = sqrt_11_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_45_cast")];
            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, []>(27462528)))];
            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, []>(27463872)))];
            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_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_45_cast)[name = tensor<string, []>("add_23_cast")];
            tensor<int32, [2]> var_3886 = const()[name = tensor<string, []>("op_3886"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3888 = const()[name = tensor<string, []>("op_3888"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27465216))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27772480))), name = tensor<string, []>("down_blocks_1_attentions_1_proj_in_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            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, []>(27772672)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_61_cast = conv(bias = down_blocks_1_attentions_1_proj_in_bias_to_fp16, dilations = var_3888, groups = var_3111, pad = hidden_states_61_pad_0, pad_type = hidden_states_61_pad_type_0, strides = var_3886, weight = down_blocks_1_attentions_1_proj_in_weight_to_fp16_palettized, x = add_23_cast)[name = tensor<string, []>("hidden_states_61_cast")];
            tensor<int32, [4]> var_3893 = const()[name = tensor<string, []>("op_3893"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<fp16, [2, 640, 1, 1024]> inputs_19_cast = reshape(shape = var_3893, x = hidden_states_61_cast)[name = tensor<string, []>("inputs_19_cast")];
            tensor<int32, [1]> var_3903 = const()[name = tensor<string, []>("op_3903"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_19_cast = reduce_mean(axes = var_3903, keep_dims = var_3106, x = inputs_19_cast)[name = tensor<string, []>("channels_mean_19_cast")];
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_19_cast = sub(x = inputs_19_cast, y = channels_mean_19_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_19_cast")];
            tensor<int32, [1]> var_3907 = const()[name = tensor<string, []>("op_3907"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_3908_cast = reduce_mean(axes = var_3907, keep_dims = var_3106, x = zero_mean_sq_19_cast)[name = tensor<string, []>("op_3908_cast")];
            tensor<fp16, []> var_3909_to_fp16 = const()[name = tensor<string, []>("op_3909_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_3910_cast = add(x = var_3908_cast, y = var_3909_to_fp16)[name = tensor<string, []>("op_3910_cast")];
            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_3910_cast)[name = tensor<string, []>("denom_19_cast")];
            tensor<fp16, [2, 640, 1, 1024]> out_19_cast = mul(x = zero_mean_19_cast, y = denom_19_cast)[name = tensor<string, []>("out_19_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_3914_to_fp16 = const()[name = tensor<string, []>("op_3914_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27774016)))];
            tensor<fp16, [2, 640, 1, 1024]> var_3915_cast = add(x = out_19_cast, y = var_3914_to_fp16)[name = tensor<string, []>("op_3915_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_3917_to_fp16 = const()[name = tensor<string, []>("op_3917_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27775360)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_63_cast = mul(x = var_3915_cast, y = var_3917_to_fp16)[name = tensor<string, []>("hidden_states_63_cast")];
            tensor<int32, [2]> var_3924 = const()[name = tensor<string, []>("op_3924"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3926 = const()[name = tensor<string, []>("op_3926"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27776704))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28083968))), name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 1024]> q_13_cast = conv(dilations = var_3926, groups = var_3111, pad = q_13_pad_0, pad_type = q_13_pad_type_0, strides = var_3924, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_63_cast)[name = tensor<string, []>("q_13_cast")];
            tensor<int32, [2]> var_3930 = const()[name = tensor<string, []>("op_3930"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3932 = const()[name = tensor<string, []>("op_3932"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28084160))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28391424))), name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 1024]> k_25_cast = conv(dilations = var_3932, groups = var_3111, pad = k_25_pad_0, pad_type = k_25_pad_type_0, strides = var_3930, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_63_cast)[name = tensor<string, []>("k_25_cast")];
            tensor<int32, [2]> var_3936 = const()[name = tensor<string, []>("op_3936"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3938 = const()[name = tensor<string, []>("op_3938"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28391616))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28698880))), name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 1024]> v_13_cast = conv(dilations = var_3938, groups = var_3111, pad = v_13_pad_0, pad_type = v_13_pad_type_0, strides = var_3936, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_63_cast)[name = tensor<string, []>("v_13_cast")];
            tensor<int32, [4]> var_3942_begin_0 = const()[name = tensor<string, []>("op_3942_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3942_end_0 = const()[name = tensor<string, []>("op_3942_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_3942_end_mask_0 = const()[name = tensor<string, []>("op_3942_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_3942_cast = slice_by_index(begin = var_3942_begin_0, end = var_3942_end_0, end_mask = var_3942_end_mask_0, x = q_13_cast)[name = tensor<string, []>("op_3942_cast")];
            tensor<int32, [4]> var_3946_begin_0 = const()[name = tensor<string, []>("op_3946_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_3946_end_0 = const()[name = tensor<string, []>("op_3946_end_0"), val = tensor<int32, [4]>([2, 160, 1, 1024])];
            tensor<bool, [4]> var_3946_end_mask_0 = const()[name = tensor<string, []>("op_3946_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_3946_cast = slice_by_index(begin = var_3946_begin_0, end = var_3946_end_0, end_mask = var_3946_end_mask_0, x = q_13_cast)[name = tensor<string, []>("op_3946_cast")];
            tensor<int32, [4]> var_3950_begin_0 = const()[name = tensor<string, []>("op_3950_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_3950_end_0 = const()[name = tensor<string, []>("op_3950_end_0"), val = tensor<int32, [4]>([2, 240, 1, 1024])];
            tensor<bool, [4]> var_3950_end_mask_0 = const()[name = tensor<string, []>("op_3950_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_3950_cast = slice_by_index(begin = var_3950_begin_0, end = var_3950_end_0, end_mask = var_3950_end_mask_0, x = q_13_cast)[name = tensor<string, []>("op_3950_cast")];
            tensor<int32, [4]> var_3954_begin_0 = const()[name = tensor<string, []>("op_3954_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_3954_end_0 = const()[name = tensor<string, []>("op_3954_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_3954_end_mask_0 = const()[name = tensor<string, []>("op_3954_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_3954_cast = slice_by_index(begin = var_3954_begin_0, end = var_3954_end_0, end_mask = var_3954_end_mask_0, x = q_13_cast)[name = tensor<string, []>("op_3954_cast")];
            tensor<int32, [4]> var_3958_begin_0 = const()[name = tensor<string, []>("op_3958_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_3958_end_0 = const()[name = tensor<string, []>("op_3958_end_0"), val = tensor<int32, [4]>([2, 400, 1, 1024])];
            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, 80, 1, 1024]> var_3958_cast = slice_by_index(begin = var_3958_begin_0, end = var_3958_end_0, end_mask = var_3958_end_mask_0, x = q_13_cast)[name = tensor<string, []>("op_3958_cast")];
            tensor<int32, [4]> var_3962_begin_0 = const()[name = tensor<string, []>("op_3962_begin_0"), val = tensor<int32, [4]>([0, 400, 0, 0])];
            tensor<int32, [4]> var_3962_end_0 = const()[name = tensor<string, []>("op_3962_end_0"), val = tensor<int32, [4]>([2, 480, 1, 1024])];
            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, 80, 1, 1024]> var_3962_cast = slice_by_index(begin = var_3962_begin_0, end = var_3962_end_0, end_mask = var_3962_end_mask_0, x = q_13_cast)[name = tensor<string, []>("op_3962_cast")];
            tensor<int32, [4]> var_3966_begin_0 = const()[name = tensor<string, []>("op_3966_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_3966_end_0 = const()[name = tensor<string, []>("op_3966_end_0"), val = tensor<int32, [4]>([2, 560, 1, 1024])];
            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, 80, 1, 1024]> var_3966_cast = slice_by_index(begin = var_3966_begin_0, end = var_3966_end_0, end_mask = var_3966_end_mask_0, x = q_13_cast)[name = tensor<string, []>("op_3966_cast")];
            tensor<int32, [4]> var_3970_begin_0 = const()[name = tensor<string, []>("op_3970_begin_0"), val = tensor<int32, [4]>([0, 560, 0, 0])];
            tensor<int32, [4]> var_3970_end_0 = const()[name = tensor<string, []>("op_3970_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            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, 80, 1, 1024]> var_3970_cast = slice_by_index(begin = var_3970_begin_0, end = var_3970_end_0, end_mask = var_3970_end_mask_0, x = q_13_cast)[name = tensor<string, []>("op_3970_cast")];
            tensor<int32, [4]> var_3973_begin_0 = const()[name = tensor<string, []>("op_3973_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3973_end_0 = const()[name = tensor<string, []>("op_3973_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_3973_end_mask_0 = const()[name = tensor<string, []>("op_3973_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3973_cast = slice_by_index(begin = var_3973_begin_0, end = var_3973_end_0, end_mask = var_3973_end_mask_0, x = var_3942_cast)[name = tensor<string, []>("op_3973_cast")];
            tensor<int32, [4]> var_3974_begin_0 = const()[name = tensor<string, []>("op_3974_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3974_end_0 = const()[name = tensor<string, []>("op_3974_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_3974_end_mask_0 = const()[name = tensor<string, []>("op_3974_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3974_cast = slice_by_index(begin = var_3974_begin_0, end = var_3974_end_0, end_mask = var_3974_end_mask_0, x = var_3942_cast)[name = tensor<string, []>("op_3974_cast")];
            tensor<int32, [4]> var_3975_begin_0 = const()[name = tensor<string, []>("op_3975_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3975_end_0 = const()[name = tensor<string, []>("op_3975_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_3975_end_mask_0 = const()[name = tensor<string, []>("op_3975_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3975_cast = slice_by_index(begin = var_3975_begin_0, end = var_3975_end_0, end_mask = var_3975_end_mask_0, x = var_3946_cast)[name = tensor<string, []>("op_3975_cast")];
            tensor<int32, [4]> var_3976_begin_0 = const()[name = tensor<string, []>("op_3976_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3976_end_0 = const()[name = tensor<string, []>("op_3976_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_3976_end_mask_0 = const()[name = tensor<string, []>("op_3976_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3976_cast = slice_by_index(begin = var_3976_begin_0, end = var_3976_end_0, end_mask = var_3976_end_mask_0, x = var_3946_cast)[name = tensor<string, []>("op_3976_cast")];
            tensor<int32, [4]> var_3977_begin_0 = const()[name = tensor<string, []>("op_3977_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3977_end_0 = const()[name = tensor<string, []>("op_3977_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_3977_end_mask_0 = const()[name = tensor<string, []>("op_3977_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3977_cast = slice_by_index(begin = var_3977_begin_0, end = var_3977_end_0, end_mask = var_3977_end_mask_0, x = var_3950_cast)[name = tensor<string, []>("op_3977_cast")];
            tensor<int32, [4]> var_3978_begin_0 = const()[name = tensor<string, []>("op_3978_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3978_end_0 = const()[name = tensor<string, []>("op_3978_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_3978_end_mask_0 = const()[name = tensor<string, []>("op_3978_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3978_cast = slice_by_index(begin = var_3978_begin_0, end = var_3978_end_0, end_mask = var_3978_end_mask_0, x = var_3950_cast)[name = tensor<string, []>("op_3978_cast")];
            tensor<int32, [4]> var_3979_begin_0 = const()[name = tensor<string, []>("op_3979_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3979_end_0 = const()[name = tensor<string, []>("op_3979_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_3979_end_mask_0 = const()[name = tensor<string, []>("op_3979_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3979_cast = slice_by_index(begin = var_3979_begin_0, end = var_3979_end_0, end_mask = var_3979_end_mask_0, x = var_3954_cast)[name = tensor<string, []>("op_3979_cast")];
            tensor<int32, [4]> var_3980_begin_0 = const()[name = tensor<string, []>("op_3980_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3980_end_0 = const()[name = tensor<string, []>("op_3980_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_3980_end_mask_0 = const()[name = tensor<string, []>("op_3980_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3980_cast = slice_by_index(begin = var_3980_begin_0, end = var_3980_end_0, end_mask = var_3980_end_mask_0, x = var_3954_cast)[name = tensor<string, []>("op_3980_cast")];
            tensor<int32, [4]> var_3981_begin_0 = const()[name = tensor<string, []>("op_3981_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3981_end_0 = const()[name = tensor<string, []>("op_3981_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_3981_end_mask_0 = const()[name = tensor<string, []>("op_3981_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3981_cast = slice_by_index(begin = var_3981_begin_0, end = var_3981_end_0, end_mask = var_3981_end_mask_0, x = var_3958_cast)[name = tensor<string, []>("op_3981_cast")];
            tensor<int32, [4]> var_3982_begin_0 = const()[name = tensor<string, []>("op_3982_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3982_end_0 = const()[name = tensor<string, []>("op_3982_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_3982_end_mask_0 = const()[name = tensor<string, []>("op_3982_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3982_cast = slice_by_index(begin = var_3982_begin_0, end = var_3982_end_0, end_mask = var_3982_end_mask_0, x = var_3958_cast)[name = tensor<string, []>("op_3982_cast")];
            tensor<int32, [4]> var_3983_begin_0 = const()[name = tensor<string, []>("op_3983_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3983_end_0 = const()[name = tensor<string, []>("op_3983_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_3983_end_mask_0 = const()[name = tensor<string, []>("op_3983_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3983_cast = slice_by_index(begin = var_3983_begin_0, end = var_3983_end_0, end_mask = var_3983_end_mask_0, x = var_3962_cast)[name = tensor<string, []>("op_3983_cast")];
            tensor<int32, [4]> var_3984_begin_0 = const()[name = tensor<string, []>("op_3984_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3984_end_0 = const()[name = tensor<string, []>("op_3984_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_3984_end_mask_0 = const()[name = tensor<string, []>("op_3984_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3984_cast = slice_by_index(begin = var_3984_begin_0, end = var_3984_end_0, end_mask = var_3984_end_mask_0, x = var_3962_cast)[name = tensor<string, []>("op_3984_cast")];
            tensor<int32, [4]> var_3985_begin_0 = const()[name = tensor<string, []>("op_3985_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3985_end_0 = const()[name = tensor<string, []>("op_3985_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_3985_end_mask_0 = const()[name = tensor<string, []>("op_3985_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3985_cast = slice_by_index(begin = var_3985_begin_0, end = var_3985_end_0, end_mask = var_3985_end_mask_0, x = var_3966_cast)[name = tensor<string, []>("op_3985_cast")];
            tensor<int32, [4]> var_3986_begin_0 = const()[name = tensor<string, []>("op_3986_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3986_end_0 = const()[name = tensor<string, []>("op_3986_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_3986_end_mask_0 = const()[name = tensor<string, []>("op_3986_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3986_cast = slice_by_index(begin = var_3986_begin_0, end = var_3986_end_0, end_mask = var_3986_end_mask_0, x = var_3966_cast)[name = tensor<string, []>("op_3986_cast")];
            tensor<int32, [4]> var_3987_begin_0 = const()[name = tensor<string, []>("op_3987_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3987_end_0 = const()[name = tensor<string, []>("op_3987_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_3987_end_mask_0 = const()[name = tensor<string, []>("op_3987_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3987_cast = slice_by_index(begin = var_3987_begin_0, end = var_3987_end_0, end_mask = var_3987_end_mask_0, x = var_3970_cast)[name = tensor<string, []>("op_3987_cast")];
            tensor<int32, [4]> var_3988_begin_0 = const()[name = tensor<string, []>("op_3988_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3988_end_0 = const()[name = tensor<string, []>("op_3988_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_3988_end_mask_0 = const()[name = tensor<string, []>("op_3988_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_3988_cast = slice_by_index(begin = var_3988_begin_0, end = var_3988_end_0, end_mask = var_3988_end_mask_0, x = var_3970_cast)[name = tensor<string, []>("op_3988_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_3993_begin_0 = const()[name = tensor<string, []>("op_3993_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3993_end_0 = const()[name = tensor<string, []>("op_3993_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 80])];
            tensor<bool, [4]> var_3993_end_mask_0 = const()[name = tensor<string, []>("op_3993_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)[name = tensor<string, []>("transpose_25")];
            tensor<fp16, [2, 1024, 1, 80]> var_3993_cast = slice_by_index(begin = var_3993_begin_0, end = var_3993_end_0, end_mask = var_3993_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_3993_cast")];
            tensor<int32, [4]> var_3997_begin_0 = const()[name = tensor<string, []>("op_3997_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 80])];
            tensor<int32, [4]> var_3997_end_0 = const()[name = tensor<string, []>("op_3997_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 160])];
            tensor<bool, [4]> var_3997_end_mask_0 = const()[name = tensor<string, []>("op_3997_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_3997_cast = slice_by_index(begin = var_3997_begin_0, end = var_3997_end_0, end_mask = var_3997_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_3997_cast")];
            tensor<int32, [4]> var_4001_begin_0 = const()[name = tensor<string, []>("op_4001_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_4001_end_0 = const()[name = tensor<string, []>("op_4001_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 240])];
            tensor<bool, [4]> var_4001_end_mask_0 = const()[name = tensor<string, []>("op_4001_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_4001_cast = slice_by_index(begin = var_4001_begin_0, end = var_4001_end_0, end_mask = var_4001_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_4001_cast")];
            tensor<int32, [4]> var_4005_begin_0 = const()[name = tensor<string, []>("op_4005_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 240])];
            tensor<int32, [4]> var_4005_end_0 = const()[name = tensor<string, []>("op_4005_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_4005_end_mask_0 = const()[name = tensor<string, []>("op_4005_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_4005_cast = slice_by_index(begin = var_4005_begin_0, end = var_4005_end_0, end_mask = var_4005_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_4005_cast")];
            tensor<int32, [4]> var_4009_begin_0 = const()[name = tensor<string, []>("op_4009_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_4009_end_0 = const()[name = tensor<string, []>("op_4009_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 400])];
            tensor<bool, [4]> var_4009_end_mask_0 = const()[name = tensor<string, []>("op_4009_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_4009_cast = slice_by_index(begin = var_4009_begin_0, end = var_4009_end_0, end_mask = var_4009_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_4009_cast")];
            tensor<int32, [4]> var_4013_begin_0 = const()[name = tensor<string, []>("op_4013_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 400])];
            tensor<int32, [4]> var_4013_end_0 = const()[name = tensor<string, []>("op_4013_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 480])];
            tensor<bool, [4]> var_4013_end_mask_0 = const()[name = tensor<string, []>("op_4013_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_4013_cast = slice_by_index(begin = var_4013_begin_0, end = var_4013_end_0, end_mask = var_4013_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_4013_cast")];
            tensor<int32, [4]> var_4017_begin_0 = const()[name = tensor<string, []>("op_4017_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 480])];
            tensor<int32, [4]> var_4017_end_0 = const()[name = tensor<string, []>("op_4017_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 560])];
            tensor<bool, [4]> var_4017_end_mask_0 = const()[name = tensor<string, []>("op_4017_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_4017_cast = slice_by_index(begin = var_4017_begin_0, end = var_4017_end_0, end_mask = var_4017_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_4017_cast")];
            tensor<int32, [4]> var_4021_begin_0 = const()[name = tensor<string, []>("op_4021_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 560])];
            tensor<int32, [4]> var_4021_end_0 = const()[name = tensor<string, []>("op_4021_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_4021_end_mask_0 = const()[name = tensor<string, []>("op_4021_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_4021_cast = slice_by_index(begin = var_4021_begin_0, end = var_4021_end_0, end_mask = var_4021_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_4021_cast")];
            tensor<int32, [4]> var_4023_begin_0 = const()[name = tensor<string, []>("op_4023_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4023_end_0 = const()[name = tensor<string, []>("op_4023_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_4023_end_mask_0 = const()[name = tensor<string, []>("op_4023_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_4023_cast = slice_by_index(begin = var_4023_begin_0, end = var_4023_end_0, end_mask = var_4023_end_mask_0, x = v_13_cast)[name = tensor<string, []>("op_4023_cast")];
            tensor<int32, [4]> var_4027_begin_0 = const()[name = tensor<string, []>("op_4027_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_4027_end_0 = const()[name = tensor<string, []>("op_4027_end_0"), val = tensor<int32, [4]>([2, 160, 1, 1024])];
            tensor<bool, [4]> var_4027_end_mask_0 = const()[name = tensor<string, []>("op_4027_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_4027_cast = slice_by_index(begin = var_4027_begin_0, end = var_4027_end_0, end_mask = var_4027_end_mask_0, x = v_13_cast)[name = tensor<string, []>("op_4027_cast")];
            tensor<int32, [4]> var_4031_begin_0 = const()[name = tensor<string, []>("op_4031_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_4031_end_0 = const()[name = tensor<string, []>("op_4031_end_0"), val = tensor<int32, [4]>([2, 240, 1, 1024])];
            tensor<bool, [4]> var_4031_end_mask_0 = const()[name = tensor<string, []>("op_4031_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_4031_cast = slice_by_index(begin = var_4031_begin_0, end = var_4031_end_0, end_mask = var_4031_end_mask_0, x = v_13_cast)[name = tensor<string, []>("op_4031_cast")];
            tensor<int32, [4]> var_4035_begin_0 = const()[name = tensor<string, []>("op_4035_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_4035_end_0 = const()[name = tensor<string, []>("op_4035_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_4035_end_mask_0 = const()[name = tensor<string, []>("op_4035_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_4035_cast = slice_by_index(begin = var_4035_begin_0, end = var_4035_end_0, end_mask = var_4035_end_mask_0, x = v_13_cast)[name = tensor<string, []>("op_4035_cast")];
            tensor<int32, [4]> var_4039_begin_0 = const()[name = tensor<string, []>("op_4039_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_4039_end_0 = const()[name = tensor<string, []>("op_4039_end_0"), val = tensor<int32, [4]>([2, 400, 1, 1024])];
            tensor<bool, [4]> var_4039_end_mask_0 = const()[name = tensor<string, []>("op_4039_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_4039_cast = slice_by_index(begin = var_4039_begin_0, end = var_4039_end_0, end_mask = var_4039_end_mask_0, x = v_13_cast)[name = tensor<string, []>("op_4039_cast")];
            tensor<int32, [4]> var_4043_begin_0 = const()[name = tensor<string, []>("op_4043_begin_0"), val = tensor<int32, [4]>([0, 400, 0, 0])];
            tensor<int32, [4]> var_4043_end_0 = const()[name = tensor<string, []>("op_4043_end_0"), val = tensor<int32, [4]>([2, 480, 1, 1024])];
            tensor<bool, [4]> var_4043_end_mask_0 = const()[name = tensor<string, []>("op_4043_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_4043_cast = slice_by_index(begin = var_4043_begin_0, end = var_4043_end_0, end_mask = var_4043_end_mask_0, x = v_13_cast)[name = tensor<string, []>("op_4043_cast")];
            tensor<int32, [4]> var_4047_begin_0 = const()[name = tensor<string, []>("op_4047_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_4047_end_0 = const()[name = tensor<string, []>("op_4047_end_0"), val = tensor<int32, [4]>([2, 560, 1, 1024])];
            tensor<bool, [4]> var_4047_end_mask_0 = const()[name = tensor<string, []>("op_4047_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_4047_cast = slice_by_index(begin = var_4047_begin_0, end = var_4047_end_0, end_mask = var_4047_end_mask_0, x = v_13_cast)[name = tensor<string, []>("op_4047_cast")];
            tensor<int32, [4]> var_4051_begin_0 = const()[name = tensor<string, []>("op_4051_begin_0"), val = tensor<int32, [4]>([0, 560, 0, 0])];
            tensor<int32, [4]> var_4051_end_0 = const()[name = tensor<string, []>("op_4051_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_4051_end_mask_0 = const()[name = tensor<string, []>("op_4051_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_4051_cast = slice_by_index(begin = var_4051_begin_0, end = var_4051_end_0, end_mask = var_4051_end_mask_0, x = v_13_cast)[name = tensor<string, []>("op_4051_cast")];
            tensor<string, []> var_4055_equation_0 = const()[name = tensor<string, []>("op_4055_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_4055_cast = einsum(equation = var_4055_equation_0, values = (var_3993_cast, var_3973_cast))[name = tensor<string, []>("op_4055_cast")];
            tensor<fp16, []> var_4056_to_fp16 = const()[name = tensor<string, []>("op_4056_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_577_cast = mul(x = var_4055_cast, y = var_4056_to_fp16)[name = tensor<string, []>("aw_chunk_577_cast")];
            tensor<string, []> var_4059_equation_0 = const()[name = tensor<string, []>("op_4059_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_4059_cast = einsum(equation = var_4059_equation_0, values = (var_3993_cast, var_3974_cast))[name = tensor<string, []>("op_4059_cast")];
            tensor<fp16, []> var_4060_to_fp16 = const()[name = tensor<string, []>("op_4060_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_579_cast = mul(x = var_4059_cast, y = var_4060_to_fp16)[name = tensor<string, []>("aw_chunk_579_cast")];
            tensor<string, []> var_4063_equation_0 = const()[name = tensor<string, []>("op_4063_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_4063_cast = einsum(equation = var_4063_equation_0, values = (var_3997_cast, var_3975_cast))[name = tensor<string, []>("op_4063_cast")];
            tensor<fp16, []> var_4064_to_fp16 = const()[name = tensor<string, []>("op_4064_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_581_cast = mul(x = var_4063_cast, y = var_4064_to_fp16)[name = tensor<string, []>("aw_chunk_581_cast")];
            tensor<string, []> var_4067_equation_0 = const()[name = tensor<string, []>("op_4067_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_4067_cast = einsum(equation = var_4067_equation_0, values = (var_3997_cast, var_3976_cast))[name = tensor<string, []>("op_4067_cast")];
            tensor<fp16, []> var_4068_to_fp16 = const()[name = tensor<string, []>("op_4068_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_583_cast = mul(x = var_4067_cast, y = var_4068_to_fp16)[name = tensor<string, []>("aw_chunk_583_cast")];
            tensor<string, []> var_4071_equation_0 = const()[name = tensor<string, []>("op_4071_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_4071_cast = einsum(equation = var_4071_equation_0, values = (var_4001_cast, var_3977_cast))[name = tensor<string, []>("op_4071_cast")];
            tensor<fp16, []> var_4072_to_fp16 = const()[name = tensor<string, []>("op_4072_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_585_cast = mul(x = var_4071_cast, y = var_4072_to_fp16)[name = tensor<string, []>("aw_chunk_585_cast")];
            tensor<string, []> var_4075_equation_0 = const()[name = tensor<string, []>("op_4075_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_4075_cast = einsum(equation = var_4075_equation_0, values = (var_4001_cast, var_3978_cast))[name = tensor<string, []>("op_4075_cast")];
            tensor<fp16, []> var_4076_to_fp16 = const()[name = tensor<string, []>("op_4076_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_587_cast = mul(x = var_4075_cast, y = var_4076_to_fp16)[name = tensor<string, []>("aw_chunk_587_cast")];
            tensor<string, []> var_4079_equation_0 = const()[name = tensor<string, []>("op_4079_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_4079_cast = einsum(equation = var_4079_equation_0, values = (var_4005_cast, var_3979_cast))[name = tensor<string, []>("op_4079_cast")];
            tensor<fp16, []> var_4080_to_fp16 = const()[name = tensor<string, []>("op_4080_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_589_cast = mul(x = var_4079_cast, y = var_4080_to_fp16)[name = tensor<string, []>("aw_chunk_589_cast")];
            tensor<string, []> var_4083_equation_0 = const()[name = tensor<string, []>("op_4083_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_4083_cast = einsum(equation = var_4083_equation_0, values = (var_4005_cast, var_3980_cast))[name = tensor<string, []>("op_4083_cast")];
            tensor<fp16, []> var_4084_to_fp16 = const()[name = tensor<string, []>("op_4084_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_591_cast = mul(x = var_4083_cast, y = var_4084_to_fp16)[name = tensor<string, []>("aw_chunk_591_cast")];
            tensor<string, []> var_4087_equation_0 = const()[name = tensor<string, []>("op_4087_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_4087_cast = einsum(equation = var_4087_equation_0, values = (var_4009_cast, var_3981_cast))[name = tensor<string, []>("op_4087_cast")];
            tensor<fp16, []> var_4088_to_fp16 = const()[name = tensor<string, []>("op_4088_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_593_cast = mul(x = var_4087_cast, y = var_4088_to_fp16)[name = tensor<string, []>("aw_chunk_593_cast")];
            tensor<string, []> var_4091_equation_0 = const()[name = tensor<string, []>("op_4091_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_4091_cast = einsum(equation = var_4091_equation_0, values = (var_4009_cast, var_3982_cast))[name = tensor<string, []>("op_4091_cast")];
            tensor<fp16, []> var_4092_to_fp16 = const()[name = tensor<string, []>("op_4092_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_595_cast = mul(x = var_4091_cast, y = var_4092_to_fp16)[name = tensor<string, []>("aw_chunk_595_cast")];
            tensor<string, []> var_4095_equation_0 = const()[name = tensor<string, []>("op_4095_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_4095_cast = einsum(equation = var_4095_equation_0, values = (var_4013_cast, var_3983_cast))[name = tensor<string, []>("op_4095_cast")];
            tensor<fp16, []> var_4096_to_fp16 = const()[name = tensor<string, []>("op_4096_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_597_cast = mul(x = var_4095_cast, y = var_4096_to_fp16)[name = tensor<string, []>("aw_chunk_597_cast")];
            tensor<string, []> var_4099_equation_0 = const()[name = tensor<string, []>("op_4099_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_4099_cast = einsum(equation = var_4099_equation_0, values = (var_4013_cast, var_3984_cast))[name = tensor<string, []>("op_4099_cast")];
            tensor<fp16, []> var_4100_to_fp16 = const()[name = tensor<string, []>("op_4100_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_599_cast = mul(x = var_4099_cast, y = var_4100_to_fp16)[name = tensor<string, []>("aw_chunk_599_cast")];
            tensor<string, []> var_4103_equation_0 = const()[name = tensor<string, []>("op_4103_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_4103_cast = einsum(equation = var_4103_equation_0, values = (var_4017_cast, var_3985_cast))[name = tensor<string, []>("op_4103_cast")];
            tensor<fp16, []> var_4104_to_fp16 = const()[name = tensor<string, []>("op_4104_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_601_cast = mul(x = var_4103_cast, y = var_4104_to_fp16)[name = tensor<string, []>("aw_chunk_601_cast")];
            tensor<string, []> var_4107_equation_0 = const()[name = tensor<string, []>("op_4107_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_4107_cast = einsum(equation = var_4107_equation_0, values = (var_4017_cast, var_3986_cast))[name = tensor<string, []>("op_4107_cast")];
            tensor<fp16, []> var_4108_to_fp16 = const()[name = tensor<string, []>("op_4108_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_603_cast = mul(x = var_4107_cast, y = var_4108_to_fp16)[name = tensor<string, []>("aw_chunk_603_cast")];
            tensor<string, []> var_4111_equation_0 = const()[name = tensor<string, []>("op_4111_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_4111_cast = einsum(equation = var_4111_equation_0, values = (var_4021_cast, var_3987_cast))[name = tensor<string, []>("op_4111_cast")];
            tensor<fp16, []> var_4112_to_fp16 = const()[name = tensor<string, []>("op_4112_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_605_cast = mul(x = var_4111_cast, y = var_4112_to_fp16)[name = tensor<string, []>("aw_chunk_605_cast")];
            tensor<string, []> var_4115_equation_0 = const()[name = tensor<string, []>("op_4115_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_4115_cast = einsum(equation = var_4115_equation_0, values = (var_4021_cast, var_3988_cast))[name = tensor<string, []>("op_4115_cast")];
            tensor<fp16, []> var_4116_to_fp16 = const()[name = tensor<string, []>("op_4116_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_607_cast = mul(x = var_4115_cast, y = var_4116_to_fp16)[name = tensor<string, []>("aw_chunk_607_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_4118_cast = softmax(axis = var_3111, x = aw_chunk_577_cast)[name = tensor<string, []>("op_4118_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_4119_cast = softmax(axis = var_3111, x = aw_chunk_579_cast)[name = tensor<string, []>("op_4119_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_4120_cast = softmax(axis = var_3111, x = aw_chunk_581_cast)[name = tensor<string, []>("op_4120_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_4121_cast = softmax(axis = var_3111, x = aw_chunk_583_cast)[name = tensor<string, []>("op_4121_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_4122_cast = softmax(axis = var_3111, x = aw_chunk_585_cast)[name = tensor<string, []>("op_4122_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_4123_cast = softmax(axis = var_3111, x = aw_chunk_587_cast)[name = tensor<string, []>("op_4123_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_4124_cast = softmax(axis = var_3111, x = aw_chunk_589_cast)[name = tensor<string, []>("op_4124_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_4125_cast = softmax(axis = var_3111, x = aw_chunk_591_cast)[name = tensor<string, []>("op_4125_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_4126_cast = softmax(axis = var_3111, x = aw_chunk_593_cast)[name = tensor<string, []>("op_4126_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_4127_cast = softmax(axis = var_3111, x = aw_chunk_595_cast)[name = tensor<string, []>("op_4127_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_4128_cast = softmax(axis = var_3111, x = aw_chunk_597_cast)[name = tensor<string, []>("op_4128_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_4129_cast = softmax(axis = var_3111, x = aw_chunk_599_cast)[name = tensor<string, []>("op_4129_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_4130_cast = softmax(axis = var_3111, x = aw_chunk_601_cast)[name = tensor<string, []>("op_4130_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_4131_cast = softmax(axis = var_3111, x = aw_chunk_603_cast)[name = tensor<string, []>("op_4131_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_4132_cast = softmax(axis = var_3111, x = aw_chunk_605_cast)[name = tensor<string, []>("op_4132_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_4133_cast = softmax(axis = var_3111, x = aw_chunk_607_cast)[name = tensor<string, []>("op_4133_cast")];
            tensor<string, []> var_4135_equation_0 = const()[name = tensor<string, []>("op_4135_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4135_cast = einsum(equation = var_4135_equation_0, values = (var_4023_cast, var_4118_cast))[name = tensor<string, []>("op_4135_cast")];
            tensor<string, []> var_4137_equation_0 = const()[name = tensor<string, []>("op_4137_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4137_cast = einsum(equation = var_4137_equation_0, values = (var_4023_cast, var_4119_cast))[name = tensor<string, []>("op_4137_cast")];
            tensor<string, []> var_4139_equation_0 = const()[name = tensor<string, []>("op_4139_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4139_cast = einsum(equation = var_4139_equation_0, values = (var_4027_cast, var_4120_cast))[name = tensor<string, []>("op_4139_cast")];
            tensor<string, []> var_4141_equation_0 = const()[name = tensor<string, []>("op_4141_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4141_cast = einsum(equation = var_4141_equation_0, values = (var_4027_cast, var_4121_cast))[name = tensor<string, []>("op_4141_cast")];
            tensor<string, []> var_4143_equation_0 = const()[name = tensor<string, []>("op_4143_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4143_cast = einsum(equation = var_4143_equation_0, values = (var_4031_cast, var_4122_cast))[name = tensor<string, []>("op_4143_cast")];
            tensor<string, []> var_4145_equation_0 = const()[name = tensor<string, []>("op_4145_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4145_cast = einsum(equation = var_4145_equation_0, values = (var_4031_cast, var_4123_cast))[name = tensor<string, []>("op_4145_cast")];
            tensor<string, []> var_4147_equation_0 = const()[name = tensor<string, []>("op_4147_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4147_cast = einsum(equation = var_4147_equation_0, values = (var_4035_cast, var_4124_cast))[name = tensor<string, []>("op_4147_cast")];
            tensor<string, []> var_4149_equation_0 = const()[name = tensor<string, []>("op_4149_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4149_cast = einsum(equation = var_4149_equation_0, values = (var_4035_cast, var_4125_cast))[name = tensor<string, []>("op_4149_cast")];
            tensor<string, []> var_4151_equation_0 = const()[name = tensor<string, []>("op_4151_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4151_cast = einsum(equation = var_4151_equation_0, values = (var_4039_cast, var_4126_cast))[name = tensor<string, []>("op_4151_cast")];
            tensor<string, []> var_4153_equation_0 = const()[name = tensor<string, []>("op_4153_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4153_cast = einsum(equation = var_4153_equation_0, values = (var_4039_cast, var_4127_cast))[name = tensor<string, []>("op_4153_cast")];
            tensor<string, []> var_4155_equation_0 = const()[name = tensor<string, []>("op_4155_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4155_cast = einsum(equation = var_4155_equation_0, values = (var_4043_cast, var_4128_cast))[name = tensor<string, []>("op_4155_cast")];
            tensor<string, []> var_4157_equation_0 = const()[name = tensor<string, []>("op_4157_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4157_cast = einsum(equation = var_4157_equation_0, values = (var_4043_cast, var_4129_cast))[name = tensor<string, []>("op_4157_cast")];
            tensor<string, []> var_4159_equation_0 = const()[name = tensor<string, []>("op_4159_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4159_cast = einsum(equation = var_4159_equation_0, values = (var_4047_cast, var_4130_cast))[name = tensor<string, []>("op_4159_cast")];
            tensor<string, []> var_4161_equation_0 = const()[name = tensor<string, []>("op_4161_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4161_cast = einsum(equation = var_4161_equation_0, values = (var_4047_cast, var_4131_cast))[name = tensor<string, []>("op_4161_cast")];
            tensor<string, []> var_4163_equation_0 = const()[name = tensor<string, []>("op_4163_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4163_cast = einsum(equation = var_4163_equation_0, values = (var_4051_cast, var_4132_cast))[name = tensor<string, []>("op_4163_cast")];
            tensor<string, []> var_4165_equation_0 = const()[name = tensor<string, []>("op_4165_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4165_cast = einsum(equation = var_4165_equation_0, values = (var_4051_cast, var_4133_cast))[name = tensor<string, []>("op_4165_cast")];
            tensor<bool, []> var_4167_interleave_0 = const()[name = tensor<string, []>("op_4167_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_4167_cast = concat(axis = var_3089, interleave = var_4167_interleave_0, values = (var_4135_cast, var_4137_cast))[name = tensor<string, []>("op_4167_cast")];
            tensor<bool, []> var_4169_interleave_0 = const()[name = tensor<string, []>("op_4169_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_4169_cast = concat(axis = var_3089, interleave = var_4169_interleave_0, values = (var_4139_cast, var_4141_cast))[name = tensor<string, []>("op_4169_cast")];
            tensor<bool, []> var_4171_interleave_0 = const()[name = tensor<string, []>("op_4171_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_4171_cast = concat(axis = var_3089, interleave = var_4171_interleave_0, values = (var_4143_cast, var_4145_cast))[name = tensor<string, []>("op_4171_cast")];
            tensor<bool, []> var_4173_interleave_0 = const()[name = tensor<string, []>("op_4173_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_4173_cast = concat(axis = var_3089, interleave = var_4173_interleave_0, values = (var_4147_cast, var_4149_cast))[name = tensor<string, []>("op_4173_cast")];
            tensor<bool, []> var_4175_interleave_0 = const()[name = tensor<string, []>("op_4175_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_4175_cast = concat(axis = var_3089, interleave = var_4175_interleave_0, values = (var_4151_cast, var_4153_cast))[name = tensor<string, []>("op_4175_cast")];
            tensor<bool, []> var_4177_interleave_0 = const()[name = tensor<string, []>("op_4177_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_4177_cast = concat(axis = var_3089, interleave = var_4177_interleave_0, values = (var_4155_cast, var_4157_cast))[name = tensor<string, []>("op_4177_cast")];
            tensor<bool, []> var_4179_interleave_0 = const()[name = tensor<string, []>("op_4179_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_4179_cast = concat(axis = var_3089, interleave = var_4179_interleave_0, values = (var_4159_cast, var_4161_cast))[name = tensor<string, []>("op_4179_cast")];
            tensor<bool, []> var_4181_interleave_0 = const()[name = tensor<string, []>("op_4181_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_4181_cast = concat(axis = var_3089, interleave = var_4181_interleave_0, values = (var_4163_cast, var_4165_cast))[name = tensor<string, []>("op_4181_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_3111, interleave = input_105_interleave_0, values = (var_4167_cast, var_4169_cast, var_4171_cast, var_4173_cast, var_4175_cast, var_4177_cast, var_4179_cast, var_4181_cast))[name = tensor<string, []>("input_105_cast")];
            tensor<int32, [2]> var_4187 = const()[name = tensor<string, []>("op_4187"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4189 = const()[name = tensor<string, []>("op_4189"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_4191_pad_type_0 = const()[name = tensor<string, []>("op_4191_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_4191_pad_0 = const()[name = tensor<string, []>("op_4191_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28699072))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29006336))), name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            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, []>(29006528)))];
            tensor<fp16, [2, 640, 1, 1024]> var_4191_cast = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_4189, groups = var_3111, pad = var_4191_pad_0, pad_type = var_4191_pad_type_0, strides = var_4187, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_105_cast)[name = tensor<string, []>("op_4191_cast")];
            tensor<fp16, [2, 640, 1, 1024]> inputs_21_cast = add(x = var_4191_cast, y = inputs_19_cast)[name = tensor<string, []>("inputs_21_cast")];
            tensor<int32, [1]> var_4195 = const()[name = tensor<string, []>("op_4195"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_21_cast = reduce_mean(axes = var_4195, keep_dims = var_3106, x = inputs_21_cast)[name = tensor<string, []>("channels_mean_21_cast")];
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_21_cast = sub(x = inputs_21_cast, y = channels_mean_21_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_21_cast")];
            tensor<int32, [1]> var_4199 = const()[name = tensor<string, []>("op_4199"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_4200_cast = reduce_mean(axes = var_4199, keep_dims = var_3106, x = zero_mean_sq_21_cast)[name = tensor<string, []>("op_4200_cast")];
            tensor<fp16, []> var_4201_to_fp16 = const()[name = tensor<string, []>("op_4201_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_4202_cast = add(x = var_4200_cast, y = var_4201_to_fp16)[name = tensor<string, []>("op_4202_cast")];
            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_4202_cast)[name = tensor<string, []>("denom_21_cast")];
            tensor<fp16, [2, 640, 1, 1024]> out_21_cast = mul(x = zero_mean_21_cast, y = denom_21_cast)[name = tensor<string, []>("out_21_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_4206_to_fp16 = const()[name = tensor<string, []>("op_4206_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29007872)))];
            tensor<fp16, [2, 640, 1, 1024]> var_4207_cast = add(x = out_21_cast, y = var_4206_to_fp16)[name = tensor<string, []>("op_4207_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_4209_to_fp16 = const()[name = tensor<string, []>("op_4209_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29009216)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_65_cast = mul(x = var_4207_cast, y = var_4209_to_fp16)[name = tensor<string, []>("hidden_states_65_cast")];
            tensor<int32, [2]> var_4216 = const()[name = tensor<string, []>("op_4216"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4218 = const()[name = tensor<string, []>("op_4218"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29010560))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29317824))), name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 1024]> q_15_cast = conv(dilations = var_4218, groups = var_3111, pad = q_15_pad_0, pad_type = q_15_pad_type_0, strides = var_4216, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_65_cast)[name = tensor<string, []>("q_15_cast")];
            tensor<int32, [2]> var_4222 = const()[name = tensor<string, []>("op_4222"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4224 = const()[name = tensor<string, []>("op_4224"), 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, 768, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [368640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29318016))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29686720))), name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 768, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> k_29_cast = conv(dilations = var_4224, groups = var_3111, pad = k_29_pad_0, pad_type = k_29_pad_type_0, strides = var_4222, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_29_cast")];
            tensor<int32, [2]> var_4228 = const()[name = tensor<string, []>("op_4228"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4230 = const()[name = tensor<string, []>("op_4230"), 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, 768, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [368640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29686912))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30055616))), name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 768, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> v_15_cast = conv(dilations = var_4230, groups = var_3111, pad = v_15_pad_0, pad_type = v_15_pad_type_0, strides = var_4228, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_15_cast")];
            tensor<int32, [4]> var_4234_begin_0 = const()[name = tensor<string, []>("op_4234_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4234_end_0 = const()[name = tensor<string, []>("op_4234_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_4234_end_mask_0 = const()[name = tensor<string, []>("op_4234_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_4234_cast = slice_by_index(begin = var_4234_begin_0, end = var_4234_end_0, end_mask = var_4234_end_mask_0, x = q_15_cast)[name = tensor<string, []>("op_4234_cast")];
            tensor<int32, [4]> var_4238_begin_0 = const()[name = tensor<string, []>("op_4238_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_4238_end_0 = const()[name = tensor<string, []>("op_4238_end_0"), val = tensor<int32, [4]>([2, 160, 1, 1024])];
            tensor<bool, [4]> var_4238_end_mask_0 = const()[name = tensor<string, []>("op_4238_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_4238_cast = slice_by_index(begin = var_4238_begin_0, end = var_4238_end_0, end_mask = var_4238_end_mask_0, x = q_15_cast)[name = tensor<string, []>("op_4238_cast")];
            tensor<int32, [4]> var_4242_begin_0 = const()[name = tensor<string, []>("op_4242_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_4242_end_0 = const()[name = tensor<string, []>("op_4242_end_0"), val = tensor<int32, [4]>([2, 240, 1, 1024])];
            tensor<bool, [4]> var_4242_end_mask_0 = const()[name = tensor<string, []>("op_4242_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_4242_cast = slice_by_index(begin = var_4242_begin_0, end = var_4242_end_0, end_mask = var_4242_end_mask_0, x = q_15_cast)[name = tensor<string, []>("op_4242_cast")];
            tensor<int32, [4]> var_4246_begin_0 = const()[name = tensor<string, []>("op_4246_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_4246_end_0 = const()[name = tensor<string, []>("op_4246_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_4246_end_mask_0 = const()[name = tensor<string, []>("op_4246_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_4246_cast = slice_by_index(begin = var_4246_begin_0, end = var_4246_end_0, end_mask = var_4246_end_mask_0, x = q_15_cast)[name = tensor<string, []>("op_4246_cast")];
            tensor<int32, [4]> var_4250_begin_0 = const()[name = tensor<string, []>("op_4250_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_4250_end_0 = const()[name = tensor<string, []>("op_4250_end_0"), val = tensor<int32, [4]>([2, 400, 1, 1024])];
            tensor<bool, [4]> var_4250_end_mask_0 = const()[name = tensor<string, []>("op_4250_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_4250_cast = slice_by_index(begin = var_4250_begin_0, end = var_4250_end_0, end_mask = var_4250_end_mask_0, x = q_15_cast)[name = tensor<string, []>("op_4250_cast")];
            tensor<int32, [4]> var_4254_begin_0 = const()[name = tensor<string, []>("op_4254_begin_0"), val = tensor<int32, [4]>([0, 400, 0, 0])];
            tensor<int32, [4]> var_4254_end_0 = const()[name = tensor<string, []>("op_4254_end_0"), val = tensor<int32, [4]>([2, 480, 1, 1024])];
            tensor<bool, [4]> var_4254_end_mask_0 = const()[name = tensor<string, []>("op_4254_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_4254_cast = slice_by_index(begin = var_4254_begin_0, end = var_4254_end_0, end_mask = var_4254_end_mask_0, x = q_15_cast)[name = tensor<string, []>("op_4254_cast")];
            tensor<int32, [4]> var_4258_begin_0 = const()[name = tensor<string, []>("op_4258_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_4258_end_0 = const()[name = tensor<string, []>("op_4258_end_0"), val = tensor<int32, [4]>([2, 560, 1, 1024])];
            tensor<bool, [4]> var_4258_end_mask_0 = const()[name = tensor<string, []>("op_4258_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_4258_cast = slice_by_index(begin = var_4258_begin_0, end = var_4258_end_0, end_mask = var_4258_end_mask_0, x = q_15_cast)[name = tensor<string, []>("op_4258_cast")];
            tensor<int32, [4]> var_4262_begin_0 = const()[name = tensor<string, []>("op_4262_begin_0"), val = tensor<int32, [4]>([0, 560, 0, 0])];
            tensor<int32, [4]> var_4262_end_0 = const()[name = tensor<string, []>("op_4262_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_4262_end_mask_0 = const()[name = tensor<string, []>("op_4262_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_4262_cast = slice_by_index(begin = var_4262_begin_0, end = var_4262_end_0, end_mask = var_4262_end_mask_0, x = q_15_cast)[name = tensor<string, []>("op_4262_cast")];
            tensor<int32, [4]> var_4265_begin_0 = const()[name = tensor<string, []>("op_4265_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4265_end_0 = const()[name = tensor<string, []>("op_4265_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_4265_end_mask_0 = const()[name = tensor<string, []>("op_4265_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_4265_cast = slice_by_index(begin = var_4265_begin_0, end = var_4265_end_0, end_mask = var_4265_end_mask_0, x = var_4234_cast)[name = tensor<string, []>("op_4265_cast")];
            tensor<int32, [4]> var_4266_begin_0 = const()[name = tensor<string, []>("op_4266_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_4266_end_0 = const()[name = tensor<string, []>("op_4266_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_4266_end_mask_0 = const()[name = tensor<string, []>("op_4266_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_4266_cast = slice_by_index(begin = var_4266_begin_0, end = var_4266_end_0, end_mask = var_4266_end_mask_0, x = var_4234_cast)[name = tensor<string, []>("op_4266_cast")];
            tensor<int32, [4]> var_4267_begin_0 = const()[name = tensor<string, []>("op_4267_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4267_end_0 = const()[name = tensor<string, []>("op_4267_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_4267_end_mask_0 = const()[name = tensor<string, []>("op_4267_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_4267_cast = slice_by_index(begin = var_4267_begin_0, end = var_4267_end_0, end_mask = var_4267_end_mask_0, x = var_4238_cast)[name = tensor<string, []>("op_4267_cast")];
            tensor<int32, [4]> var_4268_begin_0 = const()[name = tensor<string, []>("op_4268_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_4268_end_0 = const()[name = tensor<string, []>("op_4268_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_4268_end_mask_0 = const()[name = tensor<string, []>("op_4268_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_4268_cast = slice_by_index(begin = var_4268_begin_0, end = var_4268_end_0, end_mask = var_4268_end_mask_0, x = var_4238_cast)[name = tensor<string, []>("op_4268_cast")];
            tensor<int32, [4]> var_4269_begin_0 = const()[name = tensor<string, []>("op_4269_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4269_end_0 = const()[name = tensor<string, []>("op_4269_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_4269_end_mask_0 = const()[name = tensor<string, []>("op_4269_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_4269_cast = slice_by_index(begin = var_4269_begin_0, end = var_4269_end_0, end_mask = var_4269_end_mask_0, x = var_4242_cast)[name = tensor<string, []>("op_4269_cast")];
            tensor<int32, [4]> var_4270_begin_0 = const()[name = tensor<string, []>("op_4270_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_4270_end_0 = const()[name = tensor<string, []>("op_4270_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_4270_end_mask_0 = const()[name = tensor<string, []>("op_4270_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_4270_cast = slice_by_index(begin = var_4270_begin_0, end = var_4270_end_0, end_mask = var_4270_end_mask_0, x = var_4242_cast)[name = tensor<string, []>("op_4270_cast")];
            tensor<int32, [4]> var_4271_begin_0 = const()[name = tensor<string, []>("op_4271_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4271_end_0 = const()[name = tensor<string, []>("op_4271_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_4271_end_mask_0 = const()[name = tensor<string, []>("op_4271_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_4271_cast = slice_by_index(begin = var_4271_begin_0, end = var_4271_end_0, end_mask = var_4271_end_mask_0, x = var_4246_cast)[name = tensor<string, []>("op_4271_cast")];
            tensor<int32, [4]> var_4272_begin_0 = const()[name = tensor<string, []>("op_4272_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_4272_end_0 = const()[name = tensor<string, []>("op_4272_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_4272_end_mask_0 = const()[name = tensor<string, []>("op_4272_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_4272_cast = slice_by_index(begin = var_4272_begin_0, end = var_4272_end_0, end_mask = var_4272_end_mask_0, x = var_4246_cast)[name = tensor<string, []>("op_4272_cast")];
            tensor<int32, [4]> var_4273_begin_0 = const()[name = tensor<string, []>("op_4273_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4273_end_0 = const()[name = tensor<string, []>("op_4273_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_4273_end_mask_0 = const()[name = tensor<string, []>("op_4273_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_4273_cast = slice_by_index(begin = var_4273_begin_0, end = var_4273_end_0, end_mask = var_4273_end_mask_0, x = var_4250_cast)[name = tensor<string, []>("op_4273_cast")];
            tensor<int32, [4]> var_4274_begin_0 = const()[name = tensor<string, []>("op_4274_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_4274_end_0 = const()[name = tensor<string, []>("op_4274_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_4274_end_mask_0 = const()[name = tensor<string, []>("op_4274_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_4274_cast = slice_by_index(begin = var_4274_begin_0, end = var_4274_end_0, end_mask = var_4274_end_mask_0, x = var_4250_cast)[name = tensor<string, []>("op_4274_cast")];
            tensor<int32, [4]> var_4275_begin_0 = const()[name = tensor<string, []>("op_4275_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4275_end_0 = const()[name = tensor<string, []>("op_4275_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_4275_end_mask_0 = const()[name = tensor<string, []>("op_4275_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_4275_cast = slice_by_index(begin = var_4275_begin_0, end = var_4275_end_0, end_mask = var_4275_end_mask_0, x = var_4254_cast)[name = tensor<string, []>("op_4275_cast")];
            tensor<int32, [4]> var_4276_begin_0 = const()[name = tensor<string, []>("op_4276_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_4276_end_0 = const()[name = tensor<string, []>("op_4276_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_4276_end_mask_0 = const()[name = tensor<string, []>("op_4276_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_4276_cast = slice_by_index(begin = var_4276_begin_0, end = var_4276_end_0, end_mask = var_4276_end_mask_0, x = var_4254_cast)[name = tensor<string, []>("op_4276_cast")];
            tensor<int32, [4]> var_4277_begin_0 = const()[name = tensor<string, []>("op_4277_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4277_end_0 = const()[name = tensor<string, []>("op_4277_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_4277_end_mask_0 = const()[name = tensor<string, []>("op_4277_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_4277_cast = slice_by_index(begin = var_4277_begin_0, end = var_4277_end_0, end_mask = var_4277_end_mask_0, x = var_4258_cast)[name = tensor<string, []>("op_4277_cast")];
            tensor<int32, [4]> var_4278_begin_0 = const()[name = tensor<string, []>("op_4278_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_4278_end_0 = const()[name = tensor<string, []>("op_4278_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_4278_end_mask_0 = const()[name = tensor<string, []>("op_4278_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_4278_cast = slice_by_index(begin = var_4278_begin_0, end = var_4278_end_0, end_mask = var_4278_end_mask_0, x = var_4258_cast)[name = tensor<string, []>("op_4278_cast")];
            tensor<int32, [4]> var_4279_begin_0 = const()[name = tensor<string, []>("op_4279_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4279_end_0 = const()[name = tensor<string, []>("op_4279_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_4279_end_mask_0 = const()[name = tensor<string, []>("op_4279_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_4279_cast = slice_by_index(begin = var_4279_begin_0, end = var_4279_end_0, end_mask = var_4279_end_mask_0, x = var_4262_cast)[name = tensor<string, []>("op_4279_cast")];
            tensor<int32, [4]> var_4280_begin_0 = const()[name = tensor<string, []>("op_4280_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_4280_end_0 = const()[name = tensor<string, []>("op_4280_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_4280_end_mask_0 = const()[name = tensor<string, []>("op_4280_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_4280_cast = slice_by_index(begin = var_4280_begin_0, end = var_4280_end_0, end_mask = var_4280_end_mask_0, x = var_4262_cast)[name = tensor<string, []>("op_4280_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_4285_begin_0 = const()[name = tensor<string, []>("op_4285_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4285_end_0 = const()[name = tensor<string, []>("op_4285_end_0"), val = tensor<int32, [4]>([2, 77, 1, 80])];
            tensor<bool, [4]> var_4285_end_mask_0 = const()[name = tensor<string, []>("op_4285_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)[name = tensor<string, []>("transpose_24")];
            tensor<fp16, [2, 77, 1, 80]> var_4285_cast = slice_by_index(begin = var_4285_begin_0, end = var_4285_end_0, end_mask = var_4285_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_4285_cast")];
            tensor<int32, [4]> var_4289_begin_0 = const()[name = tensor<string, []>("op_4289_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 80])];
            tensor<int32, [4]> var_4289_end_0 = const()[name = tensor<string, []>("op_4289_end_0"), val = tensor<int32, [4]>([2, 77, 1, 160])];
            tensor<bool, [4]> var_4289_end_mask_0 = const()[name = tensor<string, []>("op_4289_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_4289_cast = slice_by_index(begin = var_4289_begin_0, end = var_4289_end_0, end_mask = var_4289_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_4289_cast")];
            tensor<int32, [4]> var_4293_begin_0 = const()[name = tensor<string, []>("op_4293_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_4293_end_0 = const()[name = tensor<string, []>("op_4293_end_0"), val = tensor<int32, [4]>([2, 77, 1, 240])];
            tensor<bool, [4]> var_4293_end_mask_0 = const()[name = tensor<string, []>("op_4293_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_4293_cast = slice_by_index(begin = var_4293_begin_0, end = var_4293_end_0, end_mask = var_4293_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_4293_cast")];
            tensor<int32, [4]> var_4297_begin_0 = const()[name = tensor<string, []>("op_4297_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 240])];
            tensor<int32, [4]> var_4297_end_0 = const()[name = tensor<string, []>("op_4297_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_4297_end_mask_0 = const()[name = tensor<string, []>("op_4297_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_4297_cast = slice_by_index(begin = var_4297_begin_0, end = var_4297_end_0, end_mask = var_4297_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_4297_cast")];
            tensor<int32, [4]> var_4301_begin_0 = const()[name = tensor<string, []>("op_4301_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_4301_end_0 = const()[name = tensor<string, []>("op_4301_end_0"), val = tensor<int32, [4]>([2, 77, 1, 400])];
            tensor<bool, [4]> var_4301_end_mask_0 = const()[name = tensor<string, []>("op_4301_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_4301_cast = slice_by_index(begin = var_4301_begin_0, end = var_4301_end_0, end_mask = var_4301_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_4301_cast")];
            tensor<int32, [4]> var_4305_begin_0 = const()[name = tensor<string, []>("op_4305_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 400])];
            tensor<int32, [4]> var_4305_end_0 = const()[name = tensor<string, []>("op_4305_end_0"), val = tensor<int32, [4]>([2, 77, 1, 480])];
            tensor<bool, [4]> var_4305_end_mask_0 = const()[name = tensor<string, []>("op_4305_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_4305_cast = slice_by_index(begin = var_4305_begin_0, end = var_4305_end_0, end_mask = var_4305_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_4305_cast")];
            tensor<int32, [4]> var_4309_begin_0 = const()[name = tensor<string, []>("op_4309_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 480])];
            tensor<int32, [4]> var_4309_end_0 = const()[name = tensor<string, []>("op_4309_end_0"), val = tensor<int32, [4]>([2, 77, 1, 560])];
            tensor<bool, [4]> var_4309_end_mask_0 = const()[name = tensor<string, []>("op_4309_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_4309_cast = slice_by_index(begin = var_4309_begin_0, end = var_4309_end_0, end_mask = var_4309_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_4309_cast")];
            tensor<int32, [4]> var_4313_begin_0 = const()[name = tensor<string, []>("op_4313_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 560])];
            tensor<int32, [4]> var_4313_end_0 = const()[name = tensor<string, []>("op_4313_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_4313_end_mask_0 = const()[name = tensor<string, []>("op_4313_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_4313_cast = slice_by_index(begin = var_4313_begin_0, end = var_4313_end_0, end_mask = var_4313_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_4313_cast")];
            tensor<int32, [4]> var_4315_begin_0 = const()[name = tensor<string, []>("op_4315_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4315_end_0 = const()[name = tensor<string, []>("op_4315_end_0"), val = tensor<int32, [4]>([2, 80, 1, 77])];
            tensor<bool, [4]> var_4315_end_mask_0 = const()[name = tensor<string, []>("op_4315_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_4315_cast = slice_by_index(begin = var_4315_begin_0, end = var_4315_end_0, end_mask = var_4315_end_mask_0, x = v_15_cast)[name = tensor<string, []>("op_4315_cast")];
            tensor<int32, [4]> var_4319_begin_0 = const()[name = tensor<string, []>("op_4319_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_4319_end_0 = const()[name = tensor<string, []>("op_4319_end_0"), val = tensor<int32, [4]>([2, 160, 1, 77])];
            tensor<bool, [4]> var_4319_end_mask_0 = const()[name = tensor<string, []>("op_4319_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_4319_cast = slice_by_index(begin = var_4319_begin_0, end = var_4319_end_0, end_mask = var_4319_end_mask_0, x = v_15_cast)[name = tensor<string, []>("op_4319_cast")];
            tensor<int32, [4]> var_4323_begin_0 = const()[name = tensor<string, []>("op_4323_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_4323_end_0 = const()[name = tensor<string, []>("op_4323_end_0"), val = tensor<int32, [4]>([2, 240, 1, 77])];
            tensor<bool, [4]> var_4323_end_mask_0 = const()[name = tensor<string, []>("op_4323_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_4323_cast = slice_by_index(begin = var_4323_begin_0, end = var_4323_end_0, end_mask = var_4323_end_mask_0, x = v_15_cast)[name = tensor<string, []>("op_4323_cast")];
            tensor<int32, [4]> var_4327_begin_0 = const()[name = tensor<string, []>("op_4327_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_4327_end_0 = const()[name = tensor<string, []>("op_4327_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_4327_end_mask_0 = const()[name = tensor<string, []>("op_4327_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_4327_cast = slice_by_index(begin = var_4327_begin_0, end = var_4327_end_0, end_mask = var_4327_end_mask_0, x = v_15_cast)[name = tensor<string, []>("op_4327_cast")];
            tensor<int32, [4]> var_4331_begin_0 = const()[name = tensor<string, []>("op_4331_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_4331_end_0 = const()[name = tensor<string, []>("op_4331_end_0"), val = tensor<int32, [4]>([2, 400, 1, 77])];
            tensor<bool, [4]> var_4331_end_mask_0 = const()[name = tensor<string, []>("op_4331_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_4331_cast = slice_by_index(begin = var_4331_begin_0, end = var_4331_end_0, end_mask = var_4331_end_mask_0, x = v_15_cast)[name = tensor<string, []>("op_4331_cast")];
            tensor<int32, [4]> var_4335_begin_0 = const()[name = tensor<string, []>("op_4335_begin_0"), val = tensor<int32, [4]>([0, 400, 0, 0])];
            tensor<int32, [4]> var_4335_end_0 = const()[name = tensor<string, []>("op_4335_end_0"), val = tensor<int32, [4]>([2, 480, 1, 77])];
            tensor<bool, [4]> var_4335_end_mask_0 = const()[name = tensor<string, []>("op_4335_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_4335_cast = slice_by_index(begin = var_4335_begin_0, end = var_4335_end_0, end_mask = var_4335_end_mask_0, x = v_15_cast)[name = tensor<string, []>("op_4335_cast")];
            tensor<int32, [4]> var_4339_begin_0 = const()[name = tensor<string, []>("op_4339_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_4339_end_0 = const()[name = tensor<string, []>("op_4339_end_0"), val = tensor<int32, [4]>([2, 560, 1, 77])];
            tensor<bool, [4]> var_4339_end_mask_0 = const()[name = tensor<string, []>("op_4339_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_4339_cast = slice_by_index(begin = var_4339_begin_0, end = var_4339_end_0, end_mask = var_4339_end_mask_0, x = v_15_cast)[name = tensor<string, []>("op_4339_cast")];
            tensor<int32, [4]> var_4343_begin_0 = const()[name = tensor<string, []>("op_4343_begin_0"), val = tensor<int32, [4]>([0, 560, 0, 0])];
            tensor<int32, [4]> var_4343_end_0 = const()[name = tensor<string, []>("op_4343_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_4343_end_mask_0 = const()[name = tensor<string, []>("op_4343_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_4343_cast = slice_by_index(begin = var_4343_begin_0, end = var_4343_end_0, end_mask = var_4343_end_mask_0, x = v_15_cast)[name = tensor<string, []>("op_4343_cast")];
            tensor<string, []> var_4347_equation_0 = const()[name = tensor<string, []>("op_4347_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_4347_cast = einsum(equation = var_4347_equation_0, values = (var_4285_cast, var_4265_cast))[name = tensor<string, []>("op_4347_cast")];
            tensor<fp16, []> var_4348_to_fp16 = const()[name = tensor<string, []>("op_4348_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_609_cast = mul(x = var_4347_cast, y = var_4348_to_fp16)[name = tensor<string, []>("aw_chunk_609_cast")];
            tensor<string, []> var_4351_equation_0 = const()[name = tensor<string, []>("op_4351_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_4351_cast = einsum(equation = var_4351_equation_0, values = (var_4285_cast, var_4266_cast))[name = tensor<string, []>("op_4351_cast")];
            tensor<fp16, []> var_4352_to_fp16 = const()[name = tensor<string, []>("op_4352_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_611_cast = mul(x = var_4351_cast, y = var_4352_to_fp16)[name = tensor<string, []>("aw_chunk_611_cast")];
            tensor<string, []> var_4355_equation_0 = const()[name = tensor<string, []>("op_4355_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_4355_cast = einsum(equation = var_4355_equation_0, values = (var_4289_cast, var_4267_cast))[name = tensor<string, []>("op_4355_cast")];
            tensor<fp16, []> var_4356_to_fp16 = const()[name = tensor<string, []>("op_4356_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_613_cast = mul(x = var_4355_cast, y = var_4356_to_fp16)[name = tensor<string, []>("aw_chunk_613_cast")];
            tensor<string, []> var_4359_equation_0 = const()[name = tensor<string, []>("op_4359_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_4359_cast = einsum(equation = var_4359_equation_0, values = (var_4289_cast, var_4268_cast))[name = tensor<string, []>("op_4359_cast")];
            tensor<fp16, []> var_4360_to_fp16 = const()[name = tensor<string, []>("op_4360_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_615_cast = mul(x = var_4359_cast, y = var_4360_to_fp16)[name = tensor<string, []>("aw_chunk_615_cast")];
            tensor<string, []> var_4363_equation_0 = const()[name = tensor<string, []>("op_4363_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_4363_cast = einsum(equation = var_4363_equation_0, values = (var_4293_cast, var_4269_cast))[name = tensor<string, []>("op_4363_cast")];
            tensor<fp16, []> var_4364_to_fp16 = const()[name = tensor<string, []>("op_4364_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_617_cast = mul(x = var_4363_cast, y = var_4364_to_fp16)[name = tensor<string, []>("aw_chunk_617_cast")];
            tensor<string, []> var_4367_equation_0 = const()[name = tensor<string, []>("op_4367_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_4367_cast = einsum(equation = var_4367_equation_0, values = (var_4293_cast, var_4270_cast))[name = tensor<string, []>("op_4367_cast")];
            tensor<fp16, []> var_4368_to_fp16 = const()[name = tensor<string, []>("op_4368_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_619_cast = mul(x = var_4367_cast, y = var_4368_to_fp16)[name = tensor<string, []>("aw_chunk_619_cast")];
            tensor<string, []> var_4371_equation_0 = const()[name = tensor<string, []>("op_4371_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_4371_cast = einsum(equation = var_4371_equation_0, values = (var_4297_cast, var_4271_cast))[name = tensor<string, []>("op_4371_cast")];
            tensor<fp16, []> var_4372_to_fp16 = const()[name = tensor<string, []>("op_4372_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_621_cast = mul(x = var_4371_cast, y = var_4372_to_fp16)[name = tensor<string, []>("aw_chunk_621_cast")];
            tensor<string, []> var_4375_equation_0 = const()[name = tensor<string, []>("op_4375_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_4375_cast = einsum(equation = var_4375_equation_0, values = (var_4297_cast, var_4272_cast))[name = tensor<string, []>("op_4375_cast")];
            tensor<fp16, []> var_4376_to_fp16 = const()[name = tensor<string, []>("op_4376_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_623_cast = mul(x = var_4375_cast, y = var_4376_to_fp16)[name = tensor<string, []>("aw_chunk_623_cast")];
            tensor<string, []> var_4379_equation_0 = const()[name = tensor<string, []>("op_4379_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_4379_cast = einsum(equation = var_4379_equation_0, values = (var_4301_cast, var_4273_cast))[name = tensor<string, []>("op_4379_cast")];
            tensor<fp16, []> var_4380_to_fp16 = const()[name = tensor<string, []>("op_4380_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_625_cast = mul(x = var_4379_cast, y = var_4380_to_fp16)[name = tensor<string, []>("aw_chunk_625_cast")];
            tensor<string, []> var_4383_equation_0 = const()[name = tensor<string, []>("op_4383_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_4383_cast = einsum(equation = var_4383_equation_0, values = (var_4301_cast, var_4274_cast))[name = tensor<string, []>("op_4383_cast")];
            tensor<fp16, []> var_4384_to_fp16 = const()[name = tensor<string, []>("op_4384_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_627_cast = mul(x = var_4383_cast, y = var_4384_to_fp16)[name = tensor<string, []>("aw_chunk_627_cast")];
            tensor<string, []> var_4387_equation_0 = const()[name = tensor<string, []>("op_4387_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_4387_cast = einsum(equation = var_4387_equation_0, values = (var_4305_cast, var_4275_cast))[name = tensor<string, []>("op_4387_cast")];
            tensor<fp16, []> var_4388_to_fp16 = const()[name = tensor<string, []>("op_4388_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_629_cast = mul(x = var_4387_cast, y = var_4388_to_fp16)[name = tensor<string, []>("aw_chunk_629_cast")];
            tensor<string, []> var_4391_equation_0 = const()[name = tensor<string, []>("op_4391_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_4391_cast = einsum(equation = var_4391_equation_0, values = (var_4305_cast, var_4276_cast))[name = tensor<string, []>("op_4391_cast")];
            tensor<fp16, []> var_4392_to_fp16 = const()[name = tensor<string, []>("op_4392_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_631_cast = mul(x = var_4391_cast, y = var_4392_to_fp16)[name = tensor<string, []>("aw_chunk_631_cast")];
            tensor<string, []> var_4395_equation_0 = const()[name = tensor<string, []>("op_4395_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_4395_cast = einsum(equation = var_4395_equation_0, values = (var_4309_cast, var_4277_cast))[name = tensor<string, []>("op_4395_cast")];
            tensor<fp16, []> var_4396_to_fp16 = const()[name = tensor<string, []>("op_4396_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_633_cast = mul(x = var_4395_cast, y = var_4396_to_fp16)[name = tensor<string, []>("aw_chunk_633_cast")];
            tensor<string, []> var_4399_equation_0 = const()[name = tensor<string, []>("op_4399_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_4399_cast = einsum(equation = var_4399_equation_0, values = (var_4309_cast, var_4278_cast))[name = tensor<string, []>("op_4399_cast")];
            tensor<fp16, []> var_4400_to_fp16 = const()[name = tensor<string, []>("op_4400_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_635_cast = mul(x = var_4399_cast, y = var_4400_to_fp16)[name = tensor<string, []>("aw_chunk_635_cast")];
            tensor<string, []> var_4403_equation_0 = const()[name = tensor<string, []>("op_4403_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_4403_cast = einsum(equation = var_4403_equation_0, values = (var_4313_cast, var_4279_cast))[name = tensor<string, []>("op_4403_cast")];
            tensor<fp16, []> var_4404_to_fp16 = const()[name = tensor<string, []>("op_4404_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_637_cast = mul(x = var_4403_cast, y = var_4404_to_fp16)[name = tensor<string, []>("aw_chunk_637_cast")];
            tensor<string, []> var_4407_equation_0 = const()[name = tensor<string, []>("op_4407_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_4407_cast = einsum(equation = var_4407_equation_0, values = (var_4313_cast, var_4280_cast))[name = tensor<string, []>("op_4407_cast")];
            tensor<fp16, []> var_4408_to_fp16 = const()[name = tensor<string, []>("op_4408_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_639_cast = mul(x = var_4407_cast, y = var_4408_to_fp16)[name = tensor<string, []>("aw_chunk_639_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_4410_cast = softmax(axis = var_3111, x = aw_chunk_609_cast)[name = tensor<string, []>("op_4410_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_4411_cast = softmax(axis = var_3111, x = aw_chunk_611_cast)[name = tensor<string, []>("op_4411_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_4412_cast = softmax(axis = var_3111, x = aw_chunk_613_cast)[name = tensor<string, []>("op_4412_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_4413_cast = softmax(axis = var_3111, x = aw_chunk_615_cast)[name = tensor<string, []>("op_4413_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_4414_cast = softmax(axis = var_3111, x = aw_chunk_617_cast)[name = tensor<string, []>("op_4414_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_4415_cast = softmax(axis = var_3111, x = aw_chunk_619_cast)[name = tensor<string, []>("op_4415_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_4416_cast = softmax(axis = var_3111, x = aw_chunk_621_cast)[name = tensor<string, []>("op_4416_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_4417_cast = softmax(axis = var_3111, x = aw_chunk_623_cast)[name = tensor<string, []>("op_4417_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_4418_cast = softmax(axis = var_3111, x = aw_chunk_625_cast)[name = tensor<string, []>("op_4418_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_4419_cast = softmax(axis = var_3111, x = aw_chunk_627_cast)[name = tensor<string, []>("op_4419_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_4420_cast = softmax(axis = var_3111, x = aw_chunk_629_cast)[name = tensor<string, []>("op_4420_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_4421_cast = softmax(axis = var_3111, x = aw_chunk_631_cast)[name = tensor<string, []>("op_4421_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_4422_cast = softmax(axis = var_3111, x = aw_chunk_633_cast)[name = tensor<string, []>("op_4422_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_4423_cast = softmax(axis = var_3111, x = aw_chunk_635_cast)[name = tensor<string, []>("op_4423_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_4424_cast = softmax(axis = var_3111, x = aw_chunk_637_cast)[name = tensor<string, []>("op_4424_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_4425_cast = softmax(axis = var_3111, x = aw_chunk_639_cast)[name = tensor<string, []>("op_4425_cast")];
            tensor<string, []> var_4427_equation_0 = const()[name = tensor<string, []>("op_4427_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4427_cast = einsum(equation = var_4427_equation_0, values = (var_4315_cast, var_4410_cast))[name = tensor<string, []>("op_4427_cast")];
            tensor<string, []> var_4429_equation_0 = const()[name = tensor<string, []>("op_4429_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4429_cast = einsum(equation = var_4429_equation_0, values = (var_4315_cast, var_4411_cast))[name = tensor<string, []>("op_4429_cast")];
            tensor<string, []> var_4431_equation_0 = const()[name = tensor<string, []>("op_4431_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4431_cast = einsum(equation = var_4431_equation_0, values = (var_4319_cast, var_4412_cast))[name = tensor<string, []>("op_4431_cast")];
            tensor<string, []> var_4433_equation_0 = const()[name = tensor<string, []>("op_4433_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4433_cast = einsum(equation = var_4433_equation_0, values = (var_4319_cast, var_4413_cast))[name = tensor<string, []>("op_4433_cast")];
            tensor<string, []> var_4435_equation_0 = const()[name = tensor<string, []>("op_4435_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4435_cast = einsum(equation = var_4435_equation_0, values = (var_4323_cast, var_4414_cast))[name = tensor<string, []>("op_4435_cast")];
            tensor<string, []> var_4437_equation_0 = const()[name = tensor<string, []>("op_4437_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4437_cast = einsum(equation = var_4437_equation_0, values = (var_4323_cast, var_4415_cast))[name = tensor<string, []>("op_4437_cast")];
            tensor<string, []> var_4439_equation_0 = const()[name = tensor<string, []>("op_4439_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4439_cast = einsum(equation = var_4439_equation_0, values = (var_4327_cast, var_4416_cast))[name = tensor<string, []>("op_4439_cast")];
            tensor<string, []> var_4441_equation_0 = const()[name = tensor<string, []>("op_4441_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4441_cast = einsum(equation = var_4441_equation_0, values = (var_4327_cast, var_4417_cast))[name = tensor<string, []>("op_4441_cast")];
            tensor<string, []> var_4443_equation_0 = const()[name = tensor<string, []>("op_4443_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4443_cast = einsum(equation = var_4443_equation_0, values = (var_4331_cast, var_4418_cast))[name = tensor<string, []>("op_4443_cast")];
            tensor<string, []> var_4445_equation_0 = const()[name = tensor<string, []>("op_4445_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4445_cast = einsum(equation = var_4445_equation_0, values = (var_4331_cast, var_4419_cast))[name = tensor<string, []>("op_4445_cast")];
            tensor<string, []> var_4447_equation_0 = const()[name = tensor<string, []>("op_4447_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4447_cast = einsum(equation = var_4447_equation_0, values = (var_4335_cast, var_4420_cast))[name = tensor<string, []>("op_4447_cast")];
            tensor<string, []> var_4449_equation_0 = const()[name = tensor<string, []>("op_4449_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4449_cast = einsum(equation = var_4449_equation_0, values = (var_4335_cast, var_4421_cast))[name = tensor<string, []>("op_4449_cast")];
            tensor<string, []> var_4451_equation_0 = const()[name = tensor<string, []>("op_4451_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4451_cast = einsum(equation = var_4451_equation_0, values = (var_4339_cast, var_4422_cast))[name = tensor<string, []>("op_4451_cast")];
            tensor<string, []> var_4453_equation_0 = const()[name = tensor<string, []>("op_4453_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4453_cast = einsum(equation = var_4453_equation_0, values = (var_4339_cast, var_4423_cast))[name = tensor<string, []>("op_4453_cast")];
            tensor<string, []> var_4455_equation_0 = const()[name = tensor<string, []>("op_4455_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4455_cast = einsum(equation = var_4455_equation_0, values = (var_4343_cast, var_4424_cast))[name = tensor<string, []>("op_4455_cast")];
            tensor<string, []> var_4457_equation_0 = const()[name = tensor<string, []>("op_4457_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_4457_cast = einsum(equation = var_4457_equation_0, values = (var_4343_cast, var_4425_cast))[name = tensor<string, []>("op_4457_cast")];
            tensor<bool, []> var_4459_interleave_0 = const()[name = tensor<string, []>("op_4459_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_4459_cast = concat(axis = var_3089, interleave = var_4459_interleave_0, values = (var_4427_cast, var_4429_cast))[name = tensor<string, []>("op_4459_cast")];
            tensor<bool, []> var_4461_interleave_0 = const()[name = tensor<string, []>("op_4461_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_4461_cast = concat(axis = var_3089, interleave = var_4461_interleave_0, values = (var_4431_cast, var_4433_cast))[name = tensor<string, []>("op_4461_cast")];
            tensor<bool, []> var_4463_interleave_0 = const()[name = tensor<string, []>("op_4463_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_4463_cast = concat(axis = var_3089, interleave = var_4463_interleave_0, values = (var_4435_cast, var_4437_cast))[name = tensor<string, []>("op_4463_cast")];
            tensor<bool, []> var_4465_interleave_0 = const()[name = tensor<string, []>("op_4465_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_4465_cast = concat(axis = var_3089, interleave = var_4465_interleave_0, values = (var_4439_cast, var_4441_cast))[name = tensor<string, []>("op_4465_cast")];
            tensor<bool, []> var_4467_interleave_0 = const()[name = tensor<string, []>("op_4467_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_4467_cast = concat(axis = var_3089, interleave = var_4467_interleave_0, values = (var_4443_cast, var_4445_cast))[name = tensor<string, []>("op_4467_cast")];
            tensor<bool, []> var_4469_interleave_0 = const()[name = tensor<string, []>("op_4469_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_4469_cast = concat(axis = var_3089, interleave = var_4469_interleave_0, values = (var_4447_cast, var_4449_cast))[name = tensor<string, []>("op_4469_cast")];
            tensor<bool, []> var_4471_interleave_0 = const()[name = tensor<string, []>("op_4471_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_4471_cast = concat(axis = var_3089, interleave = var_4471_interleave_0, values = (var_4451_cast, var_4453_cast))[name = tensor<string, []>("op_4471_cast")];
            tensor<bool, []> var_4473_interleave_0 = const()[name = tensor<string, []>("op_4473_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_4473_cast = concat(axis = var_3089, interleave = var_4473_interleave_0, values = (var_4455_cast, var_4457_cast))[name = tensor<string, []>("op_4473_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_3111, interleave = input_107_interleave_0, values = (var_4459_cast, var_4461_cast, var_4463_cast, var_4465_cast, var_4467_cast, var_4469_cast, var_4471_cast, var_4473_cast))[name = tensor<string, []>("input_107_cast")];
            tensor<int32, [2]> var_4479 = const()[name = tensor<string, []>("op_4479"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4481 = const()[name = tensor<string, []>("op_4481"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_4483_pad_type_0 = const()[name = tensor<string, []>("op_4483_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_4483_pad_0 = const()[name = tensor<string, []>("op_4483_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30055808))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30363072))), name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            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, []>(30363264)))];
            tensor<fp16, [2, 640, 1, 1024]> var_4483_cast = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_4481, groups = var_3111, pad = var_4483_pad_0, pad_type = var_4483_pad_type_0, strides = var_4479, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_107_cast)[name = tensor<string, []>("op_4483_cast")];
            tensor<fp16, [2, 640, 1, 1024]> inputs_23_cast = add(x = var_4483_cast, y = inputs_21_cast)[name = tensor<string, []>("inputs_23_cast")];
            tensor<int32, [1]> var_4487 = const()[name = tensor<string, []>("op_4487"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_23_cast = reduce_mean(axes = var_4487, keep_dims = var_3106, x = inputs_23_cast)[name = tensor<string, []>("channels_mean_23_cast")];
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_23_cast = sub(x = inputs_23_cast, y = channels_mean_23_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_23_cast")];
            tensor<int32, [1]> var_4491 = const()[name = tensor<string, []>("op_4491"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_4492_cast = reduce_mean(axes = var_4491, keep_dims = var_3106, x = zero_mean_sq_23_cast)[name = tensor<string, []>("op_4492_cast")];
            tensor<fp16, []> var_4493_to_fp16 = const()[name = tensor<string, []>("op_4493_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_4494_cast = add(x = var_4492_cast, y = var_4493_to_fp16)[name = tensor<string, []>("op_4494_cast")];
            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_4494_cast)[name = tensor<string, []>("denom_23_cast")];
            tensor<fp16, [2, 640, 1, 1024]> out_23_cast = mul(x = zero_mean_23_cast, y = denom_23_cast)[name = tensor<string, []>("out_23_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_4498_to_fp16 = const()[name = tensor<string, []>("op_4498_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30364608)))];
            tensor<fp16, [2, 640, 1, 1024]> var_4499_cast = add(x = out_23_cast, y = var_4498_to_fp16)[name = tensor<string, []>("op_4499_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_4501_to_fp16 = const()[name = tensor<string, []>("op_4501_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30365952)))];
            tensor<fp16, [2, 640, 1, 1024]> input_109_cast = mul(x = var_4499_cast, y = var_4501_to_fp16)[name = tensor<string, []>("input_109_cast")];
            tensor<int32, [2]> var_4509 = const()[name = tensor<string, []>("op_4509"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4511 = const()[name = tensor<string, []>("op_4511"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_4513_pad_type_0 = const()[name = tensor<string, []>("op_4513_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_4513_pad_0 = const()[name = tensor<string, []>("op_4513_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2457600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30367296))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32824960))), name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([5120, 640, 1, 1])];
            tensor<fp16, [5120]> down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3840]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32825152))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32829056))), name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([5120])];
            tensor<fp16, [2, 5120, 1, 1024]> var_4513_cast = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_4511, groups = var_3111, pad = var_4513_pad_0, pad_type = var_4513_pad_type_0, strides = var_4509, weight = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_109_cast)[name = tensor<string, []>("op_4513_cast")];
            tensor<int32, [2]> var_4514_split_sizes_0 = const()[name = tensor<string, []>("op_4514_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_4514_axis_0 = const()[name = tensor<string, []>("op_4514_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 1024]> var_4514_cast_0, tensor<fp16, [2, 2560, 1, 1024]> var_4514_cast_1 = split(axis = var_4514_axis_0, split_sizes = var_4514_split_sizes_0, x = var_4513_cast)[name = tensor<string, []>("op_4514_cast")];
            tensor<string, []> var_4516_mode_0 = const()[name = tensor<string, []>("op_4516_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 1024]> var_4516_cast = gelu(mode = var_4516_mode_0, x = var_4514_cast_1)[name = tensor<string, []>("op_4516_cast")];
            tensor<fp16, [2, 2560, 1, 1024]> input_111_cast = mul(x = var_4514_cast_0, y = var_4516_cast)[name = tensor<string, []>("input_111_cast")];
            tensor<int32, [2]> var_4520 = const()[name = tensor<string, []>("op_4520"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4522 = const()[name = tensor<string, []>("op_4522"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_4524_pad_type_0 = const()[name = tensor<string, []>("op_4524_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_4524_pad_0 = const()[name = tensor<string, []>("op_4524_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32829248))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34058112))), name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 2560, 1, 1])];
            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, []>(34058304)))];
            tensor<fp16, [2, 640, 1, 1024]> var_4524_cast = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_4522, groups = var_3111, pad = var_4524_pad_0, pad_type = var_4524_pad_type_0, strides = var_4520, weight = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_111_cast)[name = tensor<string, []>("op_4524_cast")];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_69_cast = add(x = var_4524_cast, y = inputs_23_cast)[name = tensor<string, []>("hidden_states_69_cast")];
            tensor<int32, [4]> var_4526 = const()[name = tensor<string, []>("op_4526"), val = tensor<int32, [4]>([2, 640, 32, 32])];
            tensor<fp16, [2, 640, 32, 32]> input_113_cast = reshape(shape = var_4526, x = hidden_states_69_cast)[name = tensor<string, []>("input_113_cast")];
            tensor<int32, [2]> var_4530 = const()[name = tensor<string, []>("op_4530"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4532 = const()[name = tensor<string, []>("op_4532"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34059648))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34366912))), name = tensor<string, []>("down_blocks_1_attentions_1_proj_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            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, []>(34367104)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_71_cast = conv(bias = down_blocks_1_attentions_1_proj_out_bias_to_fp16, dilations = var_4532, groups = var_3111, pad = hidden_states_71_pad_0, pad_type = hidden_states_71_pad_type_0, strides = var_4530, weight = down_blocks_1_attentions_1_proj_out_weight_to_fp16_palettized, x = input_113_cast)[name = tensor<string, []>("hidden_states_71_cast")];
            tensor<fp16, [2, 640, 32, 32]> input_115_cast = add(x = hidden_states_71_cast, y = hidden_states_59_cast)[name = tensor<string, []>("input_115_cast")];
            tensor<int32, [2]> var_4539 = const()[name = tensor<string, []>("op_4539"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [2]> var_4541 = const()[name = tensor<string, []>("op_4541"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2764800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34368448))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37133312))), name = tensor<string, []>("down_blocks_1_downsamplers_0_conv_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 3, 3])];
            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, []>(37133504)))];
            tensor<fp16, [2, 640, 16, 16]> input_117_cast = conv(bias = down_blocks_1_downsamplers_0_conv_bias_to_fp16, dilations = var_4541, groups = var_3111, pad = input_117_pad_0, pad_type = input_117_pad_type_0, strides = var_4539, weight = down_blocks_1_downsamplers_0_conv_weight_to_fp16_palettized, x = input_115_cast)[name = tensor<string, []>("input_117_cast")];
            tensor<bool, []> var_4564 = const()[name = tensor<string, []>("op_4564"), val = tensor<bool, []>(true)];
            tensor<int32, []> var_4569 = const()[name = tensor<string, []>("op_4569"), 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)[name = tensor<string, []>("reshape_48_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)[name = tensor<string, []>("reduce_mean_36_cast")];
            tensor<fp16, [2, 32, 20, 16, 16]> sub_24_cast = sub(x = reshape_48_cast, y = reduce_mean_36_cast)[name = tensor<string, []>("sub_24_cast")];
            tensor<fp16, [2, 32, 20, 16, 16]> square_12_cast = square(x = sub_24_cast)[name = tensor<string, []>("square_12_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)[name = tensor<string, []>("reduce_mean_38_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)[name = tensor<string, []>("add_24_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_12_cast = sqrt(x = add_24_cast)[name = tensor<string, []>("sqrt_12_cast")];
            tensor<fp16, [2, 32, 20, 16, 16]> real_div_12_cast = real_div(x = sub_24_cast, y = sqrt_12_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_49_cast")];
            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, []>(37134848)))];
            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, []>(37136192)))];
            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_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_49_cast)[name = tensor<string, []>("add_25_cast")];
            tensor<fp16, [2, 640, 16, 16]> input_121_cast = silu(x = add_25_cast)[name = tensor<string, []>("input_121_cast")];
            tensor<int32, [2]> var_4592 = const()[name = tensor<string, []>("op_4592"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4594 = const()[name = tensor<string, []>("op_4594"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5529600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37137536))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42667200))), name = tensor<string, []>("down_blocks_2_resnets_0_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 640, 3, 3])];
            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, []>(42667392)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_73_cast = conv(bias = down_blocks_2_resnets_0_conv1_bias_to_fp16, dilations = var_4594, groups = var_4569, pad = hidden_states_73_pad_0, pad_type = hidden_states_73_pad_type_0, strides = var_4592, weight = down_blocks_2_resnets_0_conv1_weight_to_fp16_palettized, x = input_121_cast)[name = tensor<string, []>("hidden_states_73_cast")];
            tensor<int32, [2]> var_4600 = const()[name = tensor<string, []>("op_4600"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4602 = const()[name = tensor<string, []>("op_4602"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42670016))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43898880))), name = tensor<string, []>("down_blocks_2_resnets_0_time_emb_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(43899072)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_9_cast = conv(bias = down_blocks_2_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_4602, groups = var_4569, pad = temb_9_pad_0, pad_type = temb_9_pad_type_0, strides = var_4600, weight = down_blocks_2_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor<string, []>("temb_9_cast")];
            tensor<fp16, [2, 1280, 16, 16]> input_125_cast = add(x = hidden_states_73_cast, y = temb_9_cast)[name = tensor<string, []>("input_125_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)[name = tensor<string, []>("reshape_52_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)[name = tensor<string, []>("reduce_mean_39_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> sub_26_cast = sub(x = reshape_52_cast, y = reduce_mean_39_cast)[name = tensor<string, []>("sub_26_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> square_13_cast = square(x = sub_26_cast)[name = tensor<string, []>("square_13_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)[name = tensor<string, []>("reduce_mean_41_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)[name = tensor<string, []>("add_26_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_13_cast = sqrt(x = add_26_cast)[name = tensor<string, []>("sqrt_13_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> real_div_13_cast = real_div(x = sub_26_cast, y = sqrt_13_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_53_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, []>(43901696)))];
            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, []>(43904320)))];
            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, []>(43906944)))];
            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, []>(43909568)))];
            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)[name = tensor<string, []>("add_27_cast")];
            tensor<fp16, [2, 1280, 16, 16]> input_129_cast = silu(x = add_27_cast)[name = tensor<string, []>("input_129_cast")];
            tensor<int32, [2]> var_4612 = const()[name = tensor<string, []>("op_4612"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4614 = const()[name = tensor<string, []>("op_4614"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [11059200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43912192))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54971456))), name = tensor<string, []>("down_blocks_2_resnets_0_conv2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 3, 3])];
            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, []>(54971648)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_75_cast = conv(bias = down_blocks_2_resnets_0_conv2_bias_to_fp16, dilations = var_4614, groups = var_4569, pad = hidden_states_75_pad_0, pad_type = hidden_states_75_pad_type_0, strides = var_4612, weight = down_blocks_2_resnets_0_conv2_weight_to_fp16_palettized, x = input_129_cast)[name = tensor<string, []>("hidden_states_75_cast")];
            tensor<int32, [2]> var_4619 = const()[name = tensor<string, []>("op_4619"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4621 = const()[name = tensor<string, []>("op_4621"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [614400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54974272))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55588736))), name = tensor<string, []>("down_blocks_2_resnets_0_conv_shortcut_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 640, 1, 1])];
            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, []>(55588928)))];
            tensor<fp16, [2, 1280, 16, 16]> x_3_cast = conv(bias = down_blocks_2_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_4621, groups = var_4569, pad = x_3_pad_0, pad_type = x_3_pad_type_0, strides = var_4619, weight = down_blocks_2_resnets_0_conv_shortcut_weight_to_fp16_palettized, x = input_117_cast)[name = tensor<string, []>("x_3_cast")];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_77_cast = add(x = x_3_cast, y = hidden_states_75_cast)[name = tensor<string, []>("hidden_states_77_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)[name = tensor<string, []>("reshape_56_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)[name = tensor<string, []>("reduce_mean_42_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> sub_28_cast = sub(x = reshape_56_cast, y = reduce_mean_42_cast)[name = tensor<string, []>("sub_28_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> square_14_cast = square(x = sub_28_cast)[name = tensor<string, []>("square_14_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)[name = tensor<string, []>("reduce_mean_44_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)[name = tensor<string, []>("add_28_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_14_cast = sqrt(x = add_28_cast)[name = tensor<string, []>("sqrt_14_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> real_div_14_cast = real_div(x = sub_28_cast, y = sqrt_14_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_57_cast")];
            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, []>(55591552)))];
            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, []>(55594176)))];
            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_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_57_cast)[name = tensor<string, []>("add_29_cast")];
            tensor<int32, [2]> var_4641 = const()[name = tensor<string, []>("op_4641"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4643 = const()[name = tensor<string, []>("op_4643"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55596800))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(56825664))), name = tensor<string, []>("down_blocks_2_attentions_0_proj_in_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(56825856)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_79_cast = conv(bias = down_blocks_2_attentions_0_proj_in_bias_to_fp16, dilations = var_4643, groups = var_4569, pad = hidden_states_79_pad_0, pad_type = hidden_states_79_pad_type_0, strides = var_4641, weight = down_blocks_2_attentions_0_proj_in_weight_to_fp16_palettized, x = add_29_cast)[name = tensor<string, []>("hidden_states_79_cast")];
            tensor<int32, [4]> var_4648 = const()[name = tensor<string, []>("op_4648"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<fp16, [2, 1280, 1, 256]> inputs_25_cast = reshape(shape = var_4648, x = hidden_states_79_cast)[name = tensor<string, []>("inputs_25_cast")];
            tensor<int32, [1]> var_4658 = const()[name = tensor<string, []>("op_4658"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_25_cast = reduce_mean(axes = var_4658, keep_dims = var_4564, x = inputs_25_cast)[name = tensor<string, []>("channels_mean_25_cast")];
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_25_cast = sub(x = inputs_25_cast, y = channels_mean_25_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_25_cast")];
            tensor<int32, [1]> var_4662 = const()[name = tensor<string, []>("op_4662"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_4663_cast = reduce_mean(axes = var_4662, keep_dims = var_4564, x = zero_mean_sq_25_cast)[name = tensor<string, []>("op_4663_cast")];
            tensor<fp16, []> var_4664_to_fp16 = const()[name = tensor<string, []>("op_4664_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_4665_cast = add(x = var_4663_cast, y = var_4664_to_fp16)[name = tensor<string, []>("op_4665_cast")];
            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_4665_cast)[name = tensor<string, []>("denom_25_cast")];
            tensor<fp16, [2, 1280, 1, 256]> out_25_cast = mul(x = zero_mean_25_cast, y = denom_25_cast)[name = tensor<string, []>("out_25_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_4669_to_fp16 = const()[name = tensor<string, []>("op_4669_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(56828480)))];
            tensor<fp16, [2, 1280, 1, 256]> var_4670_cast = add(x = out_25_cast, y = var_4669_to_fp16)[name = tensor<string, []>("op_4670_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_4672_to_fp16 = const()[name = tensor<string, []>("op_4672_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(56831104)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_81_cast = mul(x = var_4670_cast, y = var_4672_to_fp16)[name = tensor<string, []>("hidden_states_81_cast")];
            tensor<int32, [2]> var_4679 = const()[name = tensor<string, []>("op_4679"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4681 = const()[name = tensor<string, []>("op_4681"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(56833728))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58062592))), name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 256]> q_17_cast = conv(dilations = var_4681, groups = var_4569, pad = q_17_pad_0, pad_type = q_17_pad_type_0, strides = var_4679, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_81_cast)[name = tensor<string, []>("q_17_cast")];
            tensor<int32, [2]> var_4685 = const()[name = tensor<string, []>("op_4685"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4687 = const()[name = tensor<string, []>("op_4687"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58062784))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59291648))), name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 256]> k_33_cast = conv(dilations = var_4687, groups = var_4569, pad = k_33_pad_0, pad_type = k_33_pad_type_0, strides = var_4685, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_81_cast)[name = tensor<string, []>("k_33_cast")];
            tensor<int32, [2]> var_4691 = const()[name = tensor<string, []>("op_4691"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4693 = const()[name = tensor<string, []>("op_4693"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59291840))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60520704))), name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 256]> v_17_cast = conv(dilations = var_4693, groups = var_4569, pad = v_17_pad_0, pad_type = v_17_pad_type_0, strides = var_4691, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_81_cast)[name = tensor<string, []>("v_17_cast")];
            tensor<int32, [4]> var_4697_begin_0 = const()[name = tensor<string, []>("op_4697_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4697_end_0 = const()[name = tensor<string, []>("op_4697_end_0"), val = tensor<int32, [4]>([2, 160, 1, 256])];
            tensor<bool, [4]> var_4697_end_mask_0 = const()[name = tensor<string, []>("op_4697_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_4697_cast = slice_by_index(begin = var_4697_begin_0, end = var_4697_end_0, end_mask = var_4697_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_4697_cast")];
            tensor<int32, [4]> var_4701_begin_0 = const()[name = tensor<string, []>("op_4701_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_4701_end_0 = const()[name = tensor<string, []>("op_4701_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_4701_end_mask_0 = const()[name = tensor<string, []>("op_4701_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_4701_cast = slice_by_index(begin = var_4701_begin_0, end = var_4701_end_0, end_mask = var_4701_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_4701_cast")];
            tensor<int32, [4]> var_4705_begin_0 = const()[name = tensor<string, []>("op_4705_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_4705_end_0 = const()[name = tensor<string, []>("op_4705_end_0"), val = tensor<int32, [4]>([2, 480, 1, 256])];
            tensor<bool, [4]> var_4705_end_mask_0 = const()[name = tensor<string, []>("op_4705_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_4705_cast = slice_by_index(begin = var_4705_begin_0, end = var_4705_end_0, end_mask = var_4705_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_4705_cast")];
            tensor<int32, [4]> var_4709_begin_0 = const()[name = tensor<string, []>("op_4709_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_4709_end_0 = const()[name = tensor<string, []>("op_4709_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_4709_end_mask_0 = const()[name = tensor<string, []>("op_4709_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_4709_cast = slice_by_index(begin = var_4709_begin_0, end = var_4709_end_0, end_mask = var_4709_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_4709_cast")];
            tensor<int32, [4]> var_4713_begin_0 = const()[name = tensor<string, []>("op_4713_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_4713_end_0 = const()[name = tensor<string, []>("op_4713_end_0"), val = tensor<int32, [4]>([2, 800, 1, 256])];
            tensor<bool, [4]> var_4713_end_mask_0 = const()[name = tensor<string, []>("op_4713_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_4713_cast = slice_by_index(begin = var_4713_begin_0, end = var_4713_end_0, end_mask = var_4713_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_4713_cast")];
            tensor<int32, [4]> var_4717_begin_0 = const()[name = tensor<string, []>("op_4717_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_4717_end_0 = const()[name = tensor<string, []>("op_4717_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_4717_end_mask_0 = const()[name = tensor<string, []>("op_4717_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_4717_cast = slice_by_index(begin = var_4717_begin_0, end = var_4717_end_0, end_mask = var_4717_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_4717_cast")];
            tensor<int32, [4]> var_4721_begin_0 = const()[name = tensor<string, []>("op_4721_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_4721_end_0 = const()[name = tensor<string, []>("op_4721_end_0"), val = tensor<int32, [4]>([2, 1120, 1, 256])];
            tensor<bool, [4]> var_4721_end_mask_0 = const()[name = tensor<string, []>("op_4721_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_4721_cast = slice_by_index(begin = var_4721_begin_0, end = var_4721_end_0, end_mask = var_4721_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_4721_cast")];
            tensor<int32, [4]> var_4725_begin_0 = const()[name = tensor<string, []>("op_4725_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_4725_end_0 = const()[name = tensor<string, []>("op_4725_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_4725_end_mask_0 = const()[name = tensor<string, []>("op_4725_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_4725_cast = slice_by_index(begin = var_4725_begin_0, end = var_4725_end_0, end_mask = var_4725_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_4725_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_4732_begin_0 = const()[name = tensor<string, []>("op_4732_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4732_end_0 = const()[name = tensor<string, []>("op_4732_end_0"), val = tensor<int32, [4]>([2, 256, 1, 160])];
            tensor<bool, [4]> var_4732_end_mask_0 = const()[name = tensor<string, []>("op_4732_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)[name = tensor<string, []>("transpose_23")];
            tensor<fp16, [2, 256, 1, 160]> var_4732_cast = slice_by_index(begin = var_4732_begin_0, end = var_4732_end_0, end_mask = var_4732_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_4732_cast")];
            tensor<int32, [4]> var_4736_begin_0 = const()[name = tensor<string, []>("op_4736_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_4736_end_0 = const()[name = tensor<string, []>("op_4736_end_0"), val = tensor<int32, [4]>([2, 256, 1, 320])];
            tensor<bool, [4]> var_4736_end_mask_0 = const()[name = tensor<string, []>("op_4736_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_4736_cast = slice_by_index(begin = var_4736_begin_0, end = var_4736_end_0, end_mask = var_4736_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_4736_cast")];
            tensor<int32, [4]> var_4740_begin_0 = const()[name = tensor<string, []>("op_4740_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_4740_end_0 = const()[name = tensor<string, []>("op_4740_end_0"), val = tensor<int32, [4]>([2, 256, 1, 480])];
            tensor<bool, [4]> var_4740_end_mask_0 = const()[name = tensor<string, []>("op_4740_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_4740_cast = slice_by_index(begin = var_4740_begin_0, end = var_4740_end_0, end_mask = var_4740_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_4740_cast")];
            tensor<int32, [4]> var_4744_begin_0 = const()[name = tensor<string, []>("op_4744_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 480])];
            tensor<int32, [4]> var_4744_end_0 = const()[name = tensor<string, []>("op_4744_end_0"), val = tensor<int32, [4]>([2, 256, 1, 640])];
            tensor<bool, [4]> var_4744_end_mask_0 = const()[name = tensor<string, []>("op_4744_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_4744_cast = slice_by_index(begin = var_4744_begin_0, end = var_4744_end_0, end_mask = var_4744_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_4744_cast")];
            tensor<int32, [4]> var_4748_begin_0 = const()[name = tensor<string, []>("op_4748_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_4748_end_0 = const()[name = tensor<string, []>("op_4748_end_0"), val = tensor<int32, [4]>([2, 256, 1, 800])];
            tensor<bool, [4]> var_4748_end_mask_0 = const()[name = tensor<string, []>("op_4748_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_4748_cast = slice_by_index(begin = var_4748_begin_0, end = var_4748_end_0, end_mask = var_4748_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_4748_cast")];
            tensor<int32, [4]> var_4752_begin_0 = const()[name = tensor<string, []>("op_4752_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 800])];
            tensor<int32, [4]> var_4752_end_0 = const()[name = tensor<string, []>("op_4752_end_0"), val = tensor<int32, [4]>([2, 256, 1, 960])];
            tensor<bool, [4]> var_4752_end_mask_0 = const()[name = tensor<string, []>("op_4752_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_4752_cast = slice_by_index(begin = var_4752_begin_0, end = var_4752_end_0, end_mask = var_4752_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_4752_cast")];
            tensor<int32, [4]> var_4756_begin_0 = const()[name = tensor<string, []>("op_4756_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_4756_end_0 = const()[name = tensor<string, []>("op_4756_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1120])];
            tensor<bool, [4]> var_4756_end_mask_0 = const()[name = tensor<string, []>("op_4756_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_4756_cast = slice_by_index(begin = var_4756_begin_0, end = var_4756_end_0, end_mask = var_4756_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_4756_cast")];
            tensor<int32, [4]> var_4760_begin_0 = const()[name = tensor<string, []>("op_4760_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1120])];
            tensor<int32, [4]> var_4760_end_0 = const()[name = tensor<string, []>("op_4760_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1280])];
            tensor<bool, [4]> var_4760_end_mask_0 = const()[name = tensor<string, []>("op_4760_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_4760_cast = slice_by_index(begin = var_4760_begin_0, end = var_4760_end_0, end_mask = var_4760_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_4760_cast")];
            tensor<int32, [4]> var_4762_begin_0 = const()[name = tensor<string, []>("op_4762_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4762_end_0 = const()[name = tensor<string, []>("op_4762_end_0"), val = tensor<int32, [4]>([2, 160, 1, 256])];
            tensor<bool, [4]> var_4762_end_mask_0 = const()[name = tensor<string, []>("op_4762_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_4762_cast = slice_by_index(begin = var_4762_begin_0, end = var_4762_end_0, end_mask = var_4762_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_4762_cast")];
            tensor<int32, [4]> var_4766_begin_0 = const()[name = tensor<string, []>("op_4766_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_4766_end_0 = const()[name = tensor<string, []>("op_4766_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_4766_end_mask_0 = const()[name = tensor<string, []>("op_4766_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_4766_cast = slice_by_index(begin = var_4766_begin_0, end = var_4766_end_0, end_mask = var_4766_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_4766_cast")];
            tensor<int32, [4]> var_4770_begin_0 = const()[name = tensor<string, []>("op_4770_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_4770_end_0 = const()[name = tensor<string, []>("op_4770_end_0"), val = tensor<int32, [4]>([2, 480, 1, 256])];
            tensor<bool, [4]> var_4770_end_mask_0 = const()[name = tensor<string, []>("op_4770_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_4770_cast = slice_by_index(begin = var_4770_begin_0, end = var_4770_end_0, end_mask = var_4770_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_4770_cast")];
            tensor<int32, [4]> var_4774_begin_0 = const()[name = tensor<string, []>("op_4774_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_4774_end_0 = const()[name = tensor<string, []>("op_4774_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_4774_end_mask_0 = const()[name = tensor<string, []>("op_4774_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_4774_cast = slice_by_index(begin = var_4774_begin_0, end = var_4774_end_0, end_mask = var_4774_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_4774_cast")];
            tensor<int32, [4]> var_4778_begin_0 = const()[name = tensor<string, []>("op_4778_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_4778_end_0 = const()[name = tensor<string, []>("op_4778_end_0"), val = tensor<int32, [4]>([2, 800, 1, 256])];
            tensor<bool, [4]> var_4778_end_mask_0 = const()[name = tensor<string, []>("op_4778_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_4778_cast = slice_by_index(begin = var_4778_begin_0, end = var_4778_end_0, end_mask = var_4778_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_4778_cast")];
            tensor<int32, [4]> var_4782_begin_0 = const()[name = tensor<string, []>("op_4782_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_4782_end_0 = const()[name = tensor<string, []>("op_4782_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_4782_end_mask_0 = const()[name = tensor<string, []>("op_4782_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_4782_cast = slice_by_index(begin = var_4782_begin_0, end = var_4782_end_0, end_mask = var_4782_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_4782_cast")];
            tensor<int32, [4]> var_4786_begin_0 = const()[name = tensor<string, []>("op_4786_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_4786_end_0 = const()[name = tensor<string, []>("op_4786_end_0"), val = tensor<int32, [4]>([2, 1120, 1, 256])];
            tensor<bool, [4]> var_4786_end_mask_0 = const()[name = tensor<string, []>("op_4786_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_4786_cast = slice_by_index(begin = var_4786_begin_0, end = var_4786_end_0, end_mask = var_4786_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_4786_cast")];
            tensor<int32, [4]> var_4790_begin_0 = const()[name = tensor<string, []>("op_4790_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_4790_end_0 = const()[name = tensor<string, []>("op_4790_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_4790_end_mask_0 = const()[name = tensor<string, []>("op_4790_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_4790_cast = slice_by_index(begin = var_4790_begin_0, end = var_4790_end_0, end_mask = var_4790_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_4790_cast")];
            tensor<string, []> var_4794_equation_0 = const()[name = tensor<string, []>("op_4794_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_4794_cast = einsum(equation = var_4794_equation_0, values = (var_4732_cast, var_4697_cast))[name = tensor<string, []>("op_4794_cast")];
            tensor<fp16, []> var_4795_to_fp16 = const()[name = tensor<string, []>("op_4795_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_1_cast = mul(x = var_4794_cast, y = var_4795_to_fp16)[name = tensor<string, []>("aw_1_cast")];
            tensor<string, []> var_4798_equation_0 = const()[name = tensor<string, []>("op_4798_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_4798_cast = einsum(equation = var_4798_equation_0, values = (var_4736_cast, var_4701_cast))[name = tensor<string, []>("op_4798_cast")];
            tensor<fp16, []> var_4799_to_fp16 = const()[name = tensor<string, []>("op_4799_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_3_cast = mul(x = var_4798_cast, y = var_4799_to_fp16)[name = tensor<string, []>("aw_3_cast")];
            tensor<string, []> var_4802_equation_0 = const()[name = tensor<string, []>("op_4802_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_4802_cast = einsum(equation = var_4802_equation_0, values = (var_4740_cast, var_4705_cast))[name = tensor<string, []>("op_4802_cast")];
            tensor<fp16, []> var_4803_to_fp16 = const()[name = tensor<string, []>("op_4803_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_5_cast = mul(x = var_4802_cast, y = var_4803_to_fp16)[name = tensor<string, []>("aw_5_cast")];
            tensor<string, []> var_4806_equation_0 = const()[name = tensor<string, []>("op_4806_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_4806_cast = einsum(equation = var_4806_equation_0, values = (var_4744_cast, var_4709_cast))[name = tensor<string, []>("op_4806_cast")];
            tensor<fp16, []> var_4807_to_fp16 = const()[name = tensor<string, []>("op_4807_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_7_cast = mul(x = var_4806_cast, y = var_4807_to_fp16)[name = tensor<string, []>("aw_7_cast")];
            tensor<string, []> var_4810_equation_0 = const()[name = tensor<string, []>("op_4810_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_4810_cast = einsum(equation = var_4810_equation_0, values = (var_4748_cast, var_4713_cast))[name = tensor<string, []>("op_4810_cast")];
            tensor<fp16, []> var_4811_to_fp16 = const()[name = tensor<string, []>("op_4811_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_9_cast = mul(x = var_4810_cast, y = var_4811_to_fp16)[name = tensor<string, []>("aw_9_cast")];
            tensor<string, []> var_4814_equation_0 = const()[name = tensor<string, []>("op_4814_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_4814_cast = einsum(equation = var_4814_equation_0, values = (var_4752_cast, var_4717_cast))[name = tensor<string, []>("op_4814_cast")];
            tensor<fp16, []> var_4815_to_fp16 = const()[name = tensor<string, []>("op_4815_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_11_cast = mul(x = var_4814_cast, y = var_4815_to_fp16)[name = tensor<string, []>("aw_11_cast")];
            tensor<string, []> var_4818_equation_0 = const()[name = tensor<string, []>("op_4818_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_4818_cast = einsum(equation = var_4818_equation_0, values = (var_4756_cast, var_4721_cast))[name = tensor<string, []>("op_4818_cast")];
            tensor<fp16, []> var_4819_to_fp16 = const()[name = tensor<string, []>("op_4819_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_13_cast = mul(x = var_4818_cast, y = var_4819_to_fp16)[name = tensor<string, []>("aw_13_cast")];
            tensor<string, []> var_4822_equation_0 = const()[name = tensor<string, []>("op_4822_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_4822_cast = einsum(equation = var_4822_equation_0, values = (var_4760_cast, var_4725_cast))[name = tensor<string, []>("op_4822_cast")];
            tensor<fp16, []> var_4823_to_fp16 = const()[name = tensor<string, []>("op_4823_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_15_cast = mul(x = var_4822_cast, y = var_4823_to_fp16)[name = tensor<string, []>("aw_15_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_4825_cast = softmax(axis = var_4569, x = aw_1_cast)[name = tensor<string, []>("op_4825_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_4826_cast = softmax(axis = var_4569, x = aw_3_cast)[name = tensor<string, []>("op_4826_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_4827_cast = softmax(axis = var_4569, x = aw_5_cast)[name = tensor<string, []>("op_4827_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_4828_cast = softmax(axis = var_4569, x = aw_7_cast)[name = tensor<string, []>("op_4828_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_4829_cast = softmax(axis = var_4569, x = aw_9_cast)[name = tensor<string, []>("op_4829_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_4830_cast = softmax(axis = var_4569, x = aw_11_cast)[name = tensor<string, []>("op_4830_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_4831_cast = softmax(axis = var_4569, x = aw_13_cast)[name = tensor<string, []>("op_4831_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_4832_cast = softmax(axis = var_4569, x = aw_15_cast)[name = tensor<string, []>("op_4832_cast")];
            tensor<string, []> var_4834_equation_0 = const()[name = tensor<string, []>("op_4834_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_4834_cast = einsum(equation = var_4834_equation_0, values = (var_4762_cast, var_4825_cast))[name = tensor<string, []>("op_4834_cast")];
            tensor<string, []> var_4836_equation_0 = const()[name = tensor<string, []>("op_4836_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_4836_cast = einsum(equation = var_4836_equation_0, values = (var_4766_cast, var_4826_cast))[name = tensor<string, []>("op_4836_cast")];
            tensor<string, []> var_4838_equation_0 = const()[name = tensor<string, []>("op_4838_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_4838_cast = einsum(equation = var_4838_equation_0, values = (var_4770_cast, var_4827_cast))[name = tensor<string, []>("op_4838_cast")];
            tensor<string, []> var_4840_equation_0 = const()[name = tensor<string, []>("op_4840_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_4840_cast = einsum(equation = var_4840_equation_0, values = (var_4774_cast, var_4828_cast))[name = tensor<string, []>("op_4840_cast")];
            tensor<string, []> var_4842_equation_0 = const()[name = tensor<string, []>("op_4842_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_4842_cast = einsum(equation = var_4842_equation_0, values = (var_4778_cast, var_4829_cast))[name = tensor<string, []>("op_4842_cast")];
            tensor<string, []> var_4844_equation_0 = const()[name = tensor<string, []>("op_4844_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_4844_cast = einsum(equation = var_4844_equation_0, values = (var_4782_cast, var_4830_cast))[name = tensor<string, []>("op_4844_cast")];
            tensor<string, []> var_4846_equation_0 = const()[name = tensor<string, []>("op_4846_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_4846_cast = einsum(equation = var_4846_equation_0, values = (var_4786_cast, var_4831_cast))[name = tensor<string, []>("op_4846_cast")];
            tensor<string, []> var_4848_equation_0 = const()[name = tensor<string, []>("op_4848_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_4848_cast = einsum(equation = var_4848_equation_0, values = (var_4790_cast, var_4832_cast))[name = tensor<string, []>("op_4848_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_4569, interleave = input_133_interleave_0, values = (var_4834_cast, var_4836_cast, var_4838_cast, var_4840_cast, var_4842_cast, var_4844_cast, var_4846_cast, var_4848_cast))[name = tensor<string, []>("input_133_cast")];
            tensor<int32, [2]> var_4854 = const()[name = tensor<string, []>("op_4854"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4856 = const()[name = tensor<string, []>("op_4856"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_4858_pad_type_0 = const()[name = tensor<string, []>("op_4858_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_4858_pad_0 = const()[name = tensor<string, []>("op_4858_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60520896))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61749760))), name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(61749952)))];
            tensor<fp16, [2, 1280, 1, 256]> var_4858_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_4856, groups = var_4569, pad = var_4858_pad_0, pad_type = var_4858_pad_type_0, strides = var_4854, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_133_cast)[name = tensor<string, []>("op_4858_cast")];
            tensor<fp16, [2, 1280, 1, 256]> inputs_27_cast = add(x = var_4858_cast, y = inputs_25_cast)[name = tensor<string, []>("inputs_27_cast")];
            tensor<int32, [1]> var_4862 = const()[name = tensor<string, []>("op_4862"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_27_cast = reduce_mean(axes = var_4862, keep_dims = var_4564, x = inputs_27_cast)[name = tensor<string, []>("channels_mean_27_cast")];
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_27_cast = sub(x = inputs_27_cast, y = channels_mean_27_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_27_cast")];
            tensor<int32, [1]> var_4866 = const()[name = tensor<string, []>("op_4866"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_4867_cast = reduce_mean(axes = var_4866, keep_dims = var_4564, x = zero_mean_sq_27_cast)[name = tensor<string, []>("op_4867_cast")];
            tensor<fp16, []> var_4868_to_fp16 = const()[name = tensor<string, []>("op_4868_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_4869_cast = add(x = var_4867_cast, y = var_4868_to_fp16)[name = tensor<string, []>("op_4869_cast")];
            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_4869_cast)[name = tensor<string, []>("denom_27_cast")];
            tensor<fp16, [2, 1280, 1, 256]> out_27_cast = mul(x = zero_mean_27_cast, y = denom_27_cast)[name = tensor<string, []>("out_27_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_4873_to_fp16 = const()[name = tensor<string, []>("op_4873_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61752576)))];
            tensor<fp16, [2, 1280, 1, 256]> var_4874_cast = add(x = out_27_cast, y = var_4873_to_fp16)[name = tensor<string, []>("op_4874_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_4876_to_fp16 = const()[name = tensor<string, []>("op_4876_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61755200)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_83_cast = mul(x = var_4874_cast, y = var_4876_to_fp16)[name = tensor<string, []>("hidden_states_83_cast")];
            tensor<int32, [2]> var_4883 = const()[name = tensor<string, []>("op_4883"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4885 = const()[name = tensor<string, []>("op_4885"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61757824))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62986688))), name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 256]> q_19_cast = conv(dilations = var_4885, groups = var_4569, pad = q_19_pad_0, pad_type = q_19_pad_type_0, strides = var_4883, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_83_cast)[name = tensor<string, []>("q_19_cast")];
            tensor<int32, [2]> var_4889 = const()[name = tensor<string, []>("op_4889"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4891 = const()[name = tensor<string, []>("op_4891"), 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, 768, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [737280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62986880))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63724224))), name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 768, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_37_cast = conv(dilations = var_4891, groups = var_4569, pad = k_37_pad_0, pad_type = k_37_pad_type_0, strides = var_4889, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_37_cast")];
            tensor<int32, [2]> var_4895 = const()[name = tensor<string, []>("op_4895"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4897 = const()[name = tensor<string, []>("op_4897"), 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, 768, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [737280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63724416))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64461760))), name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 768, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_19_cast = conv(dilations = var_4897, groups = var_4569, pad = v_19_pad_0, pad_type = v_19_pad_type_0, strides = var_4895, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_19_cast")];
            tensor<int32, [4]> var_4901_begin_0 = const()[name = tensor<string, []>("op_4901_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4901_end_0 = const()[name = tensor<string, []>("op_4901_end_0"), val = tensor<int32, [4]>([2, 160, 1, 256])];
            tensor<bool, [4]> var_4901_end_mask_0 = const()[name = tensor<string, []>("op_4901_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_4901_cast = slice_by_index(begin = var_4901_begin_0, end = var_4901_end_0, end_mask = var_4901_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_4901_cast")];
            tensor<int32, [4]> var_4905_begin_0 = const()[name = tensor<string, []>("op_4905_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_4905_end_0 = const()[name = tensor<string, []>("op_4905_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_4905_end_mask_0 = const()[name = tensor<string, []>("op_4905_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_4905_cast = slice_by_index(begin = var_4905_begin_0, end = var_4905_end_0, end_mask = var_4905_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_4905_cast")];
            tensor<int32, [4]> var_4909_begin_0 = const()[name = tensor<string, []>("op_4909_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_4909_end_0 = const()[name = tensor<string, []>("op_4909_end_0"), val = tensor<int32, [4]>([2, 480, 1, 256])];
            tensor<bool, [4]> var_4909_end_mask_0 = const()[name = tensor<string, []>("op_4909_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_4909_cast = slice_by_index(begin = var_4909_begin_0, end = var_4909_end_0, end_mask = var_4909_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_4909_cast")];
            tensor<int32, [4]> var_4913_begin_0 = const()[name = tensor<string, []>("op_4913_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_4913_end_0 = const()[name = tensor<string, []>("op_4913_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_4913_end_mask_0 = const()[name = tensor<string, []>("op_4913_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_4913_cast = slice_by_index(begin = var_4913_begin_0, end = var_4913_end_0, end_mask = var_4913_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_4913_cast")];
            tensor<int32, [4]> var_4917_begin_0 = const()[name = tensor<string, []>("op_4917_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_4917_end_0 = const()[name = tensor<string, []>("op_4917_end_0"), val = tensor<int32, [4]>([2, 800, 1, 256])];
            tensor<bool, [4]> var_4917_end_mask_0 = const()[name = tensor<string, []>("op_4917_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_4917_cast = slice_by_index(begin = var_4917_begin_0, end = var_4917_end_0, end_mask = var_4917_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_4917_cast")];
            tensor<int32, [4]> var_4921_begin_0 = const()[name = tensor<string, []>("op_4921_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_4921_end_0 = const()[name = tensor<string, []>("op_4921_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_4921_end_mask_0 = const()[name = tensor<string, []>("op_4921_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_4921_cast = slice_by_index(begin = var_4921_begin_0, end = var_4921_end_0, end_mask = var_4921_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_4921_cast")];
            tensor<int32, [4]> var_4925_begin_0 = const()[name = tensor<string, []>("op_4925_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_4925_end_0 = const()[name = tensor<string, []>("op_4925_end_0"), val = tensor<int32, [4]>([2, 1120, 1, 256])];
            tensor<bool, [4]> var_4925_end_mask_0 = const()[name = tensor<string, []>("op_4925_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_4925_cast = slice_by_index(begin = var_4925_begin_0, end = var_4925_end_0, end_mask = var_4925_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_4925_cast")];
            tensor<int32, [4]> var_4929_begin_0 = const()[name = tensor<string, []>("op_4929_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_4929_end_0 = const()[name = tensor<string, []>("op_4929_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_4929_end_mask_0 = const()[name = tensor<string, []>("op_4929_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_4929_cast = slice_by_index(begin = var_4929_begin_0, end = var_4929_end_0, end_mask = var_4929_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_4929_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_4936_begin_0 = const()[name = tensor<string, []>("op_4936_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4936_end_0 = const()[name = tensor<string, []>("op_4936_end_0"), val = tensor<int32, [4]>([2, 77, 1, 160])];
            tensor<bool, [4]> var_4936_end_mask_0 = const()[name = tensor<string, []>("op_4936_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)[name = tensor<string, []>("transpose_22")];
            tensor<fp16, [2, 77, 1, 160]> var_4936_cast = slice_by_index(begin = var_4936_begin_0, end = var_4936_end_0, end_mask = var_4936_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_4936_cast")];
            tensor<int32, [4]> var_4940_begin_0 = const()[name = tensor<string, []>("op_4940_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_4940_end_0 = const()[name = tensor<string, []>("op_4940_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_4940_end_mask_0 = const()[name = tensor<string, []>("op_4940_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_4940_cast = slice_by_index(begin = var_4940_begin_0, end = var_4940_end_0, end_mask = var_4940_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_4940_cast")];
            tensor<int32, [4]> var_4944_begin_0 = const()[name = tensor<string, []>("op_4944_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_4944_end_0 = const()[name = tensor<string, []>("op_4944_end_0"), val = tensor<int32, [4]>([2, 77, 1, 480])];
            tensor<bool, [4]> var_4944_end_mask_0 = const()[name = tensor<string, []>("op_4944_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_4944_cast = slice_by_index(begin = var_4944_begin_0, end = var_4944_end_0, end_mask = var_4944_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_4944_cast")];
            tensor<int32, [4]> var_4948_begin_0 = const()[name = tensor<string, []>("op_4948_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 480])];
            tensor<int32, [4]> var_4948_end_0 = const()[name = tensor<string, []>("op_4948_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_4948_end_mask_0 = const()[name = tensor<string, []>("op_4948_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_4948_cast = slice_by_index(begin = var_4948_begin_0, end = var_4948_end_0, end_mask = var_4948_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_4948_cast")];
            tensor<int32, [4]> var_4952_begin_0 = const()[name = tensor<string, []>("op_4952_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_4952_end_0 = const()[name = tensor<string, []>("op_4952_end_0"), val = tensor<int32, [4]>([2, 77, 1, 800])];
            tensor<bool, [4]> var_4952_end_mask_0 = const()[name = tensor<string, []>("op_4952_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_4952_cast = slice_by_index(begin = var_4952_begin_0, end = var_4952_end_0, end_mask = var_4952_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_4952_cast")];
            tensor<int32, [4]> var_4956_begin_0 = const()[name = tensor<string, []>("op_4956_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 800])];
            tensor<int32, [4]> var_4956_end_0 = const()[name = tensor<string, []>("op_4956_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_4956_end_mask_0 = const()[name = tensor<string, []>("op_4956_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_4956_cast = slice_by_index(begin = var_4956_begin_0, end = var_4956_end_0, end_mask = var_4956_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_4956_cast")];
            tensor<int32, [4]> var_4960_begin_0 = const()[name = tensor<string, []>("op_4960_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_4960_end_0 = const()[name = tensor<string, []>("op_4960_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1120])];
            tensor<bool, [4]> var_4960_end_mask_0 = const()[name = tensor<string, []>("op_4960_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_4960_cast = slice_by_index(begin = var_4960_begin_0, end = var_4960_end_0, end_mask = var_4960_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_4960_cast")];
            tensor<int32, [4]> var_4964_begin_0 = const()[name = tensor<string, []>("op_4964_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1120])];
            tensor<int32, [4]> var_4964_end_0 = const()[name = tensor<string, []>("op_4964_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_4964_end_mask_0 = const()[name = tensor<string, []>("op_4964_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_4964_cast = slice_by_index(begin = var_4964_begin_0, end = var_4964_end_0, end_mask = var_4964_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_4964_cast")];
            tensor<int32, [4]> var_4966_begin_0 = const()[name = tensor<string, []>("op_4966_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4966_end_0 = const()[name = tensor<string, []>("op_4966_end_0"), val = tensor<int32, [4]>([2, 160, 1, 77])];
            tensor<bool, [4]> var_4966_end_mask_0 = const()[name = tensor<string, []>("op_4966_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_4966_cast = slice_by_index(begin = var_4966_begin_0, end = var_4966_end_0, end_mask = var_4966_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_4966_cast")];
            tensor<int32, [4]> var_4970_begin_0 = const()[name = tensor<string, []>("op_4970_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_4970_end_0 = const()[name = tensor<string, []>("op_4970_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_4970_end_mask_0 = const()[name = tensor<string, []>("op_4970_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_4970_cast = slice_by_index(begin = var_4970_begin_0, end = var_4970_end_0, end_mask = var_4970_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_4970_cast")];
            tensor<int32, [4]> var_4974_begin_0 = const()[name = tensor<string, []>("op_4974_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_4974_end_0 = const()[name = tensor<string, []>("op_4974_end_0"), val = tensor<int32, [4]>([2, 480, 1, 77])];
            tensor<bool, [4]> var_4974_end_mask_0 = const()[name = tensor<string, []>("op_4974_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_4974_cast = slice_by_index(begin = var_4974_begin_0, end = var_4974_end_0, end_mask = var_4974_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_4974_cast")];
            tensor<int32, [4]> var_4978_begin_0 = const()[name = tensor<string, []>("op_4978_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_4978_end_0 = const()[name = tensor<string, []>("op_4978_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_4978_end_mask_0 = const()[name = tensor<string, []>("op_4978_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_4978_cast = slice_by_index(begin = var_4978_begin_0, end = var_4978_end_0, end_mask = var_4978_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_4978_cast")];
            tensor<int32, [4]> var_4982_begin_0 = const()[name = tensor<string, []>("op_4982_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_4982_end_0 = const()[name = tensor<string, []>("op_4982_end_0"), val = tensor<int32, [4]>([2, 800, 1, 77])];
            tensor<bool, [4]> var_4982_end_mask_0 = const()[name = tensor<string, []>("op_4982_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_4982_cast = slice_by_index(begin = var_4982_begin_0, end = var_4982_end_0, end_mask = var_4982_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_4982_cast")];
            tensor<int32, [4]> var_4986_begin_0 = const()[name = tensor<string, []>("op_4986_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_4986_end_0 = const()[name = tensor<string, []>("op_4986_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_4986_end_mask_0 = const()[name = tensor<string, []>("op_4986_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_4986_cast = slice_by_index(begin = var_4986_begin_0, end = var_4986_end_0, end_mask = var_4986_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_4986_cast")];
            tensor<int32, [4]> var_4990_begin_0 = const()[name = tensor<string, []>("op_4990_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_4990_end_0 = const()[name = tensor<string, []>("op_4990_end_0"), val = tensor<int32, [4]>([2, 1120, 1, 77])];
            tensor<bool, [4]> var_4990_end_mask_0 = const()[name = tensor<string, []>("op_4990_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_4990_cast = slice_by_index(begin = var_4990_begin_0, end = var_4990_end_0, end_mask = var_4990_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_4990_cast")];
            tensor<int32, [4]> var_4994_begin_0 = const()[name = tensor<string, []>("op_4994_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_4994_end_0 = const()[name = tensor<string, []>("op_4994_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_4994_end_mask_0 = const()[name = tensor<string, []>("op_4994_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_4994_cast = slice_by_index(begin = var_4994_begin_0, end = var_4994_end_0, end_mask = var_4994_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_4994_cast")];
            tensor<string, []> var_4998_equation_0 = const()[name = tensor<string, []>("op_4998_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4998_cast = einsum(equation = var_4998_equation_0, values = (var_4936_cast, var_4901_cast))[name = tensor<string, []>("op_4998_cast")];
            tensor<fp16, []> var_4999_to_fp16 = const()[name = tensor<string, []>("op_4999_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_17_cast = mul(x = var_4998_cast, y = var_4999_to_fp16)[name = tensor<string, []>("aw_17_cast")];
            tensor<string, []> var_5002_equation_0 = const()[name = tensor<string, []>("op_5002_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5002_cast = einsum(equation = var_5002_equation_0, values = (var_4940_cast, var_4905_cast))[name = tensor<string, []>("op_5002_cast")];
            tensor<fp16, []> var_5003_to_fp16 = const()[name = tensor<string, []>("op_5003_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_19_cast = mul(x = var_5002_cast, y = var_5003_to_fp16)[name = tensor<string, []>("aw_19_cast")];
            tensor<string, []> var_5006_equation_0 = const()[name = tensor<string, []>("op_5006_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5006_cast = einsum(equation = var_5006_equation_0, values = (var_4944_cast, var_4909_cast))[name = tensor<string, []>("op_5006_cast")];
            tensor<fp16, []> var_5007_to_fp16 = const()[name = tensor<string, []>("op_5007_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_21_cast = mul(x = var_5006_cast, y = var_5007_to_fp16)[name = tensor<string, []>("aw_21_cast")];
            tensor<string, []> var_5010_equation_0 = const()[name = tensor<string, []>("op_5010_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5010_cast = einsum(equation = var_5010_equation_0, values = (var_4948_cast, var_4913_cast))[name = tensor<string, []>("op_5010_cast")];
            tensor<fp16, []> var_5011_to_fp16 = const()[name = tensor<string, []>("op_5011_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_23_cast = mul(x = var_5010_cast, y = var_5011_to_fp16)[name = tensor<string, []>("aw_23_cast")];
            tensor<string, []> var_5014_equation_0 = const()[name = tensor<string, []>("op_5014_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5014_cast = einsum(equation = var_5014_equation_0, values = (var_4952_cast, var_4917_cast))[name = tensor<string, []>("op_5014_cast")];
            tensor<fp16, []> var_5015_to_fp16 = const()[name = tensor<string, []>("op_5015_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_25_cast = mul(x = var_5014_cast, y = var_5015_to_fp16)[name = tensor<string, []>("aw_25_cast")];
            tensor<string, []> var_5018_equation_0 = const()[name = tensor<string, []>("op_5018_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5018_cast = einsum(equation = var_5018_equation_0, values = (var_4956_cast, var_4921_cast))[name = tensor<string, []>("op_5018_cast")];
            tensor<fp16, []> var_5019_to_fp16 = const()[name = tensor<string, []>("op_5019_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_27_cast = mul(x = var_5018_cast, y = var_5019_to_fp16)[name = tensor<string, []>("aw_27_cast")];
            tensor<string, []> var_5022_equation_0 = const()[name = tensor<string, []>("op_5022_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5022_cast = einsum(equation = var_5022_equation_0, values = (var_4960_cast, var_4925_cast))[name = tensor<string, []>("op_5022_cast")];
            tensor<fp16, []> var_5023_to_fp16 = const()[name = tensor<string, []>("op_5023_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_29_cast = mul(x = var_5022_cast, y = var_5023_to_fp16)[name = tensor<string, []>("aw_29_cast")];
            tensor<string, []> var_5026_equation_0 = const()[name = tensor<string, []>("op_5026_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5026_cast = einsum(equation = var_5026_equation_0, values = (var_4964_cast, var_4929_cast))[name = tensor<string, []>("op_5026_cast")];
            tensor<fp16, []> var_5027_to_fp16 = const()[name = tensor<string, []>("op_5027_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_31_cast = mul(x = var_5026_cast, y = var_5027_to_fp16)[name = tensor<string, []>("aw_31_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5029_cast = softmax(axis = var_4569, x = aw_17_cast)[name = tensor<string, []>("op_5029_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5030_cast = softmax(axis = var_4569, x = aw_19_cast)[name = tensor<string, []>("op_5030_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5031_cast = softmax(axis = var_4569, x = aw_21_cast)[name = tensor<string, []>("op_5031_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5032_cast = softmax(axis = var_4569, x = aw_23_cast)[name = tensor<string, []>("op_5032_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5033_cast = softmax(axis = var_4569, x = aw_25_cast)[name = tensor<string, []>("op_5033_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5034_cast = softmax(axis = var_4569, x = aw_27_cast)[name = tensor<string, []>("op_5034_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5035_cast = softmax(axis = var_4569, x = aw_29_cast)[name = tensor<string, []>("op_5035_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5036_cast = softmax(axis = var_4569, x = aw_31_cast)[name = tensor<string, []>("op_5036_cast")];
            tensor<string, []> var_5038_equation_0 = const()[name = tensor<string, []>("op_5038_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_5038_cast = einsum(equation = var_5038_equation_0, values = (var_4966_cast, var_5029_cast))[name = tensor<string, []>("op_5038_cast")];
            tensor<string, []> var_5040_equation_0 = const()[name = tensor<string, []>("op_5040_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_5040_cast = einsum(equation = var_5040_equation_0, values = (var_4970_cast, var_5030_cast))[name = tensor<string, []>("op_5040_cast")];
            tensor<string, []> var_5042_equation_0 = const()[name = tensor<string, []>("op_5042_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_5042_cast = einsum(equation = var_5042_equation_0, values = (var_4974_cast, var_5031_cast))[name = tensor<string, []>("op_5042_cast")];
            tensor<string, []> var_5044_equation_0 = const()[name = tensor<string, []>("op_5044_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_5044_cast = einsum(equation = var_5044_equation_0, values = (var_4978_cast, var_5032_cast))[name = tensor<string, []>("op_5044_cast")];
            tensor<string, []> var_5046_equation_0 = const()[name = tensor<string, []>("op_5046_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_5046_cast = einsum(equation = var_5046_equation_0, values = (var_4982_cast, var_5033_cast))[name = tensor<string, []>("op_5046_cast")];
            tensor<string, []> var_5048_equation_0 = const()[name = tensor<string, []>("op_5048_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_5048_cast = einsum(equation = var_5048_equation_0, values = (var_4986_cast, var_5034_cast))[name = tensor<string, []>("op_5048_cast")];
            tensor<string, []> var_5050_equation_0 = const()[name = tensor<string, []>("op_5050_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_5050_cast = einsum(equation = var_5050_equation_0, values = (var_4990_cast, var_5035_cast))[name = tensor<string, []>("op_5050_cast")];
            tensor<string, []> var_5052_equation_0 = const()[name = tensor<string, []>("op_5052_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_5052_cast = einsum(equation = var_5052_equation_0, values = (var_4994_cast, var_5036_cast))[name = tensor<string, []>("op_5052_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_4569, interleave = input_135_interleave_0, values = (var_5038_cast, var_5040_cast, var_5042_cast, var_5044_cast, var_5046_cast, var_5048_cast, var_5050_cast, var_5052_cast))[name = tensor<string, []>("input_135_cast")];
            tensor<int32, [2]> var_5058 = const()[name = tensor<string, []>("op_5058"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5060 = const()[name = tensor<string, []>("op_5060"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_5062_pad_type_0 = const()[name = tensor<string, []>("op_5062_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_5062_pad_0 = const()[name = tensor<string, []>("op_5062_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64461952))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65690816))), name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(65691008)))];
            tensor<fp16, [2, 1280, 1, 256]> var_5062_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_5060, groups = var_4569, pad = var_5062_pad_0, pad_type = var_5062_pad_type_0, strides = var_5058, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_135_cast)[name = tensor<string, []>("op_5062_cast")];
            tensor<fp16, [2, 1280, 1, 256]> inputs_29_cast = add(x = var_5062_cast, y = inputs_27_cast)[name = tensor<string, []>("inputs_29_cast")];
            tensor<int32, [1]> var_5066 = const()[name = tensor<string, []>("op_5066"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_29_cast = reduce_mean(axes = var_5066, keep_dims = var_4564, x = inputs_29_cast)[name = tensor<string, []>("channels_mean_29_cast")];
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_29_cast = sub(x = inputs_29_cast, y = channels_mean_29_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_29_cast")];
            tensor<int32, [1]> var_5070 = const()[name = tensor<string, []>("op_5070"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_5071_cast = reduce_mean(axes = var_5070, keep_dims = var_4564, x = zero_mean_sq_29_cast)[name = tensor<string, []>("op_5071_cast")];
            tensor<fp16, []> var_5072_to_fp16 = const()[name = tensor<string, []>("op_5072_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_5073_cast = add(x = var_5071_cast, y = var_5072_to_fp16)[name = tensor<string, []>("op_5073_cast")];
            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_5073_cast)[name = tensor<string, []>("denom_29_cast")];
            tensor<fp16, [2, 1280, 1, 256]> out_29_cast = mul(x = zero_mean_29_cast, y = denom_29_cast)[name = tensor<string, []>("out_29_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_5077_to_fp16 = const()[name = tensor<string, []>("op_5077_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65693632)))];
            tensor<fp16, [2, 1280, 1, 256]> var_5078_cast = add(x = out_29_cast, y = var_5077_to_fp16)[name = tensor<string, []>("op_5078_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_5080_to_fp16 = const()[name = tensor<string, []>("op_5080_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65696256)))];
            tensor<fp16, [2, 1280, 1, 256]> input_137_cast = mul(x = var_5078_cast, y = var_5080_to_fp16)[name = tensor<string, []>("input_137_cast")];
            tensor<int32, [2]> var_5088 = const()[name = tensor<string, []>("op_5088"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5090 = const()[name = tensor<string, []>("op_5090"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_5092_pad_type_0 = const()[name = tensor<string, []>("op_5092_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_5092_pad_0 = const()[name = tensor<string, []>("op_5092_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65698880))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75529344))), name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [7680]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75529536))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75537280))), name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([10240])];
            tensor<fp16, [2, 10240, 1, 256]> var_5092_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_5090, groups = var_4569, pad = var_5092_pad_0, pad_type = var_5092_pad_type_0, strides = var_5088, weight = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_137_cast)[name = tensor<string, []>("op_5092_cast")];
            tensor<int32, [2]> var_5093_split_sizes_0 = const()[name = tensor<string, []>("op_5093_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_5093_axis_0 = const()[name = tensor<string, []>("op_5093_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 256]> var_5093_cast_0, tensor<fp16, [2, 5120, 1, 256]> var_5093_cast_1 = split(axis = var_5093_axis_0, split_sizes = var_5093_split_sizes_0, x = var_5092_cast)[name = tensor<string, []>("op_5093_cast")];
            tensor<string, []> var_5095_mode_0 = const()[name = tensor<string, []>("op_5095_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 256]> var_5095_cast = gelu(mode = var_5095_mode_0, x = var_5093_cast_1)[name = tensor<string, []>("op_5095_cast")];
            tensor<fp16, [2, 5120, 1, 256]> input_139_cast = mul(x = var_5093_cast_0, y = var_5095_cast)[name = tensor<string, []>("input_139_cast")];
            tensor<int32, [2]> var_5099 = const()[name = tensor<string, []>("op_5099"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5101 = const()[name = tensor<string, []>("op_5101"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_5103_pad_type_0 = const()[name = tensor<string, []>("op_5103_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_5103_pad_0 = const()[name = tensor<string, []>("op_5103_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75537472))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80452736))), name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            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, []>(80452928)))];
            tensor<fp16, [2, 1280, 1, 256]> var_5103_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_5101, groups = var_4569, pad = var_5103_pad_0, pad_type = var_5103_pad_type_0, strides = var_5099, weight = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_139_cast)[name = tensor<string, []>("op_5103_cast")];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_87_cast = add(x = var_5103_cast, y = inputs_29_cast)[name = tensor<string, []>("hidden_states_87_cast")];
            tensor<int32, [4]> var_5105 = const()[name = tensor<string, []>("op_5105"), val = tensor<int32, [4]>([2, 1280, 16, 16])];
            tensor<fp16, [2, 1280, 16, 16]> input_141_cast = reshape(shape = var_5105, x = hidden_states_87_cast)[name = tensor<string, []>("input_141_cast")];
            tensor<int32, [2]> var_5109 = const()[name = tensor<string, []>("op_5109"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5111 = const()[name = tensor<string, []>("op_5111"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80455552))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81684416))), name = tensor<string, []>("down_blocks_2_attentions_0_proj_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(81684608)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_89_cast = conv(bias = down_blocks_2_attentions_0_proj_out_bias_to_fp16, dilations = var_5111, groups = var_4569, pad = hidden_states_89_pad_0, pad_type = hidden_states_89_pad_type_0, strides = var_5109, weight = down_blocks_2_attentions_0_proj_out_weight_to_fp16_palettized, x = input_141_cast)[name = tensor<string, []>("hidden_states_89_cast")];
            tensor<fp16, [2, 1280, 16, 16]> input_143_cast = add(x = hidden_states_89_cast, y = hidden_states_77_cast)[name = tensor<string, []>("input_143_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)[name = tensor<string, []>("reshape_60_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)[name = tensor<string, []>("reduce_mean_45_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> sub_30_cast = sub(x = reshape_60_cast, y = reduce_mean_45_cast)[name = tensor<string, []>("sub_30_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> square_15_cast = square(x = sub_30_cast)[name = tensor<string, []>("square_15_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)[name = tensor<string, []>("reduce_mean_47_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)[name = tensor<string, []>("add_30_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_15_cast = sqrt(x = add_30_cast)[name = tensor<string, []>("sqrt_15_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> real_div_15_cast = real_div(x = sub_30_cast, y = sqrt_15_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_61_cast")];
            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, []>(81687232)))];
            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, []>(81689856)))];
            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_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_61_cast)[name = tensor<string, []>("add_31_cast")];
            tensor<fp16, [2, 1280, 16, 16]> input_147_cast = silu(x = add_31_cast)[name = tensor<string, []>("input_147_cast")];
            tensor<int32, [2]> var_5126 = const()[name = tensor<string, []>("op_5126"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5128 = const()[name = tensor<string, []>("op_5128"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [11059200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81692480))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92751744))), name = tensor<string, []>("down_blocks_2_resnets_1_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 3, 3])];
            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, []>(92751936)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_91_cast = conv(bias = down_blocks_2_resnets_1_conv1_bias_to_fp16, dilations = var_5128, groups = var_4569, pad = hidden_states_91_pad_0, pad_type = hidden_states_91_pad_type_0, strides = var_5126, weight = down_blocks_2_resnets_1_conv1_weight_to_fp16_palettized, x = input_147_cast)[name = tensor<string, []>("hidden_states_91_cast")];
            tensor<int32, [2]> var_5134 = const()[name = tensor<string, []>("op_5134"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5136 = const()[name = tensor<string, []>("op_5136"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92754560))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93983424))), name = tensor<string, []>("down_blocks_2_resnets_1_time_emb_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(93983616)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_11_cast = conv(bias = down_blocks_2_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_5136, groups = var_4569, pad = temb_11_pad_0, pad_type = temb_11_pad_type_0, strides = var_5134, weight = down_blocks_2_resnets_1_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor<string, []>("temb_11_cast")];
            tensor<fp16, [2, 1280, 16, 16]> input_151_cast = add(x = hidden_states_91_cast, y = temb_11_cast)[name = tensor<string, []>("input_151_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)[name = tensor<string, []>("reshape_64_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)[name = tensor<string, []>("reduce_mean_48_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> sub_32_cast = sub(x = reshape_64_cast, y = reduce_mean_48_cast)[name = tensor<string, []>("sub_32_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> square_16_cast = square(x = sub_32_cast)[name = tensor<string, []>("square_16_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)[name = tensor<string, []>("reduce_mean_50_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)[name = tensor<string, []>("add_32_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_16_cast = sqrt(x = add_32_cast)[name = tensor<string, []>("sqrt_16_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> real_div_16_cast = real_div(x = sub_32_cast, y = sqrt_16_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_65_cast")];
            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, []>(93986240)))];
            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, []>(93988864)))];
            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_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_65_cast)[name = tensor<string, []>("add_33_cast")];
            tensor<fp16, [2, 1280, 16, 16]> input_155_cast = silu(x = add_33_cast)[name = tensor<string, []>("input_155_cast")];
            tensor<int32, [2]> var_5146 = const()[name = tensor<string, []>("op_5146"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5148 = const()[name = tensor<string, []>("op_5148"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [11059200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93991488))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105050752))), name = tensor<string, []>("down_blocks_2_resnets_1_conv2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 3, 3])];
            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, []>(105050944)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_93_cast = conv(bias = down_blocks_2_resnets_1_conv2_bias_to_fp16, dilations = var_5148, groups = var_4569, pad = hidden_states_93_pad_0, pad_type = hidden_states_93_pad_type_0, strides = var_5146, weight = down_blocks_2_resnets_1_conv2_weight_to_fp16_palettized, x = input_155_cast)[name = tensor<string, []>("hidden_states_93_cast")];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_95_cast = add(x = input_143_cast, y = hidden_states_93_cast)[name = tensor<string, []>("hidden_states_95_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)[name = tensor<string, []>("reshape_68_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)[name = tensor<string, []>("reduce_mean_51_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> sub_34_cast = sub(x = reshape_68_cast, y = reduce_mean_51_cast)[name = tensor<string, []>("sub_34_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> square_17_cast = square(x = sub_34_cast)[name = tensor<string, []>("square_17_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)[name = tensor<string, []>("reduce_mean_53_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)[name = tensor<string, []>("add_34_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_17_cast = sqrt(x = add_34_cast)[name = tensor<string, []>("sqrt_17_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> real_div_17_cast = real_div(x = sub_34_cast, y = sqrt_17_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_69_cast")];
            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, []>(105053568)))];
            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, []>(105056192)))];
            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_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_69_cast)[name = tensor<string, []>("add_35_cast")];
            tensor<int32, [2]> var_5168 = const()[name = tensor<string, []>("op_5168"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5170 = const()[name = tensor<string, []>("op_5170"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105058816))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106287680))), name = tensor<string, []>("down_blocks_2_attentions_1_proj_in_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(106287872)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_97_cast = conv(bias = down_blocks_2_attentions_1_proj_in_bias_to_fp16, dilations = var_5170, groups = var_4569, pad = hidden_states_97_pad_0, pad_type = hidden_states_97_pad_type_0, strides = var_5168, weight = down_blocks_2_attentions_1_proj_in_weight_to_fp16_palettized, x = add_35_cast)[name = tensor<string, []>("hidden_states_97_cast")];
            tensor<int32, [4]> var_5175 = const()[name = tensor<string, []>("op_5175"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<fp16, [2, 1280, 1, 256]> inputs_31_cast = reshape(shape = var_5175, x = hidden_states_97_cast)[name = tensor<string, []>("inputs_31_cast")];
            tensor<int32, [1]> var_5185 = const()[name = tensor<string, []>("op_5185"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_31_cast = reduce_mean(axes = var_5185, keep_dims = var_4564, x = inputs_31_cast)[name = tensor<string, []>("channels_mean_31_cast")];
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_31_cast = sub(x = inputs_31_cast, y = channels_mean_31_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_31_cast")];
            tensor<int32, [1]> var_5189 = const()[name = tensor<string, []>("op_5189"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_5190_cast = reduce_mean(axes = var_5189, keep_dims = var_4564, x = zero_mean_sq_31_cast)[name = tensor<string, []>("op_5190_cast")];
            tensor<fp16, []> var_5191_to_fp16 = const()[name = tensor<string, []>("op_5191_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_5192_cast = add(x = var_5190_cast, y = var_5191_to_fp16)[name = tensor<string, []>("op_5192_cast")];
            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_5192_cast)[name = tensor<string, []>("denom_31_cast")];
            tensor<fp16, [2, 1280, 1, 256]> out_31_cast = mul(x = zero_mean_31_cast, y = denom_31_cast)[name = tensor<string, []>("out_31_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_5196_to_fp16 = const()[name = tensor<string, []>("op_5196_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106290496)))];
            tensor<fp16, [2, 1280, 1, 256]> var_5197_cast = add(x = out_31_cast, y = var_5196_to_fp16)[name = tensor<string, []>("op_5197_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_5199_to_fp16 = const()[name = tensor<string, []>("op_5199_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106293120)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_99_cast = mul(x = var_5197_cast, y = var_5199_to_fp16)[name = tensor<string, []>("hidden_states_99_cast")];
            tensor<int32, [2]> var_5206 = const()[name = tensor<string, []>("op_5206"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5208 = const()[name = tensor<string, []>("op_5208"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106295744))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107524608))), name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 256]> q_21_cast = conv(dilations = var_5208, groups = var_4569, pad = q_21_pad_0, pad_type = q_21_pad_type_0, strides = var_5206, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_99_cast)[name = tensor<string, []>("q_21_cast")];
            tensor<int32, [2]> var_5212 = const()[name = tensor<string, []>("op_5212"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5214 = const()[name = tensor<string, []>("op_5214"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107524800))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108753664))), name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 256]> k_41_cast = conv(dilations = var_5214, groups = var_4569, pad = k_41_pad_0, pad_type = k_41_pad_type_0, strides = var_5212, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_99_cast)[name = tensor<string, []>("k_41_cast")];
            tensor<int32, [2]> var_5218 = const()[name = tensor<string, []>("op_5218"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5220 = const()[name = tensor<string, []>("op_5220"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108753856))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109982720))), name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 256]> v_21_cast = conv(dilations = var_5220, groups = var_4569, pad = v_21_pad_0, pad_type = v_21_pad_type_0, strides = var_5218, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_99_cast)[name = tensor<string, []>("v_21_cast")];
            tensor<int32, [4]> var_5224_begin_0 = const()[name = tensor<string, []>("op_5224_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5224_end_0 = const()[name = tensor<string, []>("op_5224_end_0"), val = tensor<int32, [4]>([2, 160, 1, 256])];
            tensor<bool, [4]> var_5224_end_mask_0 = const()[name = tensor<string, []>("op_5224_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_5224_cast = slice_by_index(begin = var_5224_begin_0, end = var_5224_end_0, end_mask = var_5224_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_5224_cast")];
            tensor<int32, [4]> var_5228_begin_0 = const()[name = tensor<string, []>("op_5228_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_5228_end_0 = const()[name = tensor<string, []>("op_5228_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_5228_end_mask_0 = const()[name = tensor<string, []>("op_5228_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_5228_cast = slice_by_index(begin = var_5228_begin_0, end = var_5228_end_0, end_mask = var_5228_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_5228_cast")];
            tensor<int32, [4]> var_5232_begin_0 = const()[name = tensor<string, []>("op_5232_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_5232_end_0 = const()[name = tensor<string, []>("op_5232_end_0"), val = tensor<int32, [4]>([2, 480, 1, 256])];
            tensor<bool, [4]> var_5232_end_mask_0 = const()[name = tensor<string, []>("op_5232_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_5232_cast = slice_by_index(begin = var_5232_begin_0, end = var_5232_end_0, end_mask = var_5232_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_5232_cast")];
            tensor<int32, [4]> var_5236_begin_0 = const()[name = tensor<string, []>("op_5236_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_5236_end_0 = const()[name = tensor<string, []>("op_5236_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_5236_end_mask_0 = const()[name = tensor<string, []>("op_5236_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_5236_cast = slice_by_index(begin = var_5236_begin_0, end = var_5236_end_0, end_mask = var_5236_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_5236_cast")];
            tensor<int32, [4]> var_5240_begin_0 = const()[name = tensor<string, []>("op_5240_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_5240_end_0 = const()[name = tensor<string, []>("op_5240_end_0"), val = tensor<int32, [4]>([2, 800, 1, 256])];
            tensor<bool, [4]> var_5240_end_mask_0 = const()[name = tensor<string, []>("op_5240_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_5240_cast = slice_by_index(begin = var_5240_begin_0, end = var_5240_end_0, end_mask = var_5240_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_5240_cast")];
            tensor<int32, [4]> var_5244_begin_0 = const()[name = tensor<string, []>("op_5244_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_5244_end_0 = const()[name = tensor<string, []>("op_5244_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_5244_end_mask_0 = const()[name = tensor<string, []>("op_5244_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_5244_cast = slice_by_index(begin = var_5244_begin_0, end = var_5244_end_0, end_mask = var_5244_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_5244_cast")];
            tensor<int32, [4]> var_5248_begin_0 = const()[name = tensor<string, []>("op_5248_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_5248_end_0 = const()[name = tensor<string, []>("op_5248_end_0"), val = tensor<int32, [4]>([2, 1120, 1, 256])];
            tensor<bool, [4]> var_5248_end_mask_0 = const()[name = tensor<string, []>("op_5248_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_5248_cast = slice_by_index(begin = var_5248_begin_0, end = var_5248_end_0, end_mask = var_5248_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_5248_cast")];
            tensor<int32, [4]> var_5252_begin_0 = const()[name = tensor<string, []>("op_5252_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_5252_end_0 = const()[name = tensor<string, []>("op_5252_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_5252_end_mask_0 = const()[name = tensor<string, []>("op_5252_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_5252_cast = slice_by_index(begin = var_5252_begin_0, end = var_5252_end_0, end_mask = var_5252_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_5252_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_5259_begin_0 = const()[name = tensor<string, []>("op_5259_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5259_end_0 = const()[name = tensor<string, []>("op_5259_end_0"), val = tensor<int32, [4]>([2, 256, 1, 160])];
            tensor<bool, [4]> var_5259_end_mask_0 = const()[name = tensor<string, []>("op_5259_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)[name = tensor<string, []>("transpose_21")];
            tensor<fp16, [2, 256, 1, 160]> var_5259_cast = slice_by_index(begin = var_5259_begin_0, end = var_5259_end_0, end_mask = var_5259_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_5259_cast")];
            tensor<int32, [4]> var_5263_begin_0 = const()[name = tensor<string, []>("op_5263_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_5263_end_0 = const()[name = tensor<string, []>("op_5263_end_0"), val = tensor<int32, [4]>([2, 256, 1, 320])];
            tensor<bool, [4]> var_5263_end_mask_0 = const()[name = tensor<string, []>("op_5263_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_5263_cast = slice_by_index(begin = var_5263_begin_0, end = var_5263_end_0, end_mask = var_5263_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_5263_cast")];
            tensor<int32, [4]> var_5267_begin_0 = const()[name = tensor<string, []>("op_5267_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_5267_end_0 = const()[name = tensor<string, []>("op_5267_end_0"), val = tensor<int32, [4]>([2, 256, 1, 480])];
            tensor<bool, [4]> var_5267_end_mask_0 = const()[name = tensor<string, []>("op_5267_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_5267_cast = slice_by_index(begin = var_5267_begin_0, end = var_5267_end_0, end_mask = var_5267_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_5267_cast")];
            tensor<int32, [4]> var_5271_begin_0 = const()[name = tensor<string, []>("op_5271_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 480])];
            tensor<int32, [4]> var_5271_end_0 = const()[name = tensor<string, []>("op_5271_end_0"), val = tensor<int32, [4]>([2, 256, 1, 640])];
            tensor<bool, [4]> var_5271_end_mask_0 = const()[name = tensor<string, []>("op_5271_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_5271_cast = slice_by_index(begin = var_5271_begin_0, end = var_5271_end_0, end_mask = var_5271_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_5271_cast")];
            tensor<int32, [4]> var_5275_begin_0 = const()[name = tensor<string, []>("op_5275_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_5275_end_0 = const()[name = tensor<string, []>("op_5275_end_0"), val = tensor<int32, [4]>([2, 256, 1, 800])];
            tensor<bool, [4]> var_5275_end_mask_0 = const()[name = tensor<string, []>("op_5275_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_5275_cast = slice_by_index(begin = var_5275_begin_0, end = var_5275_end_0, end_mask = var_5275_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_5275_cast")];
            tensor<int32, [4]> var_5279_begin_0 = const()[name = tensor<string, []>("op_5279_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 800])];
            tensor<int32, [4]> var_5279_end_0 = const()[name = tensor<string, []>("op_5279_end_0"), val = tensor<int32, [4]>([2, 256, 1, 960])];
            tensor<bool, [4]> var_5279_end_mask_0 = const()[name = tensor<string, []>("op_5279_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_5279_cast = slice_by_index(begin = var_5279_begin_0, end = var_5279_end_0, end_mask = var_5279_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_5279_cast")];
            tensor<int32, [4]> var_5283_begin_0 = const()[name = tensor<string, []>("op_5283_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_5283_end_0 = const()[name = tensor<string, []>("op_5283_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1120])];
            tensor<bool, [4]> var_5283_end_mask_0 = const()[name = tensor<string, []>("op_5283_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_5283_cast = slice_by_index(begin = var_5283_begin_0, end = var_5283_end_0, end_mask = var_5283_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_5283_cast")];
            tensor<int32, [4]> var_5287_begin_0 = const()[name = tensor<string, []>("op_5287_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1120])];
            tensor<int32, [4]> var_5287_end_0 = const()[name = tensor<string, []>("op_5287_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1280])];
            tensor<bool, [4]> var_5287_end_mask_0 = const()[name = tensor<string, []>("op_5287_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_5287_cast = slice_by_index(begin = var_5287_begin_0, end = var_5287_end_0, end_mask = var_5287_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_5287_cast")];
            tensor<int32, [4]> var_5289_begin_0 = const()[name = tensor<string, []>("op_5289_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5289_end_0 = const()[name = tensor<string, []>("op_5289_end_0"), val = tensor<int32, [4]>([2, 160, 1, 256])];
            tensor<bool, [4]> var_5289_end_mask_0 = const()[name = tensor<string, []>("op_5289_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_5289_cast = slice_by_index(begin = var_5289_begin_0, end = var_5289_end_0, end_mask = var_5289_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_5289_cast")];
            tensor<int32, [4]> var_5293_begin_0 = const()[name = tensor<string, []>("op_5293_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_5293_end_0 = const()[name = tensor<string, []>("op_5293_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_5293_end_mask_0 = const()[name = tensor<string, []>("op_5293_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_5293_cast = slice_by_index(begin = var_5293_begin_0, end = var_5293_end_0, end_mask = var_5293_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_5293_cast")];
            tensor<int32, [4]> var_5297_begin_0 = const()[name = tensor<string, []>("op_5297_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_5297_end_0 = const()[name = tensor<string, []>("op_5297_end_0"), val = tensor<int32, [4]>([2, 480, 1, 256])];
            tensor<bool, [4]> var_5297_end_mask_0 = const()[name = tensor<string, []>("op_5297_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_5297_cast = slice_by_index(begin = var_5297_begin_0, end = var_5297_end_0, end_mask = var_5297_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_5297_cast")];
            tensor<int32, [4]> var_5301_begin_0 = const()[name = tensor<string, []>("op_5301_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_5301_end_0 = const()[name = tensor<string, []>("op_5301_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_5301_end_mask_0 = const()[name = tensor<string, []>("op_5301_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_5301_cast = slice_by_index(begin = var_5301_begin_0, end = var_5301_end_0, end_mask = var_5301_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_5301_cast")];
            tensor<int32, [4]> var_5305_begin_0 = const()[name = tensor<string, []>("op_5305_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_5305_end_0 = const()[name = tensor<string, []>("op_5305_end_0"), val = tensor<int32, [4]>([2, 800, 1, 256])];
            tensor<bool, [4]> var_5305_end_mask_0 = const()[name = tensor<string, []>("op_5305_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_5305_cast = slice_by_index(begin = var_5305_begin_0, end = var_5305_end_0, end_mask = var_5305_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_5305_cast")];
            tensor<int32, [4]> var_5309_begin_0 = const()[name = tensor<string, []>("op_5309_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_5309_end_0 = const()[name = tensor<string, []>("op_5309_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_5309_end_mask_0 = const()[name = tensor<string, []>("op_5309_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_5309_cast = slice_by_index(begin = var_5309_begin_0, end = var_5309_end_0, end_mask = var_5309_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_5309_cast")];
            tensor<int32, [4]> var_5313_begin_0 = const()[name = tensor<string, []>("op_5313_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_5313_end_0 = const()[name = tensor<string, []>("op_5313_end_0"), val = tensor<int32, [4]>([2, 1120, 1, 256])];
            tensor<bool, [4]> var_5313_end_mask_0 = const()[name = tensor<string, []>("op_5313_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_5313_cast = slice_by_index(begin = var_5313_begin_0, end = var_5313_end_0, end_mask = var_5313_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_5313_cast")];
            tensor<int32, [4]> var_5317_begin_0 = const()[name = tensor<string, []>("op_5317_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_5317_end_0 = const()[name = tensor<string, []>("op_5317_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_5317_end_mask_0 = const()[name = tensor<string, []>("op_5317_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_5317_cast = slice_by_index(begin = var_5317_begin_0, end = var_5317_end_0, end_mask = var_5317_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_5317_cast")];
            tensor<string, []> var_5321_equation_0 = const()[name = tensor<string, []>("op_5321_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5321_cast = einsum(equation = var_5321_equation_0, values = (var_5259_cast, var_5224_cast))[name = tensor<string, []>("op_5321_cast")];
            tensor<fp16, []> var_5322_to_fp16 = const()[name = tensor<string, []>("op_5322_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_33_cast = mul(x = var_5321_cast, y = var_5322_to_fp16)[name = tensor<string, []>("aw_33_cast")];
            tensor<string, []> var_5325_equation_0 = const()[name = tensor<string, []>("op_5325_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5325_cast = einsum(equation = var_5325_equation_0, values = (var_5263_cast, var_5228_cast))[name = tensor<string, []>("op_5325_cast")];
            tensor<fp16, []> var_5326_to_fp16 = const()[name = tensor<string, []>("op_5326_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_35_cast = mul(x = var_5325_cast, y = var_5326_to_fp16)[name = tensor<string, []>("aw_35_cast")];
            tensor<string, []> var_5329_equation_0 = const()[name = tensor<string, []>("op_5329_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5329_cast = einsum(equation = var_5329_equation_0, values = (var_5267_cast, var_5232_cast))[name = tensor<string, []>("op_5329_cast")];
            tensor<fp16, []> var_5330_to_fp16 = const()[name = tensor<string, []>("op_5330_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_37_cast = mul(x = var_5329_cast, y = var_5330_to_fp16)[name = tensor<string, []>("aw_37_cast")];
            tensor<string, []> var_5333_equation_0 = const()[name = tensor<string, []>("op_5333_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5333_cast = einsum(equation = var_5333_equation_0, values = (var_5271_cast, var_5236_cast))[name = tensor<string, []>("op_5333_cast")];
            tensor<fp16, []> var_5334_to_fp16 = const()[name = tensor<string, []>("op_5334_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_39_cast = mul(x = var_5333_cast, y = var_5334_to_fp16)[name = tensor<string, []>("aw_39_cast")];
            tensor<string, []> var_5337_equation_0 = const()[name = tensor<string, []>("op_5337_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5337_cast = einsum(equation = var_5337_equation_0, values = (var_5275_cast, var_5240_cast))[name = tensor<string, []>("op_5337_cast")];
            tensor<fp16, []> var_5338_to_fp16 = const()[name = tensor<string, []>("op_5338_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_41_cast = mul(x = var_5337_cast, y = var_5338_to_fp16)[name = tensor<string, []>("aw_41_cast")];
            tensor<string, []> var_5341_equation_0 = const()[name = tensor<string, []>("op_5341_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5341_cast = einsum(equation = var_5341_equation_0, values = (var_5279_cast, var_5244_cast))[name = tensor<string, []>("op_5341_cast")];
            tensor<fp16, []> var_5342_to_fp16 = const()[name = tensor<string, []>("op_5342_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_43_cast = mul(x = var_5341_cast, y = var_5342_to_fp16)[name = tensor<string, []>("aw_43_cast")];
            tensor<string, []> var_5345_equation_0 = const()[name = tensor<string, []>("op_5345_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5345_cast = einsum(equation = var_5345_equation_0, values = (var_5283_cast, var_5248_cast))[name = tensor<string, []>("op_5345_cast")];
            tensor<fp16, []> var_5346_to_fp16 = const()[name = tensor<string, []>("op_5346_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_45_cast = mul(x = var_5345_cast, y = var_5346_to_fp16)[name = tensor<string, []>("aw_45_cast")];
            tensor<string, []> var_5349_equation_0 = const()[name = tensor<string, []>("op_5349_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5349_cast = einsum(equation = var_5349_equation_0, values = (var_5287_cast, var_5252_cast))[name = tensor<string, []>("op_5349_cast")];
            tensor<fp16, []> var_5350_to_fp16 = const()[name = tensor<string, []>("op_5350_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_47_cast = mul(x = var_5349_cast, y = var_5350_to_fp16)[name = tensor<string, []>("aw_47_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_5352_cast = softmax(axis = var_4569, x = aw_33_cast)[name = tensor<string, []>("op_5352_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_5353_cast = softmax(axis = var_4569, x = aw_35_cast)[name = tensor<string, []>("op_5353_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_5354_cast = softmax(axis = var_4569, x = aw_37_cast)[name = tensor<string, []>("op_5354_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_5355_cast = softmax(axis = var_4569, x = aw_39_cast)[name = tensor<string, []>("op_5355_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_5356_cast = softmax(axis = var_4569, x = aw_41_cast)[name = tensor<string, []>("op_5356_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_5357_cast = softmax(axis = var_4569, x = aw_43_cast)[name = tensor<string, []>("op_5357_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_5358_cast = softmax(axis = var_4569, x = aw_45_cast)[name = tensor<string, []>("op_5358_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_5359_cast = softmax(axis = var_4569, x = aw_47_cast)[name = tensor<string, []>("op_5359_cast")];
            tensor<string, []> var_5361_equation_0 = const()[name = tensor<string, []>("op_5361_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_5361_cast = einsum(equation = var_5361_equation_0, values = (var_5289_cast, var_5352_cast))[name = tensor<string, []>("op_5361_cast")];
            tensor<string, []> var_5363_equation_0 = const()[name = tensor<string, []>("op_5363_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_5363_cast = einsum(equation = var_5363_equation_0, values = (var_5293_cast, var_5353_cast))[name = tensor<string, []>("op_5363_cast")];
            tensor<string, []> var_5365_equation_0 = const()[name = tensor<string, []>("op_5365_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_5365_cast = einsum(equation = var_5365_equation_0, values = (var_5297_cast, var_5354_cast))[name = tensor<string, []>("op_5365_cast")];
            tensor<string, []> var_5367_equation_0 = const()[name = tensor<string, []>("op_5367_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_5367_cast = einsum(equation = var_5367_equation_0, values = (var_5301_cast, var_5355_cast))[name = tensor<string, []>("op_5367_cast")];
            tensor<string, []> var_5369_equation_0 = const()[name = tensor<string, []>("op_5369_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_5369_cast = einsum(equation = var_5369_equation_0, values = (var_5305_cast, var_5356_cast))[name = tensor<string, []>("op_5369_cast")];
            tensor<string, []> var_5371_equation_0 = const()[name = tensor<string, []>("op_5371_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_5371_cast = einsum(equation = var_5371_equation_0, values = (var_5309_cast, var_5357_cast))[name = tensor<string, []>("op_5371_cast")];
            tensor<string, []> var_5373_equation_0 = const()[name = tensor<string, []>("op_5373_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_5373_cast = einsum(equation = var_5373_equation_0, values = (var_5313_cast, var_5358_cast))[name = tensor<string, []>("op_5373_cast")];
            tensor<string, []> var_5375_equation_0 = const()[name = tensor<string, []>("op_5375_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_5375_cast = einsum(equation = var_5375_equation_0, values = (var_5317_cast, var_5359_cast))[name = tensor<string, []>("op_5375_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_4569, interleave = input_159_interleave_0, values = (var_5361_cast, var_5363_cast, var_5365_cast, var_5367_cast, var_5369_cast, var_5371_cast, var_5373_cast, var_5375_cast))[name = tensor<string, []>("input_159_cast")];
            tensor<int32, [2]> var_5381 = const()[name = tensor<string, []>("op_5381"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5383 = const()[name = tensor<string, []>("op_5383"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_5385_pad_type_0 = const()[name = tensor<string, []>("op_5385_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_5385_pad_0 = const()[name = tensor<string, []>("op_5385_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109982912))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111211776))), name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(111211968)))];
            tensor<fp16, [2, 1280, 1, 256]> var_5385_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_5383, groups = var_4569, pad = var_5385_pad_0, pad_type = var_5385_pad_type_0, strides = var_5381, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_159_cast)[name = tensor<string, []>("op_5385_cast")];
            tensor<fp16, [2, 1280, 1, 256]> inputs_33_cast = add(x = var_5385_cast, y = inputs_31_cast)[name = tensor<string, []>("inputs_33_cast")];
            tensor<int32, [1]> var_5389 = const()[name = tensor<string, []>("op_5389"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_33_cast = reduce_mean(axes = var_5389, keep_dims = var_4564, x = inputs_33_cast)[name = tensor<string, []>("channels_mean_33_cast")];
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_33_cast = sub(x = inputs_33_cast, y = channels_mean_33_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_33_cast")];
            tensor<int32, [1]> var_5393 = const()[name = tensor<string, []>("op_5393"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_5394_cast = reduce_mean(axes = var_5393, keep_dims = var_4564, x = zero_mean_sq_33_cast)[name = tensor<string, []>("op_5394_cast")];
            tensor<fp16, []> var_5395_to_fp16 = const()[name = tensor<string, []>("op_5395_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_5396_cast = add(x = var_5394_cast, y = var_5395_to_fp16)[name = tensor<string, []>("op_5396_cast")];
            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_5396_cast)[name = tensor<string, []>("denom_33_cast")];
            tensor<fp16, [2, 1280, 1, 256]> out_33_cast = mul(x = zero_mean_33_cast, y = denom_33_cast)[name = tensor<string, []>("out_33_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_5400_to_fp16 = const()[name = tensor<string, []>("op_5400_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111214592)))];
            tensor<fp16, [2, 1280, 1, 256]> var_5401_cast = add(x = out_33_cast, y = var_5400_to_fp16)[name = tensor<string, []>("op_5401_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_5403_to_fp16 = const()[name = tensor<string, []>("op_5403_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111217216)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_101_cast = mul(x = var_5401_cast, y = var_5403_to_fp16)[name = tensor<string, []>("hidden_states_101_cast")];
            tensor<int32, [2]> var_5410 = const()[name = tensor<string, []>("op_5410"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5412 = const()[name = tensor<string, []>("op_5412"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111219840))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112448704))), name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 256]> q_23_cast = conv(dilations = var_5412, groups = var_4569, pad = q_23_pad_0, pad_type = q_23_pad_type_0, strides = var_5410, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_101_cast)[name = tensor<string, []>("q_23_cast")];
            tensor<int32, [2]> var_5416 = const()[name = tensor<string, []>("op_5416"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5418 = const()[name = tensor<string, []>("op_5418"), 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, 768, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [737280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112448896))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113186240))), name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 768, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_45_cast = conv(dilations = var_5418, groups = var_4569, pad = k_45_pad_0, pad_type = k_45_pad_type_0, strides = var_5416, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_45_cast")];
            tensor<int32, [2]> var_5422 = const()[name = tensor<string, []>("op_5422"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5424 = const()[name = tensor<string, []>("op_5424"), 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, 768, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [737280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113186432))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113923776))), name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 768, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_23_cast = conv(dilations = var_5424, groups = var_4569, pad = v_23_pad_0, pad_type = v_23_pad_type_0, strides = var_5422, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_23_cast")];
            tensor<int32, [4]> var_5428_begin_0 = const()[name = tensor<string, []>("op_5428_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5428_end_0 = const()[name = tensor<string, []>("op_5428_end_0"), val = tensor<int32, [4]>([2, 160, 1, 256])];
            tensor<bool, [4]> var_5428_end_mask_0 = const()[name = tensor<string, []>("op_5428_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_5428_cast = slice_by_index(begin = var_5428_begin_0, end = var_5428_end_0, end_mask = var_5428_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_5428_cast")];
            tensor<int32, [4]> var_5432_begin_0 = const()[name = tensor<string, []>("op_5432_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_5432_end_0 = const()[name = tensor<string, []>("op_5432_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_5432_end_mask_0 = const()[name = tensor<string, []>("op_5432_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_5432_cast = slice_by_index(begin = var_5432_begin_0, end = var_5432_end_0, end_mask = var_5432_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_5432_cast")];
            tensor<int32, [4]> var_5436_begin_0 = const()[name = tensor<string, []>("op_5436_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_5436_end_0 = const()[name = tensor<string, []>("op_5436_end_0"), val = tensor<int32, [4]>([2, 480, 1, 256])];
            tensor<bool, [4]> var_5436_end_mask_0 = const()[name = tensor<string, []>("op_5436_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_5436_cast = slice_by_index(begin = var_5436_begin_0, end = var_5436_end_0, end_mask = var_5436_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_5436_cast")];
            tensor<int32, [4]> var_5440_begin_0 = const()[name = tensor<string, []>("op_5440_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_5440_end_0 = const()[name = tensor<string, []>("op_5440_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_5440_end_mask_0 = const()[name = tensor<string, []>("op_5440_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_5440_cast = slice_by_index(begin = var_5440_begin_0, end = var_5440_end_0, end_mask = var_5440_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_5440_cast")];
            tensor<int32, [4]> var_5444_begin_0 = const()[name = tensor<string, []>("op_5444_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_5444_end_0 = const()[name = tensor<string, []>("op_5444_end_0"), val = tensor<int32, [4]>([2, 800, 1, 256])];
            tensor<bool, [4]> var_5444_end_mask_0 = const()[name = tensor<string, []>("op_5444_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_5444_cast = slice_by_index(begin = var_5444_begin_0, end = var_5444_end_0, end_mask = var_5444_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_5444_cast")];
            tensor<int32, [4]> var_5448_begin_0 = const()[name = tensor<string, []>("op_5448_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_5448_end_0 = const()[name = tensor<string, []>("op_5448_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_5448_end_mask_0 = const()[name = tensor<string, []>("op_5448_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_5448_cast = slice_by_index(begin = var_5448_begin_0, end = var_5448_end_0, end_mask = var_5448_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_5448_cast")];
            tensor<int32, [4]> var_5452_begin_0 = const()[name = tensor<string, []>("op_5452_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_5452_end_0 = const()[name = tensor<string, []>("op_5452_end_0"), val = tensor<int32, [4]>([2, 1120, 1, 256])];
            tensor<bool, [4]> var_5452_end_mask_0 = const()[name = tensor<string, []>("op_5452_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_5452_cast = slice_by_index(begin = var_5452_begin_0, end = var_5452_end_0, end_mask = var_5452_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_5452_cast")];
            tensor<int32, [4]> var_5456_begin_0 = const()[name = tensor<string, []>("op_5456_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_5456_end_0 = const()[name = tensor<string, []>("op_5456_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_5456_end_mask_0 = const()[name = tensor<string, []>("op_5456_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_5456_cast = slice_by_index(begin = var_5456_begin_0, end = var_5456_end_0, end_mask = var_5456_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_5456_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_5463_begin_0 = const()[name = tensor<string, []>("op_5463_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5463_end_0 = const()[name = tensor<string, []>("op_5463_end_0"), val = tensor<int32, [4]>([2, 77, 1, 160])];
            tensor<bool, [4]> var_5463_end_mask_0 = const()[name = tensor<string, []>("op_5463_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)[name = tensor<string, []>("transpose_20")];
            tensor<fp16, [2, 77, 1, 160]> var_5463_cast = slice_by_index(begin = var_5463_begin_0, end = var_5463_end_0, end_mask = var_5463_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_5463_cast")];
            tensor<int32, [4]> var_5467_begin_0 = const()[name = tensor<string, []>("op_5467_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_5467_end_0 = const()[name = tensor<string, []>("op_5467_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_5467_end_mask_0 = const()[name = tensor<string, []>("op_5467_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_5467_cast = slice_by_index(begin = var_5467_begin_0, end = var_5467_end_0, end_mask = var_5467_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_5467_cast")];
            tensor<int32, [4]> var_5471_begin_0 = const()[name = tensor<string, []>("op_5471_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_5471_end_0 = const()[name = tensor<string, []>("op_5471_end_0"), val = tensor<int32, [4]>([2, 77, 1, 480])];
            tensor<bool, [4]> var_5471_end_mask_0 = const()[name = tensor<string, []>("op_5471_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_5471_cast = slice_by_index(begin = var_5471_begin_0, end = var_5471_end_0, end_mask = var_5471_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_5471_cast")];
            tensor<int32, [4]> var_5475_begin_0 = const()[name = tensor<string, []>("op_5475_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 480])];
            tensor<int32, [4]> var_5475_end_0 = const()[name = tensor<string, []>("op_5475_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_5475_end_mask_0 = const()[name = tensor<string, []>("op_5475_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_5475_cast = slice_by_index(begin = var_5475_begin_0, end = var_5475_end_0, end_mask = var_5475_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_5475_cast")];
            tensor<int32, [4]> var_5479_begin_0 = const()[name = tensor<string, []>("op_5479_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_5479_end_0 = const()[name = tensor<string, []>("op_5479_end_0"), val = tensor<int32, [4]>([2, 77, 1, 800])];
            tensor<bool, [4]> var_5479_end_mask_0 = const()[name = tensor<string, []>("op_5479_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_5479_cast = slice_by_index(begin = var_5479_begin_0, end = var_5479_end_0, end_mask = var_5479_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_5479_cast")];
            tensor<int32, [4]> var_5483_begin_0 = const()[name = tensor<string, []>("op_5483_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 800])];
            tensor<int32, [4]> var_5483_end_0 = const()[name = tensor<string, []>("op_5483_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_5483_end_mask_0 = const()[name = tensor<string, []>("op_5483_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_5483_cast = slice_by_index(begin = var_5483_begin_0, end = var_5483_end_0, end_mask = var_5483_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_5483_cast")];
            tensor<int32, [4]> var_5487_begin_0 = const()[name = tensor<string, []>("op_5487_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_5487_end_0 = const()[name = tensor<string, []>("op_5487_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1120])];
            tensor<bool, [4]> var_5487_end_mask_0 = const()[name = tensor<string, []>("op_5487_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_5487_cast = slice_by_index(begin = var_5487_begin_0, end = var_5487_end_0, end_mask = var_5487_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_5487_cast")];
            tensor<int32, [4]> var_5491_begin_0 = const()[name = tensor<string, []>("op_5491_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1120])];
            tensor<int32, [4]> var_5491_end_0 = const()[name = tensor<string, []>("op_5491_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_5491_end_mask_0 = const()[name = tensor<string, []>("op_5491_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_5491_cast = slice_by_index(begin = var_5491_begin_0, end = var_5491_end_0, end_mask = var_5491_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_5491_cast")];
            tensor<int32, [4]> var_5493_begin_0 = const()[name = tensor<string, []>("op_5493_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5493_end_0 = const()[name = tensor<string, []>("op_5493_end_0"), val = tensor<int32, [4]>([2, 160, 1, 77])];
            tensor<bool, [4]> var_5493_end_mask_0 = const()[name = tensor<string, []>("op_5493_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_5493_cast = slice_by_index(begin = var_5493_begin_0, end = var_5493_end_0, end_mask = var_5493_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_5493_cast")];
            tensor<int32, [4]> var_5497_begin_0 = const()[name = tensor<string, []>("op_5497_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_5497_end_0 = const()[name = tensor<string, []>("op_5497_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_5497_end_mask_0 = const()[name = tensor<string, []>("op_5497_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_5497_cast = slice_by_index(begin = var_5497_begin_0, end = var_5497_end_0, end_mask = var_5497_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_5497_cast")];
            tensor<int32, [4]> var_5501_begin_0 = const()[name = tensor<string, []>("op_5501_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_5501_end_0 = const()[name = tensor<string, []>("op_5501_end_0"), val = tensor<int32, [4]>([2, 480, 1, 77])];
            tensor<bool, [4]> var_5501_end_mask_0 = const()[name = tensor<string, []>("op_5501_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_5501_cast = slice_by_index(begin = var_5501_begin_0, end = var_5501_end_0, end_mask = var_5501_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_5501_cast")];
            tensor<int32, [4]> var_5505_begin_0 = const()[name = tensor<string, []>("op_5505_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_5505_end_0 = const()[name = tensor<string, []>("op_5505_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_5505_end_mask_0 = const()[name = tensor<string, []>("op_5505_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_5505_cast = slice_by_index(begin = var_5505_begin_0, end = var_5505_end_0, end_mask = var_5505_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_5505_cast")];
            tensor<int32, [4]> var_5509_begin_0 = const()[name = tensor<string, []>("op_5509_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_5509_end_0 = const()[name = tensor<string, []>("op_5509_end_0"), val = tensor<int32, [4]>([2, 800, 1, 77])];
            tensor<bool, [4]> var_5509_end_mask_0 = const()[name = tensor<string, []>("op_5509_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_5509_cast = slice_by_index(begin = var_5509_begin_0, end = var_5509_end_0, end_mask = var_5509_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_5509_cast")];
            tensor<int32, [4]> var_5513_begin_0 = const()[name = tensor<string, []>("op_5513_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_5513_end_0 = const()[name = tensor<string, []>("op_5513_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_5513_end_mask_0 = const()[name = tensor<string, []>("op_5513_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_5513_cast = slice_by_index(begin = var_5513_begin_0, end = var_5513_end_0, end_mask = var_5513_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_5513_cast")];
            tensor<int32, [4]> var_5517_begin_0 = const()[name = tensor<string, []>("op_5517_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_5517_end_0 = const()[name = tensor<string, []>("op_5517_end_0"), val = tensor<int32, [4]>([2, 1120, 1, 77])];
            tensor<bool, [4]> var_5517_end_mask_0 = const()[name = tensor<string, []>("op_5517_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_5517_cast = slice_by_index(begin = var_5517_begin_0, end = var_5517_end_0, end_mask = var_5517_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_5517_cast")];
            tensor<int32, [4]> var_5521_begin_0 = const()[name = tensor<string, []>("op_5521_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_5521_end_0 = const()[name = tensor<string, []>("op_5521_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_5521_end_mask_0 = const()[name = tensor<string, []>("op_5521_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_5521_cast = slice_by_index(begin = var_5521_begin_0, end = var_5521_end_0, end_mask = var_5521_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_5521_cast")];
            tensor<string, []> var_5525_equation_0 = const()[name = tensor<string, []>("op_5525_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5525_cast = einsum(equation = var_5525_equation_0, values = (var_5463_cast, var_5428_cast))[name = tensor<string, []>("op_5525_cast")];
            tensor<fp16, []> var_5526_to_fp16 = const()[name = tensor<string, []>("op_5526_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_49_cast = mul(x = var_5525_cast, y = var_5526_to_fp16)[name = tensor<string, []>("aw_49_cast")];
            tensor<string, []> var_5529_equation_0 = const()[name = tensor<string, []>("op_5529_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5529_cast = einsum(equation = var_5529_equation_0, values = (var_5467_cast, var_5432_cast))[name = tensor<string, []>("op_5529_cast")];
            tensor<fp16, []> var_5530_to_fp16 = const()[name = tensor<string, []>("op_5530_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_51_cast = mul(x = var_5529_cast, y = var_5530_to_fp16)[name = tensor<string, []>("aw_51_cast")];
            tensor<string, []> var_5533_equation_0 = const()[name = tensor<string, []>("op_5533_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5533_cast = einsum(equation = var_5533_equation_0, values = (var_5471_cast, var_5436_cast))[name = tensor<string, []>("op_5533_cast")];
            tensor<fp16, []> var_5534_to_fp16 = const()[name = tensor<string, []>("op_5534_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_53_cast = mul(x = var_5533_cast, y = var_5534_to_fp16)[name = tensor<string, []>("aw_53_cast")];
            tensor<string, []> var_5537_equation_0 = const()[name = tensor<string, []>("op_5537_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5537_cast = einsum(equation = var_5537_equation_0, values = (var_5475_cast, var_5440_cast))[name = tensor<string, []>("op_5537_cast")];
            tensor<fp16, []> var_5538_to_fp16 = const()[name = tensor<string, []>("op_5538_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_55_cast = mul(x = var_5537_cast, y = var_5538_to_fp16)[name = tensor<string, []>("aw_55_cast")];
            tensor<string, []> var_5541_equation_0 = const()[name = tensor<string, []>("op_5541_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5541_cast = einsum(equation = var_5541_equation_0, values = (var_5479_cast, var_5444_cast))[name = tensor<string, []>("op_5541_cast")];
            tensor<fp16, []> var_5542_to_fp16 = const()[name = tensor<string, []>("op_5542_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_57_cast = mul(x = var_5541_cast, y = var_5542_to_fp16)[name = tensor<string, []>("aw_57_cast")];
            tensor<string, []> var_5545_equation_0 = const()[name = tensor<string, []>("op_5545_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5545_cast = einsum(equation = var_5545_equation_0, values = (var_5483_cast, var_5448_cast))[name = tensor<string, []>("op_5545_cast")];
            tensor<fp16, []> var_5546_to_fp16 = const()[name = tensor<string, []>("op_5546_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_59_cast = mul(x = var_5545_cast, y = var_5546_to_fp16)[name = tensor<string, []>("aw_59_cast")];
            tensor<string, []> var_5549_equation_0 = const()[name = tensor<string, []>("op_5549_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5549_cast = einsum(equation = var_5549_equation_0, values = (var_5487_cast, var_5452_cast))[name = tensor<string, []>("op_5549_cast")];
            tensor<fp16, []> var_5550_to_fp16 = const()[name = tensor<string, []>("op_5550_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_61_cast = mul(x = var_5549_cast, y = var_5550_to_fp16)[name = tensor<string, []>("aw_61_cast")];
            tensor<string, []> var_5553_equation_0 = const()[name = tensor<string, []>("op_5553_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5553_cast = einsum(equation = var_5553_equation_0, values = (var_5491_cast, var_5456_cast))[name = tensor<string, []>("op_5553_cast")];
            tensor<fp16, []> var_5554_to_fp16 = const()[name = tensor<string, []>("op_5554_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_63_cast = mul(x = var_5553_cast, y = var_5554_to_fp16)[name = tensor<string, []>("aw_63_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5556_cast = softmax(axis = var_4569, x = aw_49_cast)[name = tensor<string, []>("op_5556_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5557_cast = softmax(axis = var_4569, x = aw_51_cast)[name = tensor<string, []>("op_5557_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5558_cast = softmax(axis = var_4569, x = aw_53_cast)[name = tensor<string, []>("op_5558_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5559_cast = softmax(axis = var_4569, x = aw_55_cast)[name = tensor<string, []>("op_5559_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5560_cast = softmax(axis = var_4569, x = aw_57_cast)[name = tensor<string, []>("op_5560_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5561_cast = softmax(axis = var_4569, x = aw_59_cast)[name = tensor<string, []>("op_5561_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5562_cast = softmax(axis = var_4569, x = aw_61_cast)[name = tensor<string, []>("op_5562_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5563_cast = softmax(axis = var_4569, x = aw_63_cast)[name = tensor<string, []>("op_5563_cast")];
            tensor<string, []> var_5565_equation_0 = const()[name = tensor<string, []>("op_5565_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_5565_cast = einsum(equation = var_5565_equation_0, values = (var_5493_cast, var_5556_cast))[name = tensor<string, []>("op_5565_cast")];
            tensor<string, []> var_5567_equation_0 = const()[name = tensor<string, []>("op_5567_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_5567_cast = einsum(equation = var_5567_equation_0, values = (var_5497_cast, var_5557_cast))[name = tensor<string, []>("op_5567_cast")];
            tensor<string, []> var_5569_equation_0 = const()[name = tensor<string, []>("op_5569_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_5569_cast = einsum(equation = var_5569_equation_0, values = (var_5501_cast, var_5558_cast))[name = tensor<string, []>("op_5569_cast")];
            tensor<string, []> var_5571_equation_0 = const()[name = tensor<string, []>("op_5571_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_5571_cast = einsum(equation = var_5571_equation_0, values = (var_5505_cast, var_5559_cast))[name = tensor<string, []>("op_5571_cast")];
            tensor<string, []> var_5573_equation_0 = const()[name = tensor<string, []>("op_5573_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_5573_cast = einsum(equation = var_5573_equation_0, values = (var_5509_cast, var_5560_cast))[name = tensor<string, []>("op_5573_cast")];
            tensor<string, []> var_5575_equation_0 = const()[name = tensor<string, []>("op_5575_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_5575_cast = einsum(equation = var_5575_equation_0, values = (var_5513_cast, var_5561_cast))[name = tensor<string, []>("op_5575_cast")];
            tensor<string, []> var_5577_equation_0 = const()[name = tensor<string, []>("op_5577_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_5577_cast = einsum(equation = var_5577_equation_0, values = (var_5517_cast, var_5562_cast))[name = tensor<string, []>("op_5577_cast")];
            tensor<string, []> var_5579_equation_0 = const()[name = tensor<string, []>("op_5579_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_5579_cast = einsum(equation = var_5579_equation_0, values = (var_5521_cast, var_5563_cast))[name = tensor<string, []>("op_5579_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_4569, interleave = input_161_interleave_0, values = (var_5565_cast, var_5567_cast, var_5569_cast, var_5571_cast, var_5573_cast, var_5575_cast, var_5577_cast, var_5579_cast))[name = tensor<string, []>("input_161_cast")];
            tensor<int32, [2]> var_5585 = const()[name = tensor<string, []>("op_5585"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5587 = const()[name = tensor<string, []>("op_5587"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_5589_pad_type_0 = const()[name = tensor<string, []>("op_5589_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_5589_pad_0 = const()[name = tensor<string, []>("op_5589_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113923968))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115152832))), name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(115153024)))];
            tensor<fp16, [2, 1280, 1, 256]> var_5589_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_5587, groups = var_4569, pad = var_5589_pad_0, pad_type = var_5589_pad_type_0, strides = var_5585, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_161_cast)[name = tensor<string, []>("op_5589_cast")];
            tensor<fp16, [2, 1280, 1, 256]> inputs_35_cast = add(x = var_5589_cast, y = inputs_33_cast)[name = tensor<string, []>("inputs_35_cast")];
            tensor<int32, [1]> var_5593 = const()[name = tensor<string, []>("op_5593"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_35_cast = reduce_mean(axes = var_5593, keep_dims = var_4564, x = inputs_35_cast)[name = tensor<string, []>("channels_mean_35_cast")];
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_35_cast = sub(x = inputs_35_cast, y = channels_mean_35_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_35_cast")];
            tensor<int32, [1]> var_5597 = const()[name = tensor<string, []>("op_5597"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_5598_cast = reduce_mean(axes = var_5597, keep_dims = var_4564, x = zero_mean_sq_35_cast)[name = tensor<string, []>("op_5598_cast")];
            tensor<fp16, []> var_5599_to_fp16 = const()[name = tensor<string, []>("op_5599_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_5600_cast = add(x = var_5598_cast, y = var_5599_to_fp16)[name = tensor<string, []>("op_5600_cast")];
            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_5600_cast)[name = tensor<string, []>("denom_35_cast")];
            tensor<fp16, [2, 1280, 1, 256]> out_35_cast = mul(x = zero_mean_35_cast, y = denom_35_cast)[name = tensor<string, []>("out_35_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_5604_to_fp16 = const()[name = tensor<string, []>("op_5604_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115155648)))];
            tensor<fp16, [2, 1280, 1, 256]> var_5605_cast = add(x = out_35_cast, y = var_5604_to_fp16)[name = tensor<string, []>("op_5605_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_5607_to_fp16 = const()[name = tensor<string, []>("op_5607_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115158272)))];
            tensor<fp16, [2, 1280, 1, 256]> input_163_cast = mul(x = var_5605_cast, y = var_5607_to_fp16)[name = tensor<string, []>("input_163_cast")];
            tensor<int32, [2]> var_5615 = const()[name = tensor<string, []>("op_5615"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5617 = const()[name = tensor<string, []>("op_5617"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_5619_pad_type_0 = const()[name = tensor<string, []>("op_5619_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_5619_pad_0 = const()[name = tensor<string, []>("op_5619_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115160896))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124991360))), name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [7680]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124991552))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124999296))), name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([10240])];
            tensor<fp16, [2, 10240, 1, 256]> var_5619_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_5617, groups = var_4569, pad = var_5619_pad_0, pad_type = var_5619_pad_type_0, strides = var_5615, weight = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_163_cast)[name = tensor<string, []>("op_5619_cast")];
            tensor<int32, [2]> var_5620_split_sizes_0 = const()[name = tensor<string, []>("op_5620_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_5620_axis_0 = const()[name = tensor<string, []>("op_5620_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 256]> var_5620_cast_0, tensor<fp16, [2, 5120, 1, 256]> var_5620_cast_1 = split(axis = var_5620_axis_0, split_sizes = var_5620_split_sizes_0, x = var_5619_cast)[name = tensor<string, []>("op_5620_cast")];
            tensor<string, []> var_5622_mode_0 = const()[name = tensor<string, []>("op_5622_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 256]> var_5622_cast = gelu(mode = var_5622_mode_0, x = var_5620_cast_1)[name = tensor<string, []>("op_5622_cast")];
            tensor<fp16, [2, 5120, 1, 256]> input_165_cast = mul(x = var_5620_cast_0, y = var_5622_cast)[name = tensor<string, []>("input_165_cast")];
            tensor<int32, [2]> var_5626 = const()[name = tensor<string, []>("op_5626"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5628 = const()[name = tensor<string, []>("op_5628"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_5630_pad_type_0 = const()[name = tensor<string, []>("op_5630_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_5630_pad_0 = const()[name = tensor<string, []>("op_5630_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124999488))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129914752))), name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            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, []>(129914944)))];
            tensor<fp16, [2, 1280, 1, 256]> var_5630_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_5628, groups = var_4569, pad = var_5630_pad_0, pad_type = var_5630_pad_type_0, strides = var_5626, weight = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_165_cast)[name = tensor<string, []>("op_5630_cast")];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_105_cast = add(x = var_5630_cast, y = inputs_35_cast)[name = tensor<string, []>("hidden_states_105_cast")];
            tensor<int32, [4]> var_5632 = const()[name = tensor<string, []>("op_5632"), val = tensor<int32, [4]>([2, 1280, 16, 16])];
            tensor<fp16, [2, 1280, 16, 16]> input_167_cast = reshape(shape = var_5632, x = hidden_states_105_cast)[name = tensor<string, []>("input_167_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, []> 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129917568))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131146432))), name = tensor<string, []>("down_blocks_2_attentions_1_proj_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(131146624)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_107_cast = conv(bias = down_blocks_2_attentions_1_proj_out_bias_to_fp16, dilations = var_5638, groups = var_4569, pad = hidden_states_107_pad_0, pad_type = hidden_states_107_pad_type_0, strides = var_5636, weight = down_blocks_2_attentions_1_proj_out_weight_to_fp16_palettized, x = input_167_cast)[name = tensor<string, []>("hidden_states_107_cast")];
            tensor<fp16, [2, 1280, 16, 16]> input_169_cast = add(x = hidden_states_107_cast, y = hidden_states_95_cast)[name = tensor<string, []>("input_169_cast")];
            tensor<int32, [2]> var_5645 = const()[name = tensor<string, []>("op_5645"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [2]> var_5647 = const()[name = tensor<string, []>("op_5647"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [11059200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131149248))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142208512))), name = tensor<string, []>("down_blocks_2_downsamplers_0_conv_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 3, 3])];
            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, []>(142208704)))];
            tensor<fp16, [2, 1280, 8, 8]> input_171_cast = conv(bias = down_blocks_2_downsamplers_0_conv_bias_to_fp16, dilations = var_5647, groups = var_4569, pad = input_171_pad_0, pad_type = input_171_pad_type_0, strides = var_5645, weight = down_blocks_2_downsamplers_0_conv_weight_to_fp16_palettized, x = input_169_cast)[name = tensor<string, []>("input_171_cast")];
            tensor<int32, []> var_5659 = const()[name = tensor<string, []>("op_5659"), 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)[name = tensor<string, []>("reshape_72_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)[name = tensor<string, []>("reduce_mean_54_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> sub_36_cast = sub(x = reshape_72_cast, y = reduce_mean_54_cast)[name = tensor<string, []>("sub_36_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> square_18_cast = square(x = sub_36_cast)[name = tensor<string, []>("square_18_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)[name = tensor<string, []>("reduce_mean_56_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)[name = tensor<string, []>("add_36_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_18_cast = sqrt(x = add_36_cast)[name = tensor<string, []>("sqrt_18_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> real_div_18_cast = real_div(x = sub_36_cast, y = sqrt_18_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_73_cast")];
            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, []>(142211328)))];
            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, []>(142213952)))];
            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_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_73_cast)[name = tensor<string, []>("add_37_cast")];
            tensor<fp16, [2, 1280, 8, 8]> input_175_cast = silu(x = add_37_cast)[name = tensor<string, []>("input_175_cast")];
            tensor<int32, [2]> var_5675 = const()[name = tensor<string, []>("op_5675"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5677 = const()[name = tensor<string, []>("op_5677"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [11059200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142216576))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153275840))), name = tensor<string, []>("down_blocks_3_resnets_0_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 3, 3])];
            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, []>(153276032)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_109_cast = conv(bias = down_blocks_3_resnets_0_conv1_bias_to_fp16, dilations = var_5677, groups = var_5659, pad = hidden_states_109_pad_0, pad_type = hidden_states_109_pad_type_0, strides = var_5675, weight = down_blocks_3_resnets_0_conv1_weight_to_fp16_palettized, x = input_175_cast)[name = tensor<string, []>("hidden_states_109_cast")];
            tensor<int32, [2]> var_5683 = const()[name = tensor<string, []>("op_5683"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5685 = const()[name = tensor<string, []>("op_5685"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153278656))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154507520))), name = tensor<string, []>("down_blocks_3_resnets_0_time_emb_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(154507712)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_13_cast = conv(bias = down_blocks_3_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_5685, groups = var_5659, pad = temb_13_pad_0, pad_type = temb_13_pad_type_0, strides = var_5683, weight = down_blocks_3_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor<string, []>("temb_13_cast")];
            tensor<fp16, [2, 1280, 8, 8]> input_179_cast = add(x = hidden_states_109_cast, y = temb_13_cast)[name = tensor<string, []>("input_179_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)[name = tensor<string, []>("reshape_76_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)[name = tensor<string, []>("reduce_mean_57_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> sub_38_cast = sub(x = reshape_76_cast, y = reduce_mean_57_cast)[name = tensor<string, []>("sub_38_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> square_19_cast = square(x = sub_38_cast)[name = tensor<string, []>("square_19_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)[name = tensor<string, []>("reduce_mean_59_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)[name = tensor<string, []>("add_38_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_19_cast = sqrt(x = add_38_cast)[name = tensor<string, []>("sqrt_19_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> real_div_19_cast = real_div(x = sub_38_cast, y = sqrt_19_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_77_cast")];
            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, []>(154510336)))];
            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, []>(154512960)))];
            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_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_77_cast)[name = tensor<string, []>("add_39_cast")];
            tensor<fp16, [2, 1280, 8, 8]> input_183_cast = silu(x = add_39_cast)[name = tensor<string, []>("input_183_cast")];
            tensor<int32, [2]> var_5695 = const()[name = tensor<string, []>("op_5695"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5697 = const()[name = tensor<string, []>("op_5697"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [11059200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154515584))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165574848))), name = tensor<string, []>("down_blocks_3_resnets_0_conv2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 3, 3])];
            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, []>(165575040)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_111_cast = conv(bias = down_blocks_3_resnets_0_conv2_bias_to_fp16, dilations = var_5697, groups = var_5659, pad = hidden_states_111_pad_0, pad_type = hidden_states_111_pad_type_0, strides = var_5695, weight = down_blocks_3_resnets_0_conv2_weight_to_fp16_palettized, x = input_183_cast)[name = tensor<string, []>("hidden_states_111_cast")];
            tensor<fp16, [2, 1280, 8, 8]> input_185_cast = add(x = input_171_cast, y = hidden_states_111_cast)[name = tensor<string, []>("input_185_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)[name = tensor<string, []>("reshape_80_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)[name = tensor<string, []>("reduce_mean_60_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> sub_40_cast = sub(x = reshape_80_cast, y = reduce_mean_60_cast)[name = tensor<string, []>("sub_40_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> square_20_cast = square(x = sub_40_cast)[name = tensor<string, []>("square_20_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)[name = tensor<string, []>("reduce_mean_62_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)[name = tensor<string, []>("add_40_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_20_cast = sqrt(x = add_40_cast)[name = tensor<string, []>("sqrt_20_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> real_div_20_cast = real_div(x = sub_40_cast, y = sqrt_20_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_81_cast")];
            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, []>(165577664)))];
            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, []>(165580288)))];
            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_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_81_cast)[name = tensor<string, []>("add_41_cast")];
            tensor<fp16, [2, 1280, 8, 8]> input_189_cast = silu(x = add_41_cast)[name = tensor<string, []>("input_189_cast")];
            tensor<int32, [2]> var_5712 = const()[name = tensor<string, []>("op_5712"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5714 = const()[name = tensor<string, []>("op_5714"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [11059200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165582912))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(176642176))), name = tensor<string, []>("down_blocks_3_resnets_1_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 3, 3])];
            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, []>(176642368)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_113_cast = conv(bias = down_blocks_3_resnets_1_conv1_bias_to_fp16, dilations = var_5714, groups = var_5659, pad = hidden_states_113_pad_0, pad_type = hidden_states_113_pad_type_0, strides = var_5712, weight = down_blocks_3_resnets_1_conv1_weight_to_fp16_palettized, x = input_189_cast)[name = tensor<string, []>("hidden_states_113_cast")];
            tensor<int32, [2]> var_5720 = const()[name = tensor<string, []>("op_5720"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5722 = const()[name = tensor<string, []>("op_5722"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(176644992))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(177873856))), name = tensor<string, []>("down_blocks_3_resnets_1_time_emb_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(177874048)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_15_cast = conv(bias = down_blocks_3_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_5722, groups = var_5659, pad = temb_15_pad_0, pad_type = temb_15_pad_type_0, strides = var_5720, weight = down_blocks_3_resnets_1_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor<string, []>("temb_15_cast")];
            tensor<fp16, [2, 1280, 8, 8]> input_193_cast = add(x = hidden_states_113_cast, y = temb_15_cast)[name = tensor<string, []>("input_193_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)[name = tensor<string, []>("reshape_84_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)[name = tensor<string, []>("reduce_mean_63_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> sub_42_cast = sub(x = reshape_84_cast, y = reduce_mean_63_cast)[name = tensor<string, []>("sub_42_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> square_21_cast = square(x = sub_42_cast)[name = tensor<string, []>("square_21_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)[name = tensor<string, []>("reduce_mean_65_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)[name = tensor<string, []>("add_42_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_21_cast = sqrt(x = add_42_cast)[name = tensor<string, []>("sqrt_21_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> real_div_21_cast = real_div(x = sub_42_cast, y = sqrt_21_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_85_cast")];
            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, []>(177876672)))];
            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, []>(177879296)))];
            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_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_85_cast)[name = tensor<string, []>("add_43_cast")];
            tensor<fp16, [2, 1280, 8, 8]> input_197_cast = silu(x = add_43_cast)[name = tensor<string, []>("input_197_cast")];
            tensor<int32, [2]> var_5732 = const()[name = tensor<string, []>("op_5732"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5734 = const()[name = tensor<string, []>("op_5734"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [11059200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(177881920))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(188941184))), name = tensor<string, []>("down_blocks_3_resnets_1_conv2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 3, 3])];
            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, []>(188941376)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_115_cast = conv(bias = down_blocks_3_resnets_1_conv2_bias_to_fp16, dilations = var_5734, groups = var_5659, pad = hidden_states_115_pad_0, pad_type = hidden_states_115_pad_type_0, strides = var_5732, weight = down_blocks_3_resnets_1_conv2_weight_to_fp16_palettized, x = input_197_cast)[name = tensor<string, []>("hidden_states_115_cast")];
            tensor<fp16, [2, 1280, 8, 8]> input_199_cast = add(x = input_185_cast, y = hidden_states_115_cast)[name = tensor<string, []>("input_199_cast")];
            tensor<bool, []> var_5757 = const()[name = tensor<string, []>("op_5757"), val = tensor<bool, []>(true)];
            tensor<int32, []> var_5762 = const()[name = tensor<string, []>("op_5762"), 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)[name = tensor<string, []>("reshape_88_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)[name = tensor<string, []>("reduce_mean_66_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> sub_44_cast = sub(x = reshape_88_cast, y = reduce_mean_66_cast)[name = tensor<string, []>("sub_44_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> square_22_cast = square(x = sub_44_cast)[name = tensor<string, []>("square_22_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)[name = tensor<string, []>("reduce_mean_68_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)[name = tensor<string, []>("add_44_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_22_cast = sqrt(x = add_44_cast)[name = tensor<string, []>("sqrt_22_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> real_div_22_cast = real_div(x = sub_44_cast, y = sqrt_22_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_89_cast")];
            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, []>(188944000)))];
            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, []>(188946624)))];
            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_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_89_cast)[name = tensor<string, []>("add_45_cast")];
            tensor<fp16, [2, 1280, 8, 8]> input_203_cast = silu(x = add_45_cast)[name = tensor<string, []>("input_203_cast")];
            tensor<int32, [2]> var_5780 = const()[name = tensor<string, []>("op_5780"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5782 = const()[name = tensor<string, []>("op_5782"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [11059200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(188949248))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(200008512))), name = tensor<string, []>("mid_block_resnets_0_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 3, 3])];
            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, []>(200008704)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_117_cast = conv(bias = mid_block_resnets_0_conv1_bias_to_fp16, dilations = var_5782, groups = var_5762, pad = hidden_states_117_pad_0, pad_type = hidden_states_117_pad_type_0, strides = var_5780, weight = mid_block_resnets_0_conv1_weight_to_fp16_palettized, x = input_203_cast)[name = tensor<string, []>("hidden_states_117_cast")];
            tensor<int32, [2]> var_5788 = const()[name = tensor<string, []>("op_5788"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5790 = const()[name = tensor<string, []>("op_5790"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(200011328))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(201240192))), name = tensor<string, []>("mid_block_resnets_0_time_emb_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(201240384)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_17_cast = conv(bias = mid_block_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_5790, groups = var_5762, pad = temb_17_pad_0, pad_type = temb_17_pad_type_0, strides = var_5788, weight = mid_block_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor<string, []>("temb_17_cast")];
            tensor<fp16, [2, 1280, 8, 8]> input_207_cast = add(x = hidden_states_117_cast, y = temb_17_cast)[name = tensor<string, []>("input_207_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)[name = tensor<string, []>("reshape_92_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)[name = tensor<string, []>("reduce_mean_69_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> sub_46_cast = sub(x = reshape_92_cast, y = reduce_mean_69_cast)[name = tensor<string, []>("sub_46_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> square_23_cast = square(x = sub_46_cast)[name = tensor<string, []>("square_23_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)[name = tensor<string, []>("reduce_mean_71_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)[name = tensor<string, []>("add_46_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_23_cast = sqrt(x = add_46_cast)[name = tensor<string, []>("sqrt_23_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> real_div_23_cast = real_div(x = sub_46_cast, y = sqrt_23_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_93_cast")];
            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, []>(201243008)))];
            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, []>(201245632)))];
            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_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_93_cast)[name = tensor<string, []>("add_47_cast")];
            tensor<fp16, [2, 1280, 8, 8]> input_211_cast = silu(x = add_47_cast)[name = tensor<string, []>("input_211_cast")];
            tensor<int32, [2]> var_5800 = const()[name = tensor<string, []>("op_5800"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5802 = const()[name = tensor<string, []>("op_5802"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [11059200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(201248256))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(212307520))), name = tensor<string, []>("mid_block_resnets_0_conv2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 3, 3])];
            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, []>(212307712)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_119_cast = conv(bias = mid_block_resnets_0_conv2_bias_to_fp16, dilations = var_5802, groups = var_5762, pad = hidden_states_119_pad_0, pad_type = hidden_states_119_pad_type_0, strides = var_5800, weight = mid_block_resnets_0_conv2_weight_to_fp16_palettized, x = input_211_cast)[name = tensor<string, []>("hidden_states_119_cast")];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_121_cast = add(x = input_199_cast, y = hidden_states_119_cast)[name = tensor<string, []>("hidden_states_121_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)[name = tensor<string, []>("reshape_96_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)[name = tensor<string, []>("reduce_mean_72_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> sub_48_cast = sub(x = reshape_96_cast, y = reduce_mean_72_cast)[name = tensor<string, []>("sub_48_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> square_24_cast = square(x = sub_48_cast)[name = tensor<string, []>("square_24_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)[name = tensor<string, []>("reduce_mean_74_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)[name = tensor<string, []>("add_48_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_24_cast = sqrt(x = add_48_cast)[name = tensor<string, []>("sqrt_24_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> real_div_24_cast = real_div(x = sub_48_cast, y = sqrt_24_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_97_cast")];
            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, []>(212310336)))];
            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, []>(212312960)))];
            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_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_97_cast)[name = tensor<string, []>("add_49_cast")];
            tensor<int32, [2]> var_5822 = const()[name = tensor<string, []>("op_5822"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5824 = const()[name = tensor<string, []>("op_5824"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(212315584))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213544448))), name = tensor<string, []>("mid_block_attentions_0_proj_in_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(213544640)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_123_cast = conv(bias = mid_block_attentions_0_proj_in_bias_to_fp16, dilations = var_5824, groups = var_5762, pad = hidden_states_123_pad_0, pad_type = hidden_states_123_pad_type_0, strides = var_5822, weight = mid_block_attentions_0_proj_in_weight_to_fp16_palettized, x = add_49_cast)[name = tensor<string, []>("hidden_states_123_cast")];
            tensor<int32, [4]> var_5829 = const()[name = tensor<string, []>("op_5829"), val = tensor<int32, [4]>([2, 1280, 1, 64])];
            tensor<fp16, [2, 1280, 1, 64]> inputs_37_cast = reshape(shape = var_5829, x = hidden_states_123_cast)[name = tensor<string, []>("inputs_37_cast")];
            tensor<int32, [1]> var_5839 = const()[name = tensor<string, []>("op_5839"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 64]> channels_mean_37_cast = reduce_mean(axes = var_5839, keep_dims = var_5757, x = inputs_37_cast)[name = tensor<string, []>("channels_mean_37_cast")];
            tensor<fp16, [2, 1280, 1, 64]> zero_mean_37_cast = sub(x = inputs_37_cast, y = channels_mean_37_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_37_cast")];
            tensor<int32, [1]> var_5843 = const()[name = tensor<string, []>("op_5843"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 64]> var_5844_cast = reduce_mean(axes = var_5843, keep_dims = var_5757, x = zero_mean_sq_37_cast)[name = tensor<string, []>("op_5844_cast")];
            tensor<fp16, []> var_5845_to_fp16 = const()[name = tensor<string, []>("op_5845_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 64]> var_5846_cast = add(x = var_5844_cast, y = var_5845_to_fp16)[name = tensor<string, []>("op_5846_cast")];
            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_5846_cast)[name = tensor<string, []>("denom_37_cast")];
            tensor<fp16, [2, 1280, 1, 64]> out_37_cast = mul(x = zero_mean_37_cast, y = denom_37_cast)[name = tensor<string, []>("out_37_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_5850_to_fp16 = const()[name = tensor<string, []>("op_5850_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213547264)))];
            tensor<fp16, [2, 1280, 1, 64]> var_5851_cast = add(x = out_37_cast, y = var_5850_to_fp16)[name = tensor<string, []>("op_5851_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_5853_to_fp16 = const()[name = tensor<string, []>("op_5853_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213549888)))];
            tensor<fp16, [2, 1280, 1, 64]> hidden_states_125_cast = mul(x = var_5851_cast, y = var_5853_to_fp16)[name = tensor<string, []>("hidden_states_125_cast")];
            tensor<int32, [2]> var_5860 = const()[name = tensor<string, []>("op_5860"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5862 = const()[name = tensor<string, []>("op_5862"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213552512))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214781376))), name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 64]> q_25_cast = conv(dilations = var_5862, groups = var_5762, pad = q_25_pad_0, pad_type = q_25_pad_type_0, strides = var_5860, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_125_cast)[name = tensor<string, []>("q_25_cast")];
            tensor<int32, [2]> var_5866 = const()[name = tensor<string, []>("op_5866"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5868 = const()[name = tensor<string, []>("op_5868"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214781568))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216010432))), name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 64]> k_49_cast = conv(dilations = var_5868, groups = var_5762, pad = k_49_pad_0, pad_type = k_49_pad_type_0, strides = var_5866, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_125_cast)[name = tensor<string, []>("k_49_cast")];
            tensor<int32, [2]> var_5872 = const()[name = tensor<string, []>("op_5872"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5874 = const()[name = tensor<string, []>("op_5874"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216010624))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217239488))), name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 64]> v_25_cast = conv(dilations = var_5874, groups = var_5762, pad = v_25_pad_0, pad_type = v_25_pad_type_0, strides = var_5872, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_125_cast)[name = tensor<string, []>("v_25_cast")];
            tensor<int32, [4]> var_5878_begin_0 = const()[name = tensor<string, []>("op_5878_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5878_end_0 = const()[name = tensor<string, []>("op_5878_end_0"), val = tensor<int32, [4]>([2, 160, 1, 64])];
            tensor<bool, [4]> var_5878_end_mask_0 = const()[name = tensor<string, []>("op_5878_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 64]> var_5878_cast = slice_by_index(begin = var_5878_begin_0, end = var_5878_end_0, end_mask = var_5878_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_5878_cast")];
            tensor<int32, [4]> var_5882_begin_0 = const()[name = tensor<string, []>("op_5882_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_5882_end_0 = const()[name = tensor<string, []>("op_5882_end_0"), val = tensor<int32, [4]>([2, 320, 1, 64])];
            tensor<bool, [4]> var_5882_end_mask_0 = const()[name = tensor<string, []>("op_5882_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 64]> var_5882_cast = slice_by_index(begin = var_5882_begin_0, end = var_5882_end_0, end_mask = var_5882_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_5882_cast")];
            tensor<int32, [4]> var_5886_begin_0 = const()[name = tensor<string, []>("op_5886_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_5886_end_0 = const()[name = tensor<string, []>("op_5886_end_0"), val = tensor<int32, [4]>([2, 480, 1, 64])];
            tensor<bool, [4]> var_5886_end_mask_0 = const()[name = tensor<string, []>("op_5886_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 64]> var_5886_cast = slice_by_index(begin = var_5886_begin_0, end = var_5886_end_0, end_mask = var_5886_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_5886_cast")];
            tensor<int32, [4]> var_5890_begin_0 = const()[name = tensor<string, []>("op_5890_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_5890_end_0 = const()[name = tensor<string, []>("op_5890_end_0"), val = tensor<int32, [4]>([2, 640, 1, 64])];
            tensor<bool, [4]> var_5890_end_mask_0 = const()[name = tensor<string, []>("op_5890_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 64]> var_5890_cast = slice_by_index(begin = var_5890_begin_0, end = var_5890_end_0, end_mask = var_5890_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_5890_cast")];
            tensor<int32, [4]> var_5894_begin_0 = const()[name = tensor<string, []>("op_5894_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_5894_end_0 = const()[name = tensor<string, []>("op_5894_end_0"), val = tensor<int32, [4]>([2, 800, 1, 64])];
            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, 160, 1, 64]> var_5894_cast = slice_by_index(begin = var_5894_begin_0, end = var_5894_end_0, end_mask = var_5894_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_5894_cast")];
            tensor<int32, [4]> var_5898_begin_0 = const()[name = tensor<string, []>("op_5898_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_5898_end_0 = const()[name = tensor<string, []>("op_5898_end_0"), val = tensor<int32, [4]>([2, 960, 1, 64])];
            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, 160, 1, 64]> var_5898_cast = slice_by_index(begin = var_5898_begin_0, end = var_5898_end_0, end_mask = var_5898_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_5898_cast")];
            tensor<int32, [4]> var_5902_begin_0 = const()[name = tensor<string, []>("op_5902_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_5902_end_0 = const()[name = tensor<string, []>("op_5902_end_0"), val = tensor<int32, [4]>([2, 1120, 1, 64])];
            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, 160, 1, 64]> var_5902_cast = slice_by_index(begin = var_5902_begin_0, end = var_5902_end_0, end_mask = var_5902_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_5902_cast")];
            tensor<int32, [4]> var_5906_begin_0 = const()[name = tensor<string, []>("op_5906_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_5906_end_0 = const()[name = tensor<string, []>("op_5906_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 64])];
            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, 160, 1, 64]> var_5906_cast = slice_by_index(begin = var_5906_begin_0, end = var_5906_end_0, end_mask = var_5906_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_5906_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_5913_begin_0 = const()[name = tensor<string, []>("op_5913_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5913_end_0 = const()[name = tensor<string, []>("op_5913_end_0"), val = tensor<int32, [4]>([2, 64, 1, 160])];
            tensor<bool, [4]> var_5913_end_mask_0 = const()[name = tensor<string, []>("op_5913_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)[name = tensor<string, []>("transpose_19")];
            tensor<fp16, [2, 64, 1, 160]> var_5913_cast = slice_by_index(begin = var_5913_begin_0, end = var_5913_end_0, end_mask = var_5913_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_5913_cast")];
            tensor<int32, [4]> var_5917_begin_0 = const()[name = tensor<string, []>("op_5917_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_5917_end_0 = const()[name = tensor<string, []>("op_5917_end_0"), val = tensor<int32, [4]>([2, 64, 1, 320])];
            tensor<bool, [4]> var_5917_end_mask_0 = const()[name = tensor<string, []>("op_5917_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 160]> var_5917_cast = slice_by_index(begin = var_5917_begin_0, end = var_5917_end_0, end_mask = var_5917_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_5917_cast")];
            tensor<int32, [4]> var_5921_begin_0 = const()[name = tensor<string, []>("op_5921_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_5921_end_0 = const()[name = tensor<string, []>("op_5921_end_0"), val = tensor<int32, [4]>([2, 64, 1, 480])];
            tensor<bool, [4]> var_5921_end_mask_0 = const()[name = tensor<string, []>("op_5921_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 160]> var_5921_cast = slice_by_index(begin = var_5921_begin_0, end = var_5921_end_0, end_mask = var_5921_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_5921_cast")];
            tensor<int32, [4]> var_5925_begin_0 = const()[name = tensor<string, []>("op_5925_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 480])];
            tensor<int32, [4]> var_5925_end_0 = const()[name = tensor<string, []>("op_5925_end_0"), val = tensor<int32, [4]>([2, 64, 1, 640])];
            tensor<bool, [4]> var_5925_end_mask_0 = const()[name = tensor<string, []>("op_5925_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 160]> var_5925_cast = slice_by_index(begin = var_5925_begin_0, end = var_5925_end_0, end_mask = var_5925_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_5925_cast")];
            tensor<int32, [4]> var_5929_begin_0 = const()[name = tensor<string, []>("op_5929_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_5929_end_0 = const()[name = tensor<string, []>("op_5929_end_0"), val = tensor<int32, [4]>([2, 64, 1, 800])];
            tensor<bool, [4]> var_5929_end_mask_0 = const()[name = tensor<string, []>("op_5929_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 160]> var_5929_cast = slice_by_index(begin = var_5929_begin_0, end = var_5929_end_0, end_mask = var_5929_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_5929_cast")];
            tensor<int32, [4]> var_5933_begin_0 = const()[name = tensor<string, []>("op_5933_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 800])];
            tensor<int32, [4]> var_5933_end_0 = const()[name = tensor<string, []>("op_5933_end_0"), val = tensor<int32, [4]>([2, 64, 1, 960])];
            tensor<bool, [4]> var_5933_end_mask_0 = const()[name = tensor<string, []>("op_5933_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 160]> var_5933_cast = slice_by_index(begin = var_5933_begin_0, end = var_5933_end_0, end_mask = var_5933_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_5933_cast")];
            tensor<int32, [4]> var_5937_begin_0 = const()[name = tensor<string, []>("op_5937_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_5937_end_0 = const()[name = tensor<string, []>("op_5937_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1120])];
            tensor<bool, [4]> var_5937_end_mask_0 = const()[name = tensor<string, []>("op_5937_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 160]> var_5937_cast = slice_by_index(begin = var_5937_begin_0, end = var_5937_end_0, end_mask = var_5937_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_5937_cast")];
            tensor<int32, [4]> var_5941_begin_0 = const()[name = tensor<string, []>("op_5941_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1120])];
            tensor<int32, [4]> var_5941_end_0 = const()[name = tensor<string, []>("op_5941_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1280])];
            tensor<bool, [4]> var_5941_end_mask_0 = const()[name = tensor<string, []>("op_5941_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 160]> var_5941_cast = slice_by_index(begin = var_5941_begin_0, end = var_5941_end_0, end_mask = var_5941_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_5941_cast")];
            tensor<int32, [4]> var_5943_begin_0 = const()[name = tensor<string, []>("op_5943_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5943_end_0 = const()[name = tensor<string, []>("op_5943_end_0"), val = tensor<int32, [4]>([2, 160, 1, 64])];
            tensor<bool, [4]> var_5943_end_mask_0 = const()[name = tensor<string, []>("op_5943_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 64]> var_5943_cast = slice_by_index(begin = var_5943_begin_0, end = var_5943_end_0, end_mask = var_5943_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_5943_cast")];
            tensor<int32, [4]> var_5947_begin_0 = const()[name = tensor<string, []>("op_5947_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_5947_end_0 = const()[name = tensor<string, []>("op_5947_end_0"), val = tensor<int32, [4]>([2, 320, 1, 64])];
            tensor<bool, [4]> var_5947_end_mask_0 = const()[name = tensor<string, []>("op_5947_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 64]> var_5947_cast = slice_by_index(begin = var_5947_begin_0, end = var_5947_end_0, end_mask = var_5947_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_5947_cast")];
            tensor<int32, [4]> var_5951_begin_0 = const()[name = tensor<string, []>("op_5951_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_5951_end_0 = const()[name = tensor<string, []>("op_5951_end_0"), val = tensor<int32, [4]>([2, 480, 1, 64])];
            tensor<bool, [4]> var_5951_end_mask_0 = const()[name = tensor<string, []>("op_5951_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 64]> var_5951_cast = slice_by_index(begin = var_5951_begin_0, end = var_5951_end_0, end_mask = var_5951_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_5951_cast")];
            tensor<int32, [4]> var_5955_begin_0 = const()[name = tensor<string, []>("op_5955_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_5955_end_0 = const()[name = tensor<string, []>("op_5955_end_0"), val = tensor<int32, [4]>([2, 640, 1, 64])];
            tensor<bool, [4]> var_5955_end_mask_0 = const()[name = tensor<string, []>("op_5955_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 64]> var_5955_cast = slice_by_index(begin = var_5955_begin_0, end = var_5955_end_0, end_mask = var_5955_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_5955_cast")];
            tensor<int32, [4]> var_5959_begin_0 = const()[name = tensor<string, []>("op_5959_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_5959_end_0 = const()[name = tensor<string, []>("op_5959_end_0"), val = tensor<int32, [4]>([2, 800, 1, 64])];
            tensor<bool, [4]> var_5959_end_mask_0 = const()[name = tensor<string, []>("op_5959_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 64]> var_5959_cast = slice_by_index(begin = var_5959_begin_0, end = var_5959_end_0, end_mask = var_5959_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_5959_cast")];
            tensor<int32, [4]> var_5963_begin_0 = const()[name = tensor<string, []>("op_5963_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_5963_end_0 = const()[name = tensor<string, []>("op_5963_end_0"), val = tensor<int32, [4]>([2, 960, 1, 64])];
            tensor<bool, [4]> var_5963_end_mask_0 = const()[name = tensor<string, []>("op_5963_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 64]> var_5963_cast = slice_by_index(begin = var_5963_begin_0, end = var_5963_end_0, end_mask = var_5963_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_5963_cast")];
            tensor<int32, [4]> var_5967_begin_0 = const()[name = tensor<string, []>("op_5967_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_5967_end_0 = const()[name = tensor<string, []>("op_5967_end_0"), val = tensor<int32, [4]>([2, 1120, 1, 64])];
            tensor<bool, [4]> var_5967_end_mask_0 = const()[name = tensor<string, []>("op_5967_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 64]> var_5967_cast = slice_by_index(begin = var_5967_begin_0, end = var_5967_end_0, end_mask = var_5967_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_5967_cast")];
            tensor<int32, [4]> var_5971_begin_0 = const()[name = tensor<string, []>("op_5971_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_5971_end_0 = const()[name = tensor<string, []>("op_5971_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 64])];
            tensor<bool, [4]> var_5971_end_mask_0 = const()[name = tensor<string, []>("op_5971_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 64]> var_5971_cast = slice_by_index(begin = var_5971_begin_0, end = var_5971_end_0, end_mask = var_5971_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_5971_cast")];
            tensor<string, []> var_5975_equation_0 = const()[name = tensor<string, []>("op_5975_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_5975_cast = einsum(equation = var_5975_equation_0, values = (var_5913_cast, var_5878_cast))[name = tensor<string, []>("op_5975_cast")];
            tensor<fp16, []> var_5976_to_fp16 = const()[name = tensor<string, []>("op_5976_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 64, 1, 64]> aw_65_cast = mul(x = var_5975_cast, y = var_5976_to_fp16)[name = tensor<string, []>("aw_65_cast")];
            tensor<string, []> var_5979_equation_0 = const()[name = tensor<string, []>("op_5979_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_5979_cast = einsum(equation = var_5979_equation_0, values = (var_5917_cast, var_5882_cast))[name = tensor<string, []>("op_5979_cast")];
            tensor<fp16, []> var_5980_to_fp16 = const()[name = tensor<string, []>("op_5980_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 64, 1, 64]> aw_67_cast = mul(x = var_5979_cast, y = var_5980_to_fp16)[name = tensor<string, []>("aw_67_cast")];
            tensor<string, []> var_5983_equation_0 = const()[name = tensor<string, []>("op_5983_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_5983_cast = einsum(equation = var_5983_equation_0, values = (var_5921_cast, var_5886_cast))[name = tensor<string, []>("op_5983_cast")];
            tensor<fp16, []> var_5984_to_fp16 = const()[name = tensor<string, []>("op_5984_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 64, 1, 64]> aw_69_cast = mul(x = var_5983_cast, y = var_5984_to_fp16)[name = tensor<string, []>("aw_69_cast")];
            tensor<string, []> var_5987_equation_0 = const()[name = tensor<string, []>("op_5987_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_5987_cast = einsum(equation = var_5987_equation_0, values = (var_5925_cast, var_5890_cast))[name = tensor<string, []>("op_5987_cast")];
            tensor<fp16, []> var_5988_to_fp16 = const()[name = tensor<string, []>("op_5988_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 64, 1, 64]> aw_71_cast = mul(x = var_5987_cast, y = var_5988_to_fp16)[name = tensor<string, []>("aw_71_cast")];
            tensor<string, []> var_5991_equation_0 = const()[name = tensor<string, []>("op_5991_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_5991_cast = einsum(equation = var_5991_equation_0, values = (var_5929_cast, var_5894_cast))[name = tensor<string, []>("op_5991_cast")];
            tensor<fp16, []> var_5992_to_fp16 = const()[name = tensor<string, []>("op_5992_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 64, 1, 64]> aw_73_cast = mul(x = var_5991_cast, y = var_5992_to_fp16)[name = tensor<string, []>("aw_73_cast")];
            tensor<string, []> var_5995_equation_0 = const()[name = tensor<string, []>("op_5995_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_5995_cast = einsum(equation = var_5995_equation_0, values = (var_5933_cast, var_5898_cast))[name = tensor<string, []>("op_5995_cast")];
            tensor<fp16, []> var_5996_to_fp16 = const()[name = tensor<string, []>("op_5996_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 64, 1, 64]> aw_75_cast = mul(x = var_5995_cast, y = var_5996_to_fp16)[name = tensor<string, []>("aw_75_cast")];
            tensor<string, []> var_5999_equation_0 = const()[name = tensor<string, []>("op_5999_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_5999_cast = einsum(equation = var_5999_equation_0, values = (var_5937_cast, var_5902_cast))[name = tensor<string, []>("op_5999_cast")];
            tensor<fp16, []> var_6000_to_fp16 = const()[name = tensor<string, []>("op_6000_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 64, 1, 64]> aw_77_cast = mul(x = var_5999_cast, y = var_6000_to_fp16)[name = tensor<string, []>("aw_77_cast")];
            tensor<string, []> var_6003_equation_0 = const()[name = tensor<string, []>("op_6003_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_6003_cast = einsum(equation = var_6003_equation_0, values = (var_5941_cast, var_5906_cast))[name = tensor<string, []>("op_6003_cast")];
            tensor<fp16, []> var_6004_to_fp16 = const()[name = tensor<string, []>("op_6004_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 64, 1, 64]> aw_79_cast = mul(x = var_6003_cast, y = var_6004_to_fp16)[name = tensor<string, []>("aw_79_cast")];
            tensor<fp16, [2, 64, 1, 64]> var_6006_cast = softmax(axis = var_5762, x = aw_65_cast)[name = tensor<string, []>("op_6006_cast")];
            tensor<fp16, [2, 64, 1, 64]> var_6007_cast = softmax(axis = var_5762, x = aw_67_cast)[name = tensor<string, []>("op_6007_cast")];
            tensor<fp16, [2, 64, 1, 64]> var_6008_cast = softmax(axis = var_5762, x = aw_69_cast)[name = tensor<string, []>("op_6008_cast")];
            tensor<fp16, [2, 64, 1, 64]> var_6009_cast = softmax(axis = var_5762, x = aw_71_cast)[name = tensor<string, []>("op_6009_cast")];
            tensor<fp16, [2, 64, 1, 64]> var_6010_cast = softmax(axis = var_5762, x = aw_73_cast)[name = tensor<string, []>("op_6010_cast")];
            tensor<fp16, [2, 64, 1, 64]> var_6011_cast = softmax(axis = var_5762, x = aw_75_cast)[name = tensor<string, []>("op_6011_cast")];
            tensor<fp16, [2, 64, 1, 64]> var_6012_cast = softmax(axis = var_5762, x = aw_77_cast)[name = tensor<string, []>("op_6012_cast")];
            tensor<fp16, [2, 64, 1, 64]> var_6013_cast = softmax(axis = var_5762, x = aw_79_cast)[name = tensor<string, []>("op_6013_cast")];
            tensor<string, []> var_6015_equation_0 = const()[name = tensor<string, []>("op_6015_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 64]> var_6015_cast = einsum(equation = var_6015_equation_0, values = (var_5943_cast, var_6006_cast))[name = tensor<string, []>("op_6015_cast")];
            tensor<string, []> var_6017_equation_0 = const()[name = tensor<string, []>("op_6017_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 64]> var_6017_cast = einsum(equation = var_6017_equation_0, values = (var_5947_cast, var_6007_cast))[name = tensor<string, []>("op_6017_cast")];
            tensor<string, []> var_6019_equation_0 = const()[name = tensor<string, []>("op_6019_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 64]> var_6019_cast = einsum(equation = var_6019_equation_0, values = (var_5951_cast, var_6008_cast))[name = tensor<string, []>("op_6019_cast")];
            tensor<string, []> var_6021_equation_0 = const()[name = tensor<string, []>("op_6021_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 64]> var_6021_cast = einsum(equation = var_6021_equation_0, values = (var_5955_cast, var_6009_cast))[name = tensor<string, []>("op_6021_cast")];
            tensor<string, []> var_6023_equation_0 = const()[name = tensor<string, []>("op_6023_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 64]> var_6023_cast = einsum(equation = var_6023_equation_0, values = (var_5959_cast, var_6010_cast))[name = tensor<string, []>("op_6023_cast")];
            tensor<string, []> var_6025_equation_0 = const()[name = tensor<string, []>("op_6025_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 64]> var_6025_cast = einsum(equation = var_6025_equation_0, values = (var_5963_cast, var_6011_cast))[name = tensor<string, []>("op_6025_cast")];
            tensor<string, []> var_6027_equation_0 = const()[name = tensor<string, []>("op_6027_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 64]> var_6027_cast = einsum(equation = var_6027_equation_0, values = (var_5967_cast, var_6012_cast))[name = tensor<string, []>("op_6027_cast")];
            tensor<string, []> var_6029_equation_0 = const()[name = tensor<string, []>("op_6029_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 64]> var_6029_cast = einsum(equation = var_6029_equation_0, values = (var_5971_cast, var_6013_cast))[name = tensor<string, []>("op_6029_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_5762, interleave = input_215_interleave_0, values = (var_6015_cast, var_6017_cast, var_6019_cast, var_6021_cast, var_6023_cast, var_6025_cast, var_6027_cast, var_6029_cast))[name = tensor<string, []>("input_215_cast")];
            tensor<int32, [2]> var_6035 = const()[name = tensor<string, []>("op_6035"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6037 = const()[name = tensor<string, []>("op_6037"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_6039_pad_type_0 = const()[name = tensor<string, []>("op_6039_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_6039_pad_0 = const()[name = tensor<string, []>("op_6039_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217239680))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(218468544))), name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(218468736)))];
            tensor<fp16, [2, 1280, 1, 64]> var_6039_cast = conv(bias = mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_6037, groups = var_5762, pad = var_6039_pad_0, pad_type = var_6039_pad_type_0, strides = var_6035, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_215_cast)[name = tensor<string, []>("op_6039_cast")];
            tensor<fp16, [2, 1280, 1, 64]> inputs_39_cast = add(x = var_6039_cast, y = inputs_37_cast)[name = tensor<string, []>("inputs_39_cast")];
            tensor<int32, [1]> var_6043 = const()[name = tensor<string, []>("op_6043"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 64]> channels_mean_39_cast = reduce_mean(axes = var_6043, keep_dims = var_5757, x = inputs_39_cast)[name = tensor<string, []>("channels_mean_39_cast")];
            tensor<fp16, [2, 1280, 1, 64]> zero_mean_39_cast = sub(x = inputs_39_cast, y = channels_mean_39_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_39_cast")];
            tensor<int32, [1]> var_6047 = const()[name = tensor<string, []>("op_6047"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 64]> var_6048_cast = reduce_mean(axes = var_6047, keep_dims = var_5757, x = zero_mean_sq_39_cast)[name = tensor<string, []>("op_6048_cast")];
            tensor<fp16, []> var_6049_to_fp16 = const()[name = tensor<string, []>("op_6049_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 64]> var_6050_cast = add(x = var_6048_cast, y = var_6049_to_fp16)[name = tensor<string, []>("op_6050_cast")];
            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_6050_cast)[name = tensor<string, []>("denom_39_cast")];
            tensor<fp16, [2, 1280, 1, 64]> out_39_cast = mul(x = zero_mean_39_cast, y = denom_39_cast)[name = tensor<string, []>("out_39_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_6054_to_fp16 = const()[name = tensor<string, []>("op_6054_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(218471360)))];
            tensor<fp16, [2, 1280, 1, 64]> var_6055_cast = add(x = out_39_cast, y = var_6054_to_fp16)[name = tensor<string, []>("op_6055_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_6057_to_fp16 = const()[name = tensor<string, []>("op_6057_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(218473984)))];
            tensor<fp16, [2, 1280, 1, 64]> hidden_states_127_cast = mul(x = var_6055_cast, y = var_6057_to_fp16)[name = tensor<string, []>("hidden_states_127_cast")];
            tensor<int32, [2]> var_6064 = const()[name = tensor<string, []>("op_6064"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6066 = const()[name = tensor<string, []>("op_6066"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(218476608))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(219705472))), name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 64]> q_27_cast = conv(dilations = var_6066, groups = var_5762, pad = q_27_pad_0, pad_type = q_27_pad_type_0, strides = var_6064, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_127_cast)[name = tensor<string, []>("q_27_cast")];
            tensor<int32, [2]> var_6070 = const()[name = tensor<string, []>("op_6070"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6072 = const()[name = tensor<string, []>("op_6072"), 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, 768, 1, 1]> mid_block_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [737280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(219705664))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(220443008))), name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 768, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_53_cast = conv(dilations = var_6072, groups = var_5762, pad = k_53_pad_0, pad_type = k_53_pad_type_0, strides = var_6070, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_53_cast")];
            tensor<int32, [2]> var_6076 = const()[name = tensor<string, []>("op_6076"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6078 = const()[name = tensor<string, []>("op_6078"), 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, 768, 1, 1]> mid_block_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [737280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(220443200))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(221180544))), name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 768, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_27_cast = conv(dilations = var_6078, groups = var_5762, pad = v_27_pad_0, pad_type = v_27_pad_type_0, strides = var_6076, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_27_cast")];
            tensor<int32, [4]> var_6082_begin_0 = const()[name = tensor<string, []>("op_6082_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_6082_end_0 = const()[name = tensor<string, []>("op_6082_end_0"), val = tensor<int32, [4]>([2, 160, 1, 64])];
            tensor<bool, [4]> var_6082_end_mask_0 = const()[name = tensor<string, []>("op_6082_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 64]> var_6082_cast = slice_by_index(begin = var_6082_begin_0, end = var_6082_end_0, end_mask = var_6082_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_6082_cast")];
            tensor<int32, [4]> var_6086_begin_0 = const()[name = tensor<string, []>("op_6086_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_6086_end_0 = const()[name = tensor<string, []>("op_6086_end_0"), val = tensor<int32, [4]>([2, 320, 1, 64])];
            tensor<bool, [4]> var_6086_end_mask_0 = const()[name = tensor<string, []>("op_6086_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 64]> var_6086_cast = slice_by_index(begin = var_6086_begin_0, end = var_6086_end_0, end_mask = var_6086_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_6086_cast")];
            tensor<int32, [4]> var_6090_begin_0 = const()[name = tensor<string, []>("op_6090_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_6090_end_0 = const()[name = tensor<string, []>("op_6090_end_0"), val = tensor<int32, [4]>([2, 480, 1, 64])];
            tensor<bool, [4]> var_6090_end_mask_0 = const()[name = tensor<string, []>("op_6090_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 64]> var_6090_cast = slice_by_index(begin = var_6090_begin_0, end = var_6090_end_0, end_mask = var_6090_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_6090_cast")];
            tensor<int32, [4]> var_6094_begin_0 = const()[name = tensor<string, []>("op_6094_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_6094_end_0 = const()[name = tensor<string, []>("op_6094_end_0"), val = tensor<int32, [4]>([2, 640, 1, 64])];
            tensor<bool, [4]> var_6094_end_mask_0 = const()[name = tensor<string, []>("op_6094_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 64]> var_6094_cast = slice_by_index(begin = var_6094_begin_0, end = var_6094_end_0, end_mask = var_6094_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_6094_cast")];
            tensor<int32, [4]> var_6098_begin_0 = const()[name = tensor<string, []>("op_6098_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_6098_end_0 = const()[name = tensor<string, []>("op_6098_end_0"), val = tensor<int32, [4]>([2, 800, 1, 64])];
            tensor<bool, [4]> var_6098_end_mask_0 = const()[name = tensor<string, []>("op_6098_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 64]> var_6098_cast = slice_by_index(begin = var_6098_begin_0, end = var_6098_end_0, end_mask = var_6098_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_6098_cast")];
            tensor<int32, [4]> var_6102_begin_0 = const()[name = tensor<string, []>("op_6102_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_6102_end_0 = const()[name = tensor<string, []>("op_6102_end_0"), val = tensor<int32, [4]>([2, 960, 1, 64])];
            tensor<bool, [4]> var_6102_end_mask_0 = const()[name = tensor<string, []>("op_6102_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 64]> var_6102_cast = slice_by_index(begin = var_6102_begin_0, end = var_6102_end_0, end_mask = var_6102_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_6102_cast")];
            tensor<int32, [4]> var_6106_begin_0 = const()[name = tensor<string, []>("op_6106_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_6106_end_0 = const()[name = tensor<string, []>("op_6106_end_0"), val = tensor<int32, [4]>([2, 1120, 1, 64])];
            tensor<bool, [4]> var_6106_end_mask_0 = const()[name = tensor<string, []>("op_6106_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 64]> var_6106_cast = slice_by_index(begin = var_6106_begin_0, end = var_6106_end_0, end_mask = var_6106_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_6106_cast")];
            tensor<int32, [4]> var_6110_begin_0 = const()[name = tensor<string, []>("op_6110_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_6110_end_0 = const()[name = tensor<string, []>("op_6110_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 64])];
            tensor<bool, [4]> var_6110_end_mask_0 = const()[name = tensor<string, []>("op_6110_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 64]> var_6110_cast = slice_by_index(begin = var_6110_begin_0, end = var_6110_end_0, end_mask = var_6110_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_6110_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_6117_begin_0 = const()[name = tensor<string, []>("op_6117_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_6117_end_0 = const()[name = tensor<string, []>("op_6117_end_0"), val = tensor<int32, [4]>([2, 77, 1, 160])];
            tensor<bool, [4]> var_6117_end_mask_0 = const()[name = tensor<string, []>("op_6117_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)[name = tensor<string, []>("transpose_18")];
            tensor<fp16, [2, 77, 1, 160]> var_6117_cast = slice_by_index(begin = var_6117_begin_0, end = var_6117_end_0, end_mask = var_6117_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_6117_cast")];
            tensor<int32, [4]> var_6121_begin_0 = const()[name = tensor<string, []>("op_6121_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_6121_end_0 = const()[name = tensor<string, []>("op_6121_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_6121_end_mask_0 = const()[name = tensor<string, []>("op_6121_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_6121_cast = slice_by_index(begin = var_6121_begin_0, end = var_6121_end_0, end_mask = var_6121_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_6121_cast")];
            tensor<int32, [4]> var_6125_begin_0 = const()[name = tensor<string, []>("op_6125_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_6125_end_0 = const()[name = tensor<string, []>("op_6125_end_0"), val = tensor<int32, [4]>([2, 77, 1, 480])];
            tensor<bool, [4]> var_6125_end_mask_0 = const()[name = tensor<string, []>("op_6125_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_6125_cast = slice_by_index(begin = var_6125_begin_0, end = var_6125_end_0, end_mask = var_6125_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_6125_cast")];
            tensor<int32, [4]> var_6129_begin_0 = const()[name = tensor<string, []>("op_6129_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 480])];
            tensor<int32, [4]> var_6129_end_0 = const()[name = tensor<string, []>("op_6129_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_6129_end_mask_0 = const()[name = tensor<string, []>("op_6129_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_6129_cast = slice_by_index(begin = var_6129_begin_0, end = var_6129_end_0, end_mask = var_6129_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_6129_cast")];
            tensor<int32, [4]> var_6133_begin_0 = const()[name = tensor<string, []>("op_6133_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_6133_end_0 = const()[name = tensor<string, []>("op_6133_end_0"), val = tensor<int32, [4]>([2, 77, 1, 800])];
            tensor<bool, [4]> var_6133_end_mask_0 = const()[name = tensor<string, []>("op_6133_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_6133_cast = slice_by_index(begin = var_6133_begin_0, end = var_6133_end_0, end_mask = var_6133_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_6133_cast")];
            tensor<int32, [4]> var_6137_begin_0 = const()[name = tensor<string, []>("op_6137_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 800])];
            tensor<int32, [4]> var_6137_end_0 = const()[name = tensor<string, []>("op_6137_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_6137_end_mask_0 = const()[name = tensor<string, []>("op_6137_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_6137_cast = slice_by_index(begin = var_6137_begin_0, end = var_6137_end_0, end_mask = var_6137_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_6137_cast")];
            tensor<int32, [4]> var_6141_begin_0 = const()[name = tensor<string, []>("op_6141_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_6141_end_0 = const()[name = tensor<string, []>("op_6141_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1120])];
            tensor<bool, [4]> var_6141_end_mask_0 = const()[name = tensor<string, []>("op_6141_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_6141_cast = slice_by_index(begin = var_6141_begin_0, end = var_6141_end_0, end_mask = var_6141_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_6141_cast")];
            tensor<int32, [4]> var_6145_begin_0 = const()[name = tensor<string, []>("op_6145_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1120])];
            tensor<int32, [4]> var_6145_end_0 = const()[name = tensor<string, []>("op_6145_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_6145_end_mask_0 = const()[name = tensor<string, []>("op_6145_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_6145_cast = slice_by_index(begin = var_6145_begin_0, end = var_6145_end_0, end_mask = var_6145_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_6145_cast")];
            tensor<int32, [4]> var_6147_begin_0 = const()[name = tensor<string, []>("op_6147_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_6147_end_0 = const()[name = tensor<string, []>("op_6147_end_0"), val = tensor<int32, [4]>([2, 160, 1, 77])];
            tensor<bool, [4]> var_6147_end_mask_0 = const()[name = tensor<string, []>("op_6147_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_6147_cast = slice_by_index(begin = var_6147_begin_0, end = var_6147_end_0, end_mask = var_6147_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_6147_cast")];
            tensor<int32, [4]> var_6151_begin_0 = const()[name = tensor<string, []>("op_6151_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_6151_end_0 = const()[name = tensor<string, []>("op_6151_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_6151_end_mask_0 = const()[name = tensor<string, []>("op_6151_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_6151_cast = slice_by_index(begin = var_6151_begin_0, end = var_6151_end_0, end_mask = var_6151_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_6151_cast")];
            tensor<int32, [4]> var_6155_begin_0 = const()[name = tensor<string, []>("op_6155_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_6155_end_0 = const()[name = tensor<string, []>("op_6155_end_0"), val = tensor<int32, [4]>([2, 480, 1, 77])];
            tensor<bool, [4]> var_6155_end_mask_0 = const()[name = tensor<string, []>("op_6155_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_6155_cast = slice_by_index(begin = var_6155_begin_0, end = var_6155_end_0, end_mask = var_6155_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_6155_cast")];
            tensor<int32, [4]> var_6159_begin_0 = const()[name = tensor<string, []>("op_6159_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_6159_end_0 = const()[name = tensor<string, []>("op_6159_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_6159_end_mask_0 = const()[name = tensor<string, []>("op_6159_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_6159_cast = slice_by_index(begin = var_6159_begin_0, end = var_6159_end_0, end_mask = var_6159_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_6159_cast")];
            tensor<int32, [4]> var_6163_begin_0 = const()[name = tensor<string, []>("op_6163_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_6163_end_0 = const()[name = tensor<string, []>("op_6163_end_0"), val = tensor<int32, [4]>([2, 800, 1, 77])];
            tensor<bool, [4]> var_6163_end_mask_0 = const()[name = tensor<string, []>("op_6163_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_6163_cast = slice_by_index(begin = var_6163_begin_0, end = var_6163_end_0, end_mask = var_6163_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_6163_cast")];
            tensor<int32, [4]> var_6167_begin_0 = const()[name = tensor<string, []>("op_6167_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_6167_end_0 = const()[name = tensor<string, []>("op_6167_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_6167_end_mask_0 = const()[name = tensor<string, []>("op_6167_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_6167_cast = slice_by_index(begin = var_6167_begin_0, end = var_6167_end_0, end_mask = var_6167_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_6167_cast")];
            tensor<int32, [4]> var_6171_begin_0 = const()[name = tensor<string, []>("op_6171_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_6171_end_0 = const()[name = tensor<string, []>("op_6171_end_0"), val = tensor<int32, [4]>([2, 1120, 1, 77])];
            tensor<bool, [4]> var_6171_end_mask_0 = const()[name = tensor<string, []>("op_6171_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_6171_cast = slice_by_index(begin = var_6171_begin_0, end = var_6171_end_0, end_mask = var_6171_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_6171_cast")];
            tensor<int32, [4]> var_6175_begin_0 = const()[name = tensor<string, []>("op_6175_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_6175_end_0 = const()[name = tensor<string, []>("op_6175_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_6175_end_mask_0 = const()[name = tensor<string, []>("op_6175_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_6175_cast = slice_by_index(begin = var_6175_begin_0, end = var_6175_end_0, end_mask = var_6175_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_6175_cast")];
            tensor<string, []> var_6179_equation_0 = const()[name = tensor<string, []>("op_6179_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_6179_cast = einsum(equation = var_6179_equation_0, values = (var_6117_cast, var_6082_cast))[name = tensor<string, []>("op_6179_cast")];
            tensor<fp16, []> var_6180_to_fp16 = const()[name = tensor<string, []>("op_6180_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 64]> aw_81_cast = mul(x = var_6179_cast, y = var_6180_to_fp16)[name = tensor<string, []>("aw_81_cast")];
            tensor<string, []> var_6183_equation_0 = const()[name = tensor<string, []>("op_6183_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_6183_cast = einsum(equation = var_6183_equation_0, values = (var_6121_cast, var_6086_cast))[name = tensor<string, []>("op_6183_cast")];
            tensor<fp16, []> var_6184_to_fp16 = const()[name = tensor<string, []>("op_6184_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 64]> aw_83_cast = mul(x = var_6183_cast, y = var_6184_to_fp16)[name = tensor<string, []>("aw_83_cast")];
            tensor<string, []> var_6187_equation_0 = const()[name = tensor<string, []>("op_6187_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_6187_cast = einsum(equation = var_6187_equation_0, values = (var_6125_cast, var_6090_cast))[name = tensor<string, []>("op_6187_cast")];
            tensor<fp16, []> var_6188_to_fp16 = const()[name = tensor<string, []>("op_6188_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 64]> aw_85_cast = mul(x = var_6187_cast, y = var_6188_to_fp16)[name = tensor<string, []>("aw_85_cast")];
            tensor<string, []> var_6191_equation_0 = const()[name = tensor<string, []>("op_6191_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_6191_cast = einsum(equation = var_6191_equation_0, values = (var_6129_cast, var_6094_cast))[name = tensor<string, []>("op_6191_cast")];
            tensor<fp16, []> var_6192_to_fp16 = const()[name = tensor<string, []>("op_6192_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 64]> aw_87_cast = mul(x = var_6191_cast, y = var_6192_to_fp16)[name = tensor<string, []>("aw_87_cast")];
            tensor<string, []> var_6195_equation_0 = const()[name = tensor<string, []>("op_6195_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_6195_cast = einsum(equation = var_6195_equation_0, values = (var_6133_cast, var_6098_cast))[name = tensor<string, []>("op_6195_cast")];
            tensor<fp16, []> var_6196_to_fp16 = const()[name = tensor<string, []>("op_6196_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 64]> aw_89_cast = mul(x = var_6195_cast, y = var_6196_to_fp16)[name = tensor<string, []>("aw_89_cast")];
            tensor<string, []> var_6199_equation_0 = const()[name = tensor<string, []>("op_6199_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_6199_cast = einsum(equation = var_6199_equation_0, values = (var_6137_cast, var_6102_cast))[name = tensor<string, []>("op_6199_cast")];
            tensor<fp16, []> var_6200_to_fp16 = const()[name = tensor<string, []>("op_6200_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 64]> aw_91_cast = mul(x = var_6199_cast, y = var_6200_to_fp16)[name = tensor<string, []>("aw_91_cast")];
            tensor<string, []> var_6203_equation_0 = const()[name = tensor<string, []>("op_6203_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_6203_cast = einsum(equation = var_6203_equation_0, values = (var_6141_cast, var_6106_cast))[name = tensor<string, []>("op_6203_cast")];
            tensor<fp16, []> var_6204_to_fp16 = const()[name = tensor<string, []>("op_6204_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 64]> aw_93_cast = mul(x = var_6203_cast, y = var_6204_to_fp16)[name = tensor<string, []>("aw_93_cast")];
            tensor<string, []> var_6207_equation_0 = const()[name = tensor<string, []>("op_6207_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_6207_cast = einsum(equation = var_6207_equation_0, values = (var_6145_cast, var_6110_cast))[name = tensor<string, []>("op_6207_cast")];
            tensor<fp16, []> var_6208_to_fp16 = const()[name = tensor<string, []>("op_6208_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 64]> aw_95_cast = mul(x = var_6207_cast, y = var_6208_to_fp16)[name = tensor<string, []>("aw_95_cast")];
            tensor<fp16, [2, 77, 1, 64]> var_6210_cast = softmax(axis = var_5762, x = aw_81_cast)[name = tensor<string, []>("op_6210_cast")];
            tensor<fp16, [2, 77, 1, 64]> var_6211_cast = softmax(axis = var_5762, x = aw_83_cast)[name = tensor<string, []>("op_6211_cast")];
            tensor<fp16, [2, 77, 1, 64]> var_6212_cast = softmax(axis = var_5762, x = aw_85_cast)[name = tensor<string, []>("op_6212_cast")];
            tensor<fp16, [2, 77, 1, 64]> var_6213_cast = softmax(axis = var_5762, x = aw_87_cast)[name = tensor<string, []>("op_6213_cast")];
            tensor<fp16, [2, 77, 1, 64]> var_6214_cast = softmax(axis = var_5762, x = aw_89_cast)[name = tensor<string, []>("op_6214_cast")];
            tensor<fp16, [2, 77, 1, 64]> var_6215_cast = softmax(axis = var_5762, x = aw_91_cast)[name = tensor<string, []>("op_6215_cast")];
            tensor<fp16, [2, 77, 1, 64]> var_6216_cast = softmax(axis = var_5762, x = aw_93_cast)[name = tensor<string, []>("op_6216_cast")];
            tensor<fp16, [2, 77, 1, 64]> var_6217_cast = softmax(axis = var_5762, x = aw_95_cast)[name = tensor<string, []>("op_6217_cast")];
            tensor<string, []> var_6219_equation_0 = const()[name = tensor<string, []>("op_6219_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 64]> var_6219_cast = einsum(equation = var_6219_equation_0, values = (var_6147_cast, var_6210_cast))[name = tensor<string, []>("op_6219_cast")];
            tensor<string, []> var_6221_equation_0 = const()[name = tensor<string, []>("op_6221_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 64]> var_6221_cast = einsum(equation = var_6221_equation_0, values = (var_6151_cast, var_6211_cast))[name = tensor<string, []>("op_6221_cast")];
            tensor<string, []> var_6223_equation_0 = const()[name = tensor<string, []>("op_6223_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 64]> var_6223_cast = einsum(equation = var_6223_equation_0, values = (var_6155_cast, var_6212_cast))[name = tensor<string, []>("op_6223_cast")];
            tensor<string, []> var_6225_equation_0 = const()[name = tensor<string, []>("op_6225_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 64]> var_6225_cast = einsum(equation = var_6225_equation_0, values = (var_6159_cast, var_6213_cast))[name = tensor<string, []>("op_6225_cast")];
            tensor<string, []> var_6227_equation_0 = const()[name = tensor<string, []>("op_6227_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 64]> var_6227_cast = einsum(equation = var_6227_equation_0, values = (var_6163_cast, var_6214_cast))[name = tensor<string, []>("op_6227_cast")];
            tensor<string, []> var_6229_equation_0 = const()[name = tensor<string, []>("op_6229_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 64]> var_6229_cast = einsum(equation = var_6229_equation_0, values = (var_6167_cast, var_6215_cast))[name = tensor<string, []>("op_6229_cast")];
            tensor<string, []> var_6231_equation_0 = const()[name = tensor<string, []>("op_6231_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 64]> var_6231_cast = einsum(equation = var_6231_equation_0, values = (var_6171_cast, var_6216_cast))[name = tensor<string, []>("op_6231_cast")];
            tensor<string, []> var_6233_equation_0 = const()[name = tensor<string, []>("op_6233_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 64]> var_6233_cast = einsum(equation = var_6233_equation_0, values = (var_6175_cast, var_6217_cast))[name = tensor<string, []>("op_6233_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_5762, interleave = input_217_interleave_0, values = (var_6219_cast, var_6221_cast, var_6223_cast, var_6225_cast, var_6227_cast, var_6229_cast, var_6231_cast, var_6233_cast))[name = tensor<string, []>("input_217_cast")];
            tensor<int32, [2]> var_6239 = const()[name = tensor<string, []>("op_6239"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6241 = const()[name = tensor<string, []>("op_6241"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_6243_pad_type_0 = const()[name = tensor<string, []>("op_6243_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_6243_pad_0 = const()[name = tensor<string, []>("op_6243_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(221180736))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(222409600))), name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(222409792)))];
            tensor<fp16, [2, 1280, 1, 64]> var_6243_cast = conv(bias = mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_6241, groups = var_5762, pad = var_6243_pad_0, pad_type = var_6243_pad_type_0, strides = var_6239, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_217_cast)[name = tensor<string, []>("op_6243_cast")];
            tensor<fp16, [2, 1280, 1, 64]> inputs_41_cast = add(x = var_6243_cast, y = inputs_39_cast)[name = tensor<string, []>("inputs_41_cast")];
            tensor<int32, [1]> var_6247 = const()[name = tensor<string, []>("op_6247"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 64]> channels_mean_41_cast = reduce_mean(axes = var_6247, keep_dims = var_5757, x = inputs_41_cast)[name = tensor<string, []>("channels_mean_41_cast")];
            tensor<fp16, [2, 1280, 1, 64]> zero_mean_41_cast = sub(x = inputs_41_cast, y = channels_mean_41_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_41_cast")];
            tensor<int32, [1]> var_6251 = const()[name = tensor<string, []>("op_6251"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 64]> var_6252_cast = reduce_mean(axes = var_6251, keep_dims = var_5757, x = zero_mean_sq_41_cast)[name = tensor<string, []>("op_6252_cast")];
            tensor<fp16, []> var_6253_to_fp16 = const()[name = tensor<string, []>("op_6253_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 64]> var_6254_cast = add(x = var_6252_cast, y = var_6253_to_fp16)[name = tensor<string, []>("op_6254_cast")];
            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_6254_cast)[name = tensor<string, []>("denom_41_cast")];
            tensor<fp16, [2, 1280, 1, 64]> out_41_cast = mul(x = zero_mean_41_cast, y = denom_41_cast)[name = tensor<string, []>("out_41_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_6258_to_fp16 = const()[name = tensor<string, []>("op_6258_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(222412416)))];
            tensor<fp16, [2, 1280, 1, 64]> var_6259_cast = add(x = out_41_cast, y = var_6258_to_fp16)[name = tensor<string, []>("op_6259_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_6261_to_fp16 = const()[name = tensor<string, []>("op_6261_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(222415040)))];
            tensor<fp16, [2, 1280, 1, 64]> input_219_cast = mul(x = var_6259_cast, y = var_6261_to_fp16)[name = tensor<string, []>("input_219_cast")];
            tensor<int32, [2]> var_6269 = const()[name = tensor<string, []>("op_6269"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6271 = const()[name = tensor<string, []>("op_6271"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_6273_pad_type_0 = const()[name = tensor<string, []>("op_6273_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_6273_pad_0 = const()[name = tensor<string, []>("op_6273_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(222417664))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(232248128))), name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [7680]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(232248320))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(232256064))), name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([10240])];
            tensor<fp16, [2, 10240, 1, 64]> var_6273_cast = conv(bias = mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_6271, groups = var_5762, pad = var_6273_pad_0, pad_type = var_6273_pad_type_0, strides = var_6269, weight = mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_219_cast)[name = tensor<string, []>("op_6273_cast")];
            tensor<int32, [2]> var_6274_split_sizes_0 = const()[name = tensor<string, []>("op_6274_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_6274_axis_0 = const()[name = tensor<string, []>("op_6274_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 64]> var_6274_cast_0, tensor<fp16, [2, 5120, 1, 64]> var_6274_cast_1 = split(axis = var_6274_axis_0, split_sizes = var_6274_split_sizes_0, x = var_6273_cast)[name = tensor<string, []>("op_6274_cast")];
            tensor<string, []> var_6276_mode_0 = const()[name = tensor<string, []>("op_6276_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 64]> var_6276_cast = gelu(mode = var_6276_mode_0, x = var_6274_cast_1)[name = tensor<string, []>("op_6276_cast")];
            tensor<fp16, [2, 5120, 1, 64]> input_221_cast = mul(x = var_6274_cast_0, y = var_6276_cast)[name = tensor<string, []>("input_221_cast")];
            tensor<int32, [2]> var_6280 = const()[name = tensor<string, []>("op_6280"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6282 = const()[name = tensor<string, []>("op_6282"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_6284_pad_type_0 = const()[name = tensor<string, []>("op_6284_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_6284_pad_0 = const()[name = tensor<string, []>("op_6284_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(232256256))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(237171520))), name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            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, []>(237171712)))];
            tensor<fp16, [2, 1280, 1, 64]> var_6284_cast = conv(bias = mid_block_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_6282, groups = var_5762, pad = var_6284_pad_0, pad_type = var_6284_pad_type_0, strides = var_6280, weight = mid_block_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_221_cast)[name = tensor<string, []>("op_6284_cast")];
            tensor<fp16, [2, 1280, 1, 64]> hidden_states_131_cast = add(x = var_6284_cast, y = inputs_41_cast)[name = tensor<string, []>("hidden_states_131_cast")];
            tensor<int32, [4]> var_6286 = const()[name = tensor<string, []>("op_6286"), val = tensor<int32, [4]>([2, 1280, 8, 8])];
            tensor<fp16, [2, 1280, 8, 8]> input_223_cast = reshape(shape = var_6286, x = hidden_states_131_cast)[name = tensor<string, []>("input_223_cast")];
            tensor<int32, [2]> var_6290 = const()[name = tensor<string, []>("op_6290"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6292 = const()[name = tensor<string, []>("op_6292"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(237174336))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(238403200))), name = tensor<string, []>("mid_block_attentions_0_proj_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(238403392)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_133_cast = conv(bias = mid_block_attentions_0_proj_out_bias_to_fp16, dilations = var_6292, groups = var_5762, pad = hidden_states_133_pad_0, pad_type = hidden_states_133_pad_type_0, strides = var_6290, weight = mid_block_attentions_0_proj_out_weight_to_fp16_palettized, x = input_223_cast)[name = tensor<string, []>("hidden_states_133_cast")];
            tensor<fp16, [2, 1280, 8, 8]> input_225_cast = add(x = hidden_states_133_cast, y = hidden_states_121_cast)[name = tensor<string, []>("input_225_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)[name = tensor<string, []>("reshape_100_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)[name = tensor<string, []>("reduce_mean_75_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> sub_50_cast = sub(x = reshape_100_cast, y = reduce_mean_75_cast)[name = tensor<string, []>("sub_50_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> square_25_cast = square(x = sub_50_cast)[name = tensor<string, []>("square_25_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)[name = tensor<string, []>("reduce_mean_77_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)[name = tensor<string, []>("add_50_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_25_cast = sqrt(x = add_50_cast)[name = tensor<string, []>("sqrt_25_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> real_div_25_cast = real_div(x = sub_50_cast, y = sqrt_25_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_101_cast")];
            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, []>(238406016)))];
            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, []>(238408640)))];
            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_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_101_cast)[name = tensor<string, []>("add_51_cast")];
            tensor<fp16, [2, 1280, 8, 8]> input_229_cast = silu(x = add_51_cast)[name = tensor<string, []>("input_229_cast")];
            tensor<int32, [2]> var_6307 = const()[name = tensor<string, []>("op_6307"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6309 = const()[name = tensor<string, []>("op_6309"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [11059200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(238411264))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(249470528))), name = tensor<string, []>("mid_block_resnets_1_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 3, 3])];
            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, []>(249470720)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_135_cast = conv(bias = mid_block_resnets_1_conv1_bias_to_fp16, dilations = var_6309, groups = var_5762, pad = hidden_states_135_pad_0, pad_type = hidden_states_135_pad_type_0, strides = var_6307, weight = mid_block_resnets_1_conv1_weight_to_fp16_palettized, x = input_229_cast)[name = tensor<string, []>("hidden_states_135_cast")];
            tensor<int32, [2]> var_6315 = const()[name = tensor<string, []>("op_6315"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6317 = const()[name = tensor<string, []>("op_6317"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(249473344))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(250702208))), name = tensor<string, []>("mid_block_resnets_1_time_emb_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(250702400)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_19_cast = conv(bias = mid_block_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_6317, groups = var_5762, pad = temb_19_pad_0, pad_type = temb_19_pad_type_0, strides = var_6315, weight = mid_block_resnets_1_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor<string, []>("temb_19_cast")];
            tensor<fp16, [2, 1280, 8, 8]> input_233_cast = add(x = hidden_states_135_cast, y = temb_19_cast)[name = tensor<string, []>("input_233_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)[name = tensor<string, []>("reshape_104_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)[name = tensor<string, []>("reduce_mean_78_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> sub_52_cast = sub(x = reshape_104_cast, y = reduce_mean_78_cast)[name = tensor<string, []>("sub_52_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> square_26_cast = square(x = sub_52_cast)[name = tensor<string, []>("square_26_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)[name = tensor<string, []>("reduce_mean_80_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)[name = tensor<string, []>("add_52_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_26_cast = sqrt(x = add_52_cast)[name = tensor<string, []>("sqrt_26_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> real_div_26_cast = real_div(x = sub_52_cast, y = sqrt_26_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_105_cast")];
            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, []>(250705024)))];
            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, []>(250707648)))];
            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_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_105_cast)[name = tensor<string, []>("add_53_cast")];
            tensor<fp16, [2, 1280, 8, 8]> input_237_cast = silu(x = add_53_cast)[name = tensor<string, []>("input_237_cast")];
            tensor<int32, [2]> var_6327 = const()[name = tensor<string, []>("op_6327"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6329 = const()[name = tensor<string, []>("op_6329"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [11059200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(250710272))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(261769536))), name = tensor<string, []>("mid_block_resnets_1_conv2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 3, 3])];
            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, []>(261769728)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_137_cast = conv(bias = mid_block_resnets_1_conv2_bias_to_fp16, dilations = var_6329, groups = var_5762, pad = hidden_states_137_pad_0, pad_type = hidden_states_137_pad_type_0, strides = var_6327, weight = mid_block_resnets_1_conv2_weight_to_fp16_palettized, x = input_237_cast)[name = tensor<string, []>("hidden_states_137_cast")];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_139_cast = add(x = input_225_cast, y = hidden_states_137_cast)[name = tensor<string, []>("hidden_states_139_cast")];
            tensor<int32, []> var_6340 = const()[name = tensor<string, []>("op_6340"), 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_6340, interleave = input_239_interleave_0, values = (hidden_states_139_cast, input_199_cast))[name = tensor<string, []>("input_239_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)[name = tensor<string, []>("reshape_108_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)[name = tensor<string, []>("reduce_mean_81_cast")];
            tensor<fp16, [2, 32, 80, 8, 8]> sub_54_cast = sub(x = reshape_108_cast, y = reduce_mean_81_cast)[name = tensor<string, []>("sub_54_cast")];
            tensor<fp16, [2, 32, 80, 8, 8]> square_27_cast = square(x = sub_54_cast)[name = tensor<string, []>("square_27_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)[name = tensor<string, []>("reduce_mean_83_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)[name = tensor<string, []>("add_54_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_27_cast = sqrt(x = add_54_cast)[name = tensor<string, []>("sqrt_27_cast")];
            tensor<fp16, [2, 32, 80, 8, 8]> real_div_27_cast = real_div(x = sub_54_cast, y = sqrt_27_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_109_cast")];
            tensor<fp16, [2560]> add_55_mean_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(261772352))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(261774336))), name = tensor<string, []>("add_55_mean_0_to_fp16_palettized"), shape = tensor<uint32, [1]>([2560])];
            tensor<fp16, [2560]> add_55_variance_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(261774528))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(261776512))), name = tensor<string, []>("add_55_variance_0_to_fp16_palettized"), shape = tensor<uint32, [1]>([2560])];
            tensor<fp16, [2560]> add_55_gamma_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(261776704))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(261778688))), name = tensor<string, []>("add_55_gamma_0_to_fp16_palettized"), shape = tensor<uint32, [1]>([2560])];
            tensor<fp16, [2560]> add_55_beta_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(261778880))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(261780864))), name = tensor<string, []>("add_55_beta_0_to_fp16_palettized"), shape = tensor<uint32, [1]>([2560])];
            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_palettized, epsilon = add_55_epsilon_0_to_fp16, gamma = add_55_gamma_0_to_fp16_palettized, mean = add_55_mean_0_to_fp16_palettized, variance = add_55_variance_0_to_fp16_palettized, x = reshape_109_cast)[name = tensor<string, []>("add_55_cast")];
            tensor<fp16, [2, 2560, 8, 8]> input_243_cast = silu(x = add_55_cast)[name = tensor<string, []>("input_243_cast")];
            tensor<int32, [2]> var_6363 = const()[name = tensor<string, []>("op_6363"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6365 = const()[name = tensor<string, []>("op_6365"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [22118400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(261781056))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(283899520))), name = tensor<string, []>("up_blocks_0_resnets_0_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2560, 3, 3])];
            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, []>(283899712)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_141_cast = conv(bias = up_blocks_0_resnets_0_conv1_bias_to_fp16, dilations = var_6365, groups = var_6340, pad = hidden_states_141_pad_0, pad_type = hidden_states_141_pad_type_0, strides = var_6363, weight = up_blocks_0_resnets_0_conv1_weight_to_fp16_palettized, x = input_243_cast)[name = tensor<string, []>("hidden_states_141_cast")];
            tensor<int32, [2]> var_6371 = const()[name = tensor<string, []>("op_6371"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6373 = const()[name = tensor<string, []>("op_6373"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(283902336))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(285131200))), name = tensor<string, []>("up_blocks_0_resnets_0_time_emb_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(285131392)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_21_cast = conv(bias = up_blocks_0_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_6373, groups = var_6340, pad = temb_21_pad_0, pad_type = temb_21_pad_type_0, strides = var_6371, weight = up_blocks_0_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor<string, []>("temb_21_cast")];
            tensor<fp16, [2, 1280, 8, 8]> input_247_cast = add(x = hidden_states_141_cast, y = temb_21_cast)[name = tensor<string, []>("input_247_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)[name = tensor<string, []>("reshape_112_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)[name = tensor<string, []>("reduce_mean_84_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> sub_56_cast = sub(x = reshape_112_cast, y = reduce_mean_84_cast)[name = tensor<string, []>("sub_56_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> square_28_cast = square(x = sub_56_cast)[name = tensor<string, []>("square_28_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)[name = tensor<string, []>("reduce_mean_86_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)[name = tensor<string, []>("add_56_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_28_cast = sqrt(x = add_56_cast)[name = tensor<string, []>("sqrt_28_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> real_div_28_cast = real_div(x = sub_56_cast, y = sqrt_28_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_113_cast")];
            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, []>(285134016)))];
            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, []>(285136640)))];
            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_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_113_cast)[name = tensor<string, []>("add_57_cast")];
            tensor<fp16, [2, 1280, 8, 8]> input_251_cast = silu(x = add_57_cast)[name = tensor<string, []>("input_251_cast")];
            tensor<int32, [2]> var_6383 = const()[name = tensor<string, []>("op_6383"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6385 = const()[name = tensor<string, []>("op_6385"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [11059200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(285139264))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(296198528))), name = tensor<string, []>("up_blocks_0_resnets_0_conv2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 3, 3])];
            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, []>(296198720)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_143_cast = conv(bias = up_blocks_0_resnets_0_conv2_bias_to_fp16, dilations = var_6385, groups = var_6340, pad = hidden_states_143_pad_0, pad_type = hidden_states_143_pad_type_0, strides = var_6383, weight = up_blocks_0_resnets_0_conv2_weight_to_fp16_palettized, x = input_251_cast)[name = tensor<string, []>("hidden_states_143_cast")];
            tensor<int32, [2]> var_6390 = const()[name = tensor<string, []>("op_6390"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6392 = const()[name = tensor<string, []>("op_6392"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2457600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(296201344))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(298659008))), name = tensor<string, []>("up_blocks_0_resnets_0_conv_shortcut_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2560, 1, 1])];
            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, []>(298659200)))];
            tensor<fp16, [2, 1280, 8, 8]> x_5_cast = conv(bias = up_blocks_0_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_6392, groups = var_6340, pad = x_5_pad_0, pad_type = x_5_pad_type_0, strides = var_6390, weight = up_blocks_0_resnets_0_conv_shortcut_weight_to_fp16_palettized, x = input_239_cast)[name = tensor<string, []>("x_5_cast")];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_145_cast = add(x = x_5_cast, y = hidden_states_143_cast)[name = tensor<string, []>("hidden_states_145_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_6340, interleave = input_253_interleave_0, values = (hidden_states_145_cast, input_185_cast))[name = tensor<string, []>("input_253_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)[name = tensor<string, []>("reshape_116_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)[name = tensor<string, []>("reduce_mean_87_cast")];
            tensor<fp16, [2, 32, 80, 8, 8]> sub_58_cast = sub(x = reshape_116_cast, y = reduce_mean_87_cast)[name = tensor<string, []>("sub_58_cast")];
            tensor<fp16, [2, 32, 80, 8, 8]> square_29_cast = square(x = sub_58_cast)[name = tensor<string, []>("square_29_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)[name = tensor<string, []>("reduce_mean_89_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)[name = tensor<string, []>("add_58_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_29_cast = sqrt(x = add_58_cast)[name = tensor<string, []>("sqrt_29_cast")];
            tensor<fp16, [2, 32, 80, 8, 8]> real_div_29_cast = real_div(x = sub_58_cast, y = sqrt_29_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_117_cast")];
            tensor<fp16, [2560]> add_59_gamma_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(298661824))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(298663808))), name = tensor<string, []>("add_59_gamma_0_to_fp16_palettized"), shape = tensor<uint32, [1]>([2560])];
            tensor<fp16, [2560]> add_59_beta_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(298664000))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(298665984))), name = tensor<string, []>("add_59_beta_0_to_fp16_palettized"), shape = tensor<uint32, [1]>([2560])];
            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_palettized, epsilon = add_59_epsilon_0_to_fp16, gamma = add_59_gamma_0_to_fp16_palettized, mean = add_55_mean_0_to_fp16_palettized, variance = add_55_variance_0_to_fp16_palettized, x = reshape_117_cast)[name = tensor<string, []>("add_59_cast")];
            tensor<fp16, [2, 2560, 8, 8]> input_257_cast = silu(x = add_59_cast)[name = tensor<string, []>("input_257_cast")];
            tensor<int32, [2]> var_6410 = const()[name = tensor<string, []>("op_6410"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6412 = const()[name = tensor<string, []>("op_6412"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [22118400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(298666176))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(320784640))), name = tensor<string, []>("up_blocks_0_resnets_1_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2560, 3, 3])];
            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, []>(320784832)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_147_cast = conv(bias = up_blocks_0_resnets_1_conv1_bias_to_fp16, dilations = var_6412, groups = var_6340, pad = hidden_states_147_pad_0, pad_type = hidden_states_147_pad_type_0, strides = var_6410, weight = up_blocks_0_resnets_1_conv1_weight_to_fp16_palettized, x = input_257_cast)[name = tensor<string, []>("hidden_states_147_cast")];
            tensor<int32, [2]> var_6418 = const()[name = tensor<string, []>("op_6418"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6420 = const()[name = tensor<string, []>("op_6420"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(320787456))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(322016320))), name = tensor<string, []>("up_blocks_0_resnets_1_time_emb_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(322016512)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_23_cast = conv(bias = up_blocks_0_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_6420, groups = var_6340, pad = temb_23_pad_0, pad_type = temb_23_pad_type_0, strides = var_6418, weight = up_blocks_0_resnets_1_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor<string, []>("temb_23_cast")];
            tensor<fp16, [2, 1280, 8, 8]> input_261_cast = add(x = hidden_states_147_cast, y = temb_23_cast)[name = tensor<string, []>("input_261_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)[name = tensor<string, []>("reshape_120_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)[name = tensor<string, []>("reduce_mean_90_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> sub_60_cast = sub(x = reshape_120_cast, y = reduce_mean_90_cast)[name = tensor<string, []>("sub_60_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> square_30_cast = square(x = sub_60_cast)[name = tensor<string, []>("square_30_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)[name = tensor<string, []>("reduce_mean_92_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)[name = tensor<string, []>("add_60_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_30_cast = sqrt(x = add_60_cast)[name = tensor<string, []>("sqrt_30_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> real_div_30_cast = real_div(x = sub_60_cast, y = sqrt_30_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_121_cast")];
            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, []>(322019136)))];
            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, []>(322021760)))];
            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_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_121_cast)[name = tensor<string, []>("add_61_cast")];
            tensor<fp16, [2, 1280, 8, 8]> input_265_cast = silu(x = add_61_cast)[name = tensor<string, []>("input_265_cast")];
            tensor<int32, [2]> var_6430 = const()[name = tensor<string, []>("op_6430"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6432 = const()[name = tensor<string, []>("op_6432"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [11059200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(322024384))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(333083648))), name = tensor<string, []>("up_blocks_0_resnets_1_conv2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 3, 3])];
            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, []>(333083840)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_149_cast = conv(bias = up_blocks_0_resnets_1_conv2_bias_to_fp16, dilations = var_6432, groups = var_6340, pad = hidden_states_149_pad_0, pad_type = hidden_states_149_pad_type_0, strides = var_6430, weight = up_blocks_0_resnets_1_conv2_weight_to_fp16_palettized, x = input_265_cast)[name = tensor<string, []>("hidden_states_149_cast")];
            tensor<int32, [2]> var_6437 = const()[name = tensor<string, []>("op_6437"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6439 = const()[name = tensor<string, []>("op_6439"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2457600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(333086464))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(335544128))), name = tensor<string, []>("up_blocks_0_resnets_1_conv_shortcut_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2560, 1, 1])];
            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, []>(335544320)))];
            tensor<fp16, [2, 1280, 8, 8]> x_7_cast = conv(bias = up_blocks_0_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_6439, groups = var_6340, pad = x_7_pad_0, pad_type = x_7_pad_type_0, strides = var_6437, weight = up_blocks_0_resnets_1_conv_shortcut_weight_to_fp16_palettized, x = input_253_cast)[name = tensor<string, []>("x_7_cast")];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_151_cast = add(x = x_7_cast, y = hidden_states_149_cast)[name = tensor<string, []>("hidden_states_151_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_6340, interleave = input_267_interleave_0, values = (hidden_states_151_cast, input_171_cast))[name = tensor<string, []>("input_267_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)[name = tensor<string, []>("reshape_124_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)[name = tensor<string, []>("reduce_mean_93_cast")];
            tensor<fp16, [2, 32, 80, 8, 8]> sub_62_cast = sub(x = reshape_124_cast, y = reduce_mean_93_cast)[name = tensor<string, []>("sub_62_cast")];
            tensor<fp16, [2, 32, 80, 8, 8]> square_31_cast = square(x = sub_62_cast)[name = tensor<string, []>("square_31_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)[name = tensor<string, []>("reduce_mean_95_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)[name = tensor<string, []>("add_62_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_31_cast = sqrt(x = add_62_cast)[name = tensor<string, []>("sqrt_31_cast")];
            tensor<fp16, [2, 32, 80, 8, 8]> real_div_31_cast = real_div(x = sub_62_cast, y = sqrt_31_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_125_cast")];
            tensor<fp16, [2560]> add_63_gamma_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(335546944))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(335548928))), name = tensor<string, []>("add_63_gamma_0_to_fp16_palettized"), shape = tensor<uint32, [1]>([2560])];
            tensor<fp16, [2560]> add_63_beta_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(335549120))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(335551104))), name = tensor<string, []>("add_63_beta_0_to_fp16_palettized"), shape = tensor<uint32, [1]>([2560])];
            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_palettized, epsilon = add_63_epsilon_0_to_fp16, gamma = add_63_gamma_0_to_fp16_palettized, mean = add_55_mean_0_to_fp16_palettized, variance = add_55_variance_0_to_fp16_palettized, x = reshape_125_cast)[name = tensor<string, []>("add_63_cast")];
            tensor<fp16, [2, 2560, 8, 8]> input_271_cast = silu(x = add_63_cast)[name = tensor<string, []>("input_271_cast")];
            tensor<int32, [2]> var_6457 = const()[name = tensor<string, []>("op_6457"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6459 = const()[name = tensor<string, []>("op_6459"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [22118400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(335551296))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(357669760))), name = tensor<string, []>("up_blocks_0_resnets_2_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2560, 3, 3])];
            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, []>(357669952)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_153_cast = conv(bias = up_blocks_0_resnets_2_conv1_bias_to_fp16, dilations = var_6459, groups = var_6340, pad = hidden_states_153_pad_0, pad_type = hidden_states_153_pad_type_0, strides = var_6457, weight = up_blocks_0_resnets_2_conv1_weight_to_fp16_palettized, x = input_271_cast)[name = tensor<string, []>("hidden_states_153_cast")];
            tensor<int32, [2]> var_6465 = const()[name = tensor<string, []>("op_6465"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6467 = const()[name = tensor<string, []>("op_6467"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(357672576))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(358901440))), name = tensor<string, []>("up_blocks_0_resnets_2_time_emb_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(358901632)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_25_cast = conv(bias = up_blocks_0_resnets_2_time_emb_proj_bias_to_fp16, dilations = var_6467, groups = var_6340, pad = temb_25_pad_0, pad_type = temb_25_pad_type_0, strides = var_6465, weight = up_blocks_0_resnets_2_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor<string, []>("temb_25_cast")];
            tensor<fp16, [2, 1280, 8, 8]> input_275_cast = add(x = hidden_states_153_cast, y = temb_25_cast)[name = tensor<string, []>("input_275_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)[name = tensor<string, []>("reshape_128_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)[name = tensor<string, []>("reduce_mean_96_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> sub_64_cast = sub(x = reshape_128_cast, y = reduce_mean_96_cast)[name = tensor<string, []>("sub_64_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> square_32_cast = square(x = sub_64_cast)[name = tensor<string, []>("square_32_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)[name = tensor<string, []>("reduce_mean_98_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)[name = tensor<string, []>("add_64_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_32_cast = sqrt(x = add_64_cast)[name = tensor<string, []>("sqrt_32_cast")];
            tensor<fp16, [2, 32, 40, 8, 8]> real_div_32_cast = real_div(x = sub_64_cast, y = sqrt_32_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_129_cast")];
            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, []>(358904256)))];
            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, []>(358906880)))];
            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_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_129_cast)[name = tensor<string, []>("add_65_cast")];
            tensor<fp16, [2, 1280, 8, 8]> input_279_cast = silu(x = add_65_cast)[name = tensor<string, []>("input_279_cast")];
            tensor<int32, [2]> var_6477 = const()[name = tensor<string, []>("op_6477"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6479 = const()[name = tensor<string, []>("op_6479"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [11059200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(358909504))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(369968768))), name = tensor<string, []>("up_blocks_0_resnets_2_conv2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 3, 3])];
            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, []>(369968960)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_155_cast = conv(bias = up_blocks_0_resnets_2_conv2_bias_to_fp16, dilations = var_6479, groups = var_6340, pad = hidden_states_155_pad_0, pad_type = hidden_states_155_pad_type_0, strides = var_6477, weight = up_blocks_0_resnets_2_conv2_weight_to_fp16_palettized, x = input_279_cast)[name = tensor<string, []>("hidden_states_155_cast")];
            tensor<int32, [2]> var_6484 = const()[name = tensor<string, []>("op_6484"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6486 = const()[name = tensor<string, []>("op_6486"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2457600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(369971584))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(372429248))), name = tensor<string, []>("up_blocks_0_resnets_2_conv_shortcut_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2560, 1, 1])];
            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, []>(372429440)))];
            tensor<fp16, [2, 1280, 8, 8]> x_9_cast = conv(bias = up_blocks_0_resnets_2_conv_shortcut_bias_to_fp16, dilations = var_6486, groups = var_6340, pad = x_9_pad_0, pad_type = x_9_pad_type_0, strides = var_6484, weight = up_blocks_0_resnets_2_conv_shortcut_weight_to_fp16_palettized, x = input_267_cast)[name = tensor<string, []>("x_9_cast")];
            tensor<fp16, [2, 1280, 8, 8]> input_281_cast = add(x = x_9_cast, y = hidden_states_155_cast)[name = tensor<string, []>("input_281_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)[name = tensor<string, []>("input_283_cast")];
            tensor<int32, [2]> var_6495 = const()[name = tensor<string, []>("op_6495"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6497 = const()[name = tensor<string, []>("op_6497"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [11059200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(372432064))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(383491328))), name = tensor<string, []>("up_blocks_0_upsamplers_0_conv_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 3, 3])];
            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, []>(383491520)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_157_cast = conv(bias = up_blocks_0_upsamplers_0_conv_bias_to_fp16, dilations = var_6497, groups = var_6340, pad = hidden_states_157_pad_0, pad_type = hidden_states_157_pad_type_0, strides = var_6495, weight = up_blocks_0_upsamplers_0_conv_weight_to_fp16_palettized, x = input_283_cast)[name = tensor<string, []>("hidden_states_157_cast")];
            tensor<bool, []> var_6517 = const()[name = tensor<string, []>("op_6517"), val = tensor<bool, []>(true)];
            tensor<int32, []> var_6522 = const()[name = tensor<string, []>("op_6522"), 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_6522, interleave = input_285_interleave_0, values = (hidden_states_157_cast, input_169_cast))[name = tensor<string, []>("input_285_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)[name = tensor<string, []>("reshape_132_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)[name = tensor<string, []>("reduce_mean_99_cast")];
            tensor<fp16, [2, 32, 80, 16, 16]> sub_66_cast = sub(x = reshape_132_cast, y = reduce_mean_99_cast)[name = tensor<string, []>("sub_66_cast")];
            tensor<fp16, [2, 32, 80, 16, 16]> square_33_cast = square(x = sub_66_cast)[name = tensor<string, []>("square_33_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)[name = tensor<string, []>("reduce_mean_101_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)[name = tensor<string, []>("add_66_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_33_cast = sqrt(x = add_66_cast)[name = tensor<string, []>("sqrt_33_cast")];
            tensor<fp16, [2, 32, 80, 16, 16]> real_div_33_cast = real_div(x = sub_66_cast, y = sqrt_33_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_133_cast")];
            tensor<fp16, [2560]> add_67_gamma_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(383494144))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(383496128))), name = tensor<string, []>("add_67_gamma_0_to_fp16_palettized"), shape = tensor<uint32, [1]>([2560])];
            tensor<fp16, [2560]> add_67_beta_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(383496320))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(383498304))), name = tensor<string, []>("add_67_beta_0_to_fp16_palettized"), shape = tensor<uint32, [1]>([2560])];
            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_palettized, epsilon = add_67_epsilon_0_to_fp16, gamma = add_67_gamma_0_to_fp16_palettized, mean = add_55_mean_0_to_fp16_palettized, variance = add_55_variance_0_to_fp16_palettized, x = reshape_133_cast)[name = tensor<string, []>("add_67_cast")];
            tensor<fp16, [2, 2560, 16, 16]> input_289_cast = silu(x = add_67_cast)[name = tensor<string, []>("input_289_cast")];
            tensor<int32, [2]> var_6551 = const()[name = tensor<string, []>("op_6551"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6553 = const()[name = tensor<string, []>("op_6553"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [22118400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(383498496))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(405616960))), name = tensor<string, []>("up_blocks_1_resnets_0_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2560, 3, 3])];
            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, []>(405617152)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_159_cast = conv(bias = up_blocks_1_resnets_0_conv1_bias_to_fp16, dilations = var_6553, groups = var_6522, pad = hidden_states_159_pad_0, pad_type = hidden_states_159_pad_type_0, strides = var_6551, weight = up_blocks_1_resnets_0_conv1_weight_to_fp16_palettized, x = input_289_cast)[name = tensor<string, []>("hidden_states_159_cast")];
            tensor<int32, [2]> var_6559 = const()[name = tensor<string, []>("op_6559"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6561 = const()[name = tensor<string, []>("op_6561"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(405619776))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(406848640))), name = tensor<string, []>("up_blocks_1_resnets_0_time_emb_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(406848832)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_27_cast = conv(bias = up_blocks_1_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_6561, groups = var_6522, pad = temb_27_pad_0, pad_type = temb_27_pad_type_0, strides = var_6559, weight = up_blocks_1_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor<string, []>("temb_27_cast")];
            tensor<fp16, [2, 1280, 16, 16]> input_293_cast = add(x = hidden_states_159_cast, y = temb_27_cast)[name = tensor<string, []>("input_293_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)[name = tensor<string, []>("reshape_136_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)[name = tensor<string, []>("reduce_mean_102_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> sub_68_cast = sub(x = reshape_136_cast, y = reduce_mean_102_cast)[name = tensor<string, []>("sub_68_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> square_34_cast = square(x = sub_68_cast)[name = tensor<string, []>("square_34_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)[name = tensor<string, []>("reduce_mean_104_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)[name = tensor<string, []>("add_68_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_34_cast = sqrt(x = add_68_cast)[name = tensor<string, []>("sqrt_34_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> real_div_34_cast = real_div(x = sub_68_cast, y = sqrt_34_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_137_cast")];
            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, []>(406851456)))];
            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, []>(406854080)))];
            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_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_137_cast)[name = tensor<string, []>("add_69_cast")];
            tensor<fp16, [2, 1280, 16, 16]> input_297_cast = silu(x = add_69_cast)[name = tensor<string, []>("input_297_cast")];
            tensor<int32, [2]> var_6571 = const()[name = tensor<string, []>("op_6571"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6573 = const()[name = tensor<string, []>("op_6573"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [11059200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(406856704))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(417915968))), name = tensor<string, []>("up_blocks_1_resnets_0_conv2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 3, 3])];
            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, []>(417916160)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_161_cast = conv(bias = up_blocks_1_resnets_0_conv2_bias_to_fp16, dilations = var_6573, groups = var_6522, pad = hidden_states_161_pad_0, pad_type = hidden_states_161_pad_type_0, strides = var_6571, weight = up_blocks_1_resnets_0_conv2_weight_to_fp16_palettized, x = input_297_cast)[name = tensor<string, []>("hidden_states_161_cast")];
            tensor<int32, [2]> var_6578 = const()[name = tensor<string, []>("op_6578"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6580 = const()[name = tensor<string, []>("op_6580"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2457600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(417918784))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(420376448))), name = tensor<string, []>("up_blocks_1_resnets_0_conv_shortcut_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2560, 1, 1])];
            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, []>(420376640)))];
            tensor<fp16, [2, 1280, 16, 16]> x_11_cast = conv(bias = up_blocks_1_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_6580, groups = var_6522, pad = x_11_pad_0, pad_type = x_11_pad_type_0, strides = var_6578, weight = up_blocks_1_resnets_0_conv_shortcut_weight_to_fp16_palettized, x = input_285_cast)[name = tensor<string, []>("x_11_cast")];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_163_cast = add(x = x_11_cast, y = hidden_states_161_cast)[name = tensor<string, []>("hidden_states_163_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)[name = tensor<string, []>("reshape_140_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)[name = tensor<string, []>("reduce_mean_105_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> sub_70_cast = sub(x = reshape_140_cast, y = reduce_mean_105_cast)[name = tensor<string, []>("sub_70_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> square_35_cast = square(x = sub_70_cast)[name = tensor<string, []>("square_35_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)[name = tensor<string, []>("reduce_mean_107_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)[name = tensor<string, []>("add_70_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_35_cast = sqrt(x = add_70_cast)[name = tensor<string, []>("sqrt_35_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> real_div_35_cast = real_div(x = sub_70_cast, y = sqrt_35_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_141_cast")];
            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, []>(420379264)))];
            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, []>(420381888)))];
            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_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_141_cast)[name = tensor<string, []>("add_71_cast")];
            tensor<int32, [2]> var_6600 = const()[name = tensor<string, []>("op_6600"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6602 = const()[name = tensor<string, []>("op_6602"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(420384512))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(421613376))), name = tensor<string, []>("up_blocks_1_attentions_0_proj_in_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(421613568)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_165_cast = conv(bias = up_blocks_1_attentions_0_proj_in_bias_to_fp16, dilations = var_6602, groups = var_6522, pad = hidden_states_165_pad_0, pad_type = hidden_states_165_pad_type_0, strides = var_6600, weight = up_blocks_1_attentions_0_proj_in_weight_to_fp16_palettized, x = add_71_cast)[name = tensor<string, []>("hidden_states_165_cast")];
            tensor<int32, [4]> var_6607 = const()[name = tensor<string, []>("op_6607"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<fp16, [2, 1280, 1, 256]> inputs_43_cast = reshape(shape = var_6607, x = hidden_states_165_cast)[name = tensor<string, []>("inputs_43_cast")];
            tensor<int32, [1]> var_6617 = const()[name = tensor<string, []>("op_6617"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_43_cast = reduce_mean(axes = var_6617, keep_dims = var_6517, x = inputs_43_cast)[name = tensor<string, []>("channels_mean_43_cast")];
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_43_cast = sub(x = inputs_43_cast, y = channels_mean_43_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_43_cast")];
            tensor<int32, [1]> var_6621 = const()[name = tensor<string, []>("op_6621"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_6622_cast = reduce_mean(axes = var_6621, keep_dims = var_6517, x = zero_mean_sq_43_cast)[name = tensor<string, []>("op_6622_cast")];
            tensor<fp16, []> var_6623_to_fp16 = const()[name = tensor<string, []>("op_6623_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_6624_cast = add(x = var_6622_cast, y = var_6623_to_fp16)[name = tensor<string, []>("op_6624_cast")];
            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_6624_cast)[name = tensor<string, []>("denom_43_cast")];
            tensor<fp16, [2, 1280, 1, 256]> out_43_cast = mul(x = zero_mean_43_cast, y = denom_43_cast)[name = tensor<string, []>("out_43_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_6628_to_fp16 = const()[name = tensor<string, []>("op_6628_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(421616192)))];
            tensor<fp16, [2, 1280, 1, 256]> var_6629_cast = add(x = out_43_cast, y = var_6628_to_fp16)[name = tensor<string, []>("op_6629_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_6631_to_fp16 = const()[name = tensor<string, []>("op_6631_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(421618816)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_167_cast = mul(x = var_6629_cast, y = var_6631_to_fp16)[name = tensor<string, []>("hidden_states_167_cast")];
            tensor<int32, [2]> var_6638 = const()[name = tensor<string, []>("op_6638"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6640 = const()[name = tensor<string, []>("op_6640"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(421621440))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(422850304))), name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 256]> q_29_cast = conv(dilations = var_6640, groups = var_6522, pad = q_29_pad_0, pad_type = q_29_pad_type_0, strides = var_6638, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_167_cast)[name = tensor<string, []>("q_29_cast")];
            tensor<int32, [2]> var_6644 = const()[name = tensor<string, []>("op_6644"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6646 = const()[name = tensor<string, []>("op_6646"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(422850496))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(424079360))), name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 256]> k_57_cast = conv(dilations = var_6646, groups = var_6522, pad = k_57_pad_0, pad_type = k_57_pad_type_0, strides = var_6644, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_167_cast)[name = tensor<string, []>("k_57_cast")];
            tensor<int32, [2]> var_6650 = const()[name = tensor<string, []>("op_6650"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6652 = const()[name = tensor<string, []>("op_6652"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(424079552))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(425308416))), name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 256]> v_29_cast = conv(dilations = var_6652, groups = var_6522, pad = v_29_pad_0, pad_type = v_29_pad_type_0, strides = var_6650, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_167_cast)[name = tensor<string, []>("v_29_cast")];
            tensor<int32, [4]> var_6656_begin_0 = const()[name = tensor<string, []>("op_6656_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_6656_end_0 = const()[name = tensor<string, []>("op_6656_end_0"), val = tensor<int32, [4]>([2, 160, 1, 256])];
            tensor<bool, [4]> var_6656_end_mask_0 = const()[name = tensor<string, []>("op_6656_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_6656_cast = slice_by_index(begin = var_6656_begin_0, end = var_6656_end_0, end_mask = var_6656_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_6656_cast")];
            tensor<int32, [4]> var_6660_begin_0 = const()[name = tensor<string, []>("op_6660_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_6660_end_0 = const()[name = tensor<string, []>("op_6660_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_6660_end_mask_0 = const()[name = tensor<string, []>("op_6660_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_6660_cast = slice_by_index(begin = var_6660_begin_0, end = var_6660_end_0, end_mask = var_6660_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_6660_cast")];
            tensor<int32, [4]> var_6664_begin_0 = const()[name = tensor<string, []>("op_6664_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_6664_end_0 = const()[name = tensor<string, []>("op_6664_end_0"), val = tensor<int32, [4]>([2, 480, 1, 256])];
            tensor<bool, [4]> var_6664_end_mask_0 = const()[name = tensor<string, []>("op_6664_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_6664_cast = slice_by_index(begin = var_6664_begin_0, end = var_6664_end_0, end_mask = var_6664_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_6664_cast")];
            tensor<int32, [4]> var_6668_begin_0 = const()[name = tensor<string, []>("op_6668_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_6668_end_0 = const()[name = tensor<string, []>("op_6668_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_6668_end_mask_0 = const()[name = tensor<string, []>("op_6668_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_6668_cast = slice_by_index(begin = var_6668_begin_0, end = var_6668_end_0, end_mask = var_6668_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_6668_cast")];
            tensor<int32, [4]> var_6672_begin_0 = const()[name = tensor<string, []>("op_6672_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_6672_end_0 = const()[name = tensor<string, []>("op_6672_end_0"), val = tensor<int32, [4]>([2, 800, 1, 256])];
            tensor<bool, [4]> var_6672_end_mask_0 = const()[name = tensor<string, []>("op_6672_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_6672_cast = slice_by_index(begin = var_6672_begin_0, end = var_6672_end_0, end_mask = var_6672_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_6672_cast")];
            tensor<int32, [4]> var_6676_begin_0 = const()[name = tensor<string, []>("op_6676_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_6676_end_0 = const()[name = tensor<string, []>("op_6676_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_6676_end_mask_0 = const()[name = tensor<string, []>("op_6676_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_6676_cast = slice_by_index(begin = var_6676_begin_0, end = var_6676_end_0, end_mask = var_6676_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_6676_cast")];
            tensor<int32, [4]> var_6680_begin_0 = const()[name = tensor<string, []>("op_6680_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_6680_end_0 = const()[name = tensor<string, []>("op_6680_end_0"), val = tensor<int32, [4]>([2, 1120, 1, 256])];
            tensor<bool, [4]> var_6680_end_mask_0 = const()[name = tensor<string, []>("op_6680_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_6680_cast = slice_by_index(begin = var_6680_begin_0, end = var_6680_end_0, end_mask = var_6680_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_6680_cast")];
            tensor<int32, [4]> var_6684_begin_0 = const()[name = tensor<string, []>("op_6684_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_6684_end_0 = const()[name = tensor<string, []>("op_6684_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_6684_end_mask_0 = const()[name = tensor<string, []>("op_6684_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_6684_cast = slice_by_index(begin = var_6684_begin_0, end = var_6684_end_0, end_mask = var_6684_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_6684_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_6691_begin_0 = const()[name = tensor<string, []>("op_6691_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_6691_end_0 = const()[name = tensor<string, []>("op_6691_end_0"), val = tensor<int32, [4]>([2, 256, 1, 160])];
            tensor<bool, [4]> var_6691_end_mask_0 = const()[name = tensor<string, []>("op_6691_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)[name = tensor<string, []>("transpose_17")];
            tensor<fp16, [2, 256, 1, 160]> var_6691_cast = slice_by_index(begin = var_6691_begin_0, end = var_6691_end_0, end_mask = var_6691_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_6691_cast")];
            tensor<int32, [4]> var_6695_begin_0 = const()[name = tensor<string, []>("op_6695_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_6695_end_0 = const()[name = tensor<string, []>("op_6695_end_0"), val = tensor<int32, [4]>([2, 256, 1, 320])];
            tensor<bool, [4]> var_6695_end_mask_0 = const()[name = tensor<string, []>("op_6695_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_6695_cast = slice_by_index(begin = var_6695_begin_0, end = var_6695_end_0, end_mask = var_6695_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_6695_cast")];
            tensor<int32, [4]> var_6699_begin_0 = const()[name = tensor<string, []>("op_6699_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_6699_end_0 = const()[name = tensor<string, []>("op_6699_end_0"), val = tensor<int32, [4]>([2, 256, 1, 480])];
            tensor<bool, [4]> var_6699_end_mask_0 = const()[name = tensor<string, []>("op_6699_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_6699_cast = slice_by_index(begin = var_6699_begin_0, end = var_6699_end_0, end_mask = var_6699_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_6699_cast")];
            tensor<int32, [4]> var_6703_begin_0 = const()[name = tensor<string, []>("op_6703_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 480])];
            tensor<int32, [4]> var_6703_end_0 = const()[name = tensor<string, []>("op_6703_end_0"), val = tensor<int32, [4]>([2, 256, 1, 640])];
            tensor<bool, [4]> var_6703_end_mask_0 = const()[name = tensor<string, []>("op_6703_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_6703_cast = slice_by_index(begin = var_6703_begin_0, end = var_6703_end_0, end_mask = var_6703_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_6703_cast")];
            tensor<int32, [4]> var_6707_begin_0 = const()[name = tensor<string, []>("op_6707_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_6707_end_0 = const()[name = tensor<string, []>("op_6707_end_0"), val = tensor<int32, [4]>([2, 256, 1, 800])];
            tensor<bool, [4]> var_6707_end_mask_0 = const()[name = tensor<string, []>("op_6707_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_6707_cast = slice_by_index(begin = var_6707_begin_0, end = var_6707_end_0, end_mask = var_6707_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_6707_cast")];
            tensor<int32, [4]> var_6711_begin_0 = const()[name = tensor<string, []>("op_6711_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 800])];
            tensor<int32, [4]> var_6711_end_0 = const()[name = tensor<string, []>("op_6711_end_0"), val = tensor<int32, [4]>([2, 256, 1, 960])];
            tensor<bool, [4]> var_6711_end_mask_0 = const()[name = tensor<string, []>("op_6711_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_6711_cast = slice_by_index(begin = var_6711_begin_0, end = var_6711_end_0, end_mask = var_6711_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_6711_cast")];
            tensor<int32, [4]> var_6715_begin_0 = const()[name = tensor<string, []>("op_6715_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_6715_end_0 = const()[name = tensor<string, []>("op_6715_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1120])];
            tensor<bool, [4]> var_6715_end_mask_0 = const()[name = tensor<string, []>("op_6715_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_6715_cast = slice_by_index(begin = var_6715_begin_0, end = var_6715_end_0, end_mask = var_6715_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_6715_cast")];
            tensor<int32, [4]> var_6719_begin_0 = const()[name = tensor<string, []>("op_6719_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1120])];
            tensor<int32, [4]> var_6719_end_0 = const()[name = tensor<string, []>("op_6719_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1280])];
            tensor<bool, [4]> var_6719_end_mask_0 = const()[name = tensor<string, []>("op_6719_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_6719_cast = slice_by_index(begin = var_6719_begin_0, end = var_6719_end_0, end_mask = var_6719_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_6719_cast")];
            tensor<int32, [4]> var_6721_begin_0 = const()[name = tensor<string, []>("op_6721_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_6721_end_0 = const()[name = tensor<string, []>("op_6721_end_0"), val = tensor<int32, [4]>([2, 160, 1, 256])];
            tensor<bool, [4]> var_6721_end_mask_0 = const()[name = tensor<string, []>("op_6721_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_6721_cast = slice_by_index(begin = var_6721_begin_0, end = var_6721_end_0, end_mask = var_6721_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_6721_cast")];
            tensor<int32, [4]> var_6725_begin_0 = const()[name = tensor<string, []>("op_6725_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_6725_end_0 = const()[name = tensor<string, []>("op_6725_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_6725_end_mask_0 = const()[name = tensor<string, []>("op_6725_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_6725_cast = slice_by_index(begin = var_6725_begin_0, end = var_6725_end_0, end_mask = var_6725_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_6725_cast")];
            tensor<int32, [4]> var_6729_begin_0 = const()[name = tensor<string, []>("op_6729_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_6729_end_0 = const()[name = tensor<string, []>("op_6729_end_0"), val = tensor<int32, [4]>([2, 480, 1, 256])];
            tensor<bool, [4]> var_6729_end_mask_0 = const()[name = tensor<string, []>("op_6729_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_6729_cast = slice_by_index(begin = var_6729_begin_0, end = var_6729_end_0, end_mask = var_6729_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_6729_cast")];
            tensor<int32, [4]> var_6733_begin_0 = const()[name = tensor<string, []>("op_6733_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_6733_end_0 = const()[name = tensor<string, []>("op_6733_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_6733_end_mask_0 = const()[name = tensor<string, []>("op_6733_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_6733_cast = slice_by_index(begin = var_6733_begin_0, end = var_6733_end_0, end_mask = var_6733_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_6733_cast")];
            tensor<int32, [4]> var_6737_begin_0 = const()[name = tensor<string, []>("op_6737_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_6737_end_0 = const()[name = tensor<string, []>("op_6737_end_0"), val = tensor<int32, [4]>([2, 800, 1, 256])];
            tensor<bool, [4]> var_6737_end_mask_0 = const()[name = tensor<string, []>("op_6737_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_6737_cast = slice_by_index(begin = var_6737_begin_0, end = var_6737_end_0, end_mask = var_6737_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_6737_cast")];
            tensor<int32, [4]> var_6741_begin_0 = const()[name = tensor<string, []>("op_6741_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_6741_end_0 = const()[name = tensor<string, []>("op_6741_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_6741_end_mask_0 = const()[name = tensor<string, []>("op_6741_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_6741_cast = slice_by_index(begin = var_6741_begin_0, end = var_6741_end_0, end_mask = var_6741_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_6741_cast")];
            tensor<int32, [4]> var_6745_begin_0 = const()[name = tensor<string, []>("op_6745_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_6745_end_0 = const()[name = tensor<string, []>("op_6745_end_0"), val = tensor<int32, [4]>([2, 1120, 1, 256])];
            tensor<bool, [4]> var_6745_end_mask_0 = const()[name = tensor<string, []>("op_6745_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_6745_cast = slice_by_index(begin = var_6745_begin_0, end = var_6745_end_0, end_mask = var_6745_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_6745_cast")];
            tensor<int32, [4]> var_6749_begin_0 = const()[name = tensor<string, []>("op_6749_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_6749_end_0 = const()[name = tensor<string, []>("op_6749_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_6749_end_mask_0 = const()[name = tensor<string, []>("op_6749_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_6749_cast = slice_by_index(begin = var_6749_begin_0, end = var_6749_end_0, end_mask = var_6749_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_6749_cast")];
            tensor<string, []> var_6753_equation_0 = const()[name = tensor<string, []>("op_6753_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6753_cast = einsum(equation = var_6753_equation_0, values = (var_6691_cast, var_6656_cast))[name = tensor<string, []>("op_6753_cast")];
            tensor<fp16, []> var_6754_to_fp16 = const()[name = tensor<string, []>("op_6754_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_97_cast = mul(x = var_6753_cast, y = var_6754_to_fp16)[name = tensor<string, []>("aw_97_cast")];
            tensor<string, []> var_6757_equation_0 = const()[name = tensor<string, []>("op_6757_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6757_cast = einsum(equation = var_6757_equation_0, values = (var_6695_cast, var_6660_cast))[name = tensor<string, []>("op_6757_cast")];
            tensor<fp16, []> var_6758_to_fp16 = const()[name = tensor<string, []>("op_6758_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_99_cast = mul(x = var_6757_cast, y = var_6758_to_fp16)[name = tensor<string, []>("aw_99_cast")];
            tensor<string, []> var_6761_equation_0 = const()[name = tensor<string, []>("op_6761_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6761_cast = einsum(equation = var_6761_equation_0, values = (var_6699_cast, var_6664_cast))[name = tensor<string, []>("op_6761_cast")];
            tensor<fp16, []> var_6762_to_fp16 = const()[name = tensor<string, []>("op_6762_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_101_cast = mul(x = var_6761_cast, y = var_6762_to_fp16)[name = tensor<string, []>("aw_101_cast")];
            tensor<string, []> var_6765_equation_0 = const()[name = tensor<string, []>("op_6765_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6765_cast = einsum(equation = var_6765_equation_0, values = (var_6703_cast, var_6668_cast))[name = tensor<string, []>("op_6765_cast")];
            tensor<fp16, []> var_6766_to_fp16 = const()[name = tensor<string, []>("op_6766_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_103_cast = mul(x = var_6765_cast, y = var_6766_to_fp16)[name = tensor<string, []>("aw_103_cast")];
            tensor<string, []> var_6769_equation_0 = const()[name = tensor<string, []>("op_6769_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6769_cast = einsum(equation = var_6769_equation_0, values = (var_6707_cast, var_6672_cast))[name = tensor<string, []>("op_6769_cast")];
            tensor<fp16, []> var_6770_to_fp16 = const()[name = tensor<string, []>("op_6770_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_105_cast = mul(x = var_6769_cast, y = var_6770_to_fp16)[name = tensor<string, []>("aw_105_cast")];
            tensor<string, []> var_6773_equation_0 = const()[name = tensor<string, []>("op_6773_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6773_cast = einsum(equation = var_6773_equation_0, values = (var_6711_cast, var_6676_cast))[name = tensor<string, []>("op_6773_cast")];
            tensor<fp16, []> var_6774_to_fp16 = const()[name = tensor<string, []>("op_6774_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_107_cast = mul(x = var_6773_cast, y = var_6774_to_fp16)[name = tensor<string, []>("aw_107_cast")];
            tensor<string, []> var_6777_equation_0 = const()[name = tensor<string, []>("op_6777_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6777_cast = einsum(equation = var_6777_equation_0, values = (var_6715_cast, var_6680_cast))[name = tensor<string, []>("op_6777_cast")];
            tensor<fp16, []> var_6778_to_fp16 = const()[name = tensor<string, []>("op_6778_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_109_cast = mul(x = var_6777_cast, y = var_6778_to_fp16)[name = tensor<string, []>("aw_109_cast")];
            tensor<string, []> var_6781_equation_0 = const()[name = tensor<string, []>("op_6781_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_6781_cast = einsum(equation = var_6781_equation_0, values = (var_6719_cast, var_6684_cast))[name = tensor<string, []>("op_6781_cast")];
            tensor<fp16, []> var_6782_to_fp16 = const()[name = tensor<string, []>("op_6782_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_111_cast = mul(x = var_6781_cast, y = var_6782_to_fp16)[name = tensor<string, []>("aw_111_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_6784_cast = softmax(axis = var_6522, x = aw_97_cast)[name = tensor<string, []>("op_6784_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_6785_cast = softmax(axis = var_6522, x = aw_99_cast)[name = tensor<string, []>("op_6785_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_6786_cast = softmax(axis = var_6522, x = aw_101_cast)[name = tensor<string, []>("op_6786_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_6787_cast = softmax(axis = var_6522, x = aw_103_cast)[name = tensor<string, []>("op_6787_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_6788_cast = softmax(axis = var_6522, x = aw_105_cast)[name = tensor<string, []>("op_6788_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_6789_cast = softmax(axis = var_6522, x = aw_107_cast)[name = tensor<string, []>("op_6789_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_6790_cast = softmax(axis = var_6522, x = aw_109_cast)[name = tensor<string, []>("op_6790_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_6791_cast = softmax(axis = var_6522, x = aw_111_cast)[name = tensor<string, []>("op_6791_cast")];
            tensor<string, []> var_6793_equation_0 = const()[name = tensor<string, []>("op_6793_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_6793_cast = einsum(equation = var_6793_equation_0, values = (var_6721_cast, var_6784_cast))[name = tensor<string, []>("op_6793_cast")];
            tensor<string, []> var_6795_equation_0 = const()[name = tensor<string, []>("op_6795_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_6795_cast = einsum(equation = var_6795_equation_0, values = (var_6725_cast, var_6785_cast))[name = tensor<string, []>("op_6795_cast")];
            tensor<string, []> var_6797_equation_0 = const()[name = tensor<string, []>("op_6797_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_6797_cast = einsum(equation = var_6797_equation_0, values = (var_6729_cast, var_6786_cast))[name = tensor<string, []>("op_6797_cast")];
            tensor<string, []> var_6799_equation_0 = const()[name = tensor<string, []>("op_6799_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_6799_cast = einsum(equation = var_6799_equation_0, values = (var_6733_cast, var_6787_cast))[name = tensor<string, []>("op_6799_cast")];
            tensor<string, []> var_6801_equation_0 = const()[name = tensor<string, []>("op_6801_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_6801_cast = einsum(equation = var_6801_equation_0, values = (var_6737_cast, var_6788_cast))[name = tensor<string, []>("op_6801_cast")];
            tensor<string, []> var_6803_equation_0 = const()[name = tensor<string, []>("op_6803_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_6803_cast = einsum(equation = var_6803_equation_0, values = (var_6741_cast, var_6789_cast))[name = tensor<string, []>("op_6803_cast")];
            tensor<string, []> var_6805_equation_0 = const()[name = tensor<string, []>("op_6805_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_6805_cast = einsum(equation = var_6805_equation_0, values = (var_6745_cast, var_6790_cast))[name = tensor<string, []>("op_6805_cast")];
            tensor<string, []> var_6807_equation_0 = const()[name = tensor<string, []>("op_6807_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_6807_cast = einsum(equation = var_6807_equation_0, values = (var_6749_cast, var_6791_cast))[name = tensor<string, []>("op_6807_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_6522, interleave = input_301_interleave_0, values = (var_6793_cast, var_6795_cast, var_6797_cast, var_6799_cast, var_6801_cast, var_6803_cast, var_6805_cast, var_6807_cast))[name = tensor<string, []>("input_301_cast")];
            tensor<int32, [2]> var_6813 = const()[name = tensor<string, []>("op_6813"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6815 = const()[name = tensor<string, []>("op_6815"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_6817_pad_type_0 = const()[name = tensor<string, []>("op_6817_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_6817_pad_0 = const()[name = tensor<string, []>("op_6817_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(425308608))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(426537472))), name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(426537664)))];
            tensor<fp16, [2, 1280, 1, 256]> var_6817_cast = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_6815, groups = var_6522, pad = var_6817_pad_0, pad_type = var_6817_pad_type_0, strides = var_6813, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_301_cast)[name = tensor<string, []>("op_6817_cast")];
            tensor<fp16, [2, 1280, 1, 256]> inputs_45_cast = add(x = var_6817_cast, y = inputs_43_cast)[name = tensor<string, []>("inputs_45_cast")];
            tensor<int32, [1]> var_6821 = const()[name = tensor<string, []>("op_6821"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_45_cast = reduce_mean(axes = var_6821, keep_dims = var_6517, x = inputs_45_cast)[name = tensor<string, []>("channels_mean_45_cast")];
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_45_cast = sub(x = inputs_45_cast, y = channels_mean_45_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_45_cast")];
            tensor<int32, [1]> var_6825 = const()[name = tensor<string, []>("op_6825"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_6826_cast = reduce_mean(axes = var_6825, keep_dims = var_6517, x = zero_mean_sq_45_cast)[name = tensor<string, []>("op_6826_cast")];
            tensor<fp16, []> var_6827_to_fp16 = const()[name = tensor<string, []>("op_6827_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_6828_cast = add(x = var_6826_cast, y = var_6827_to_fp16)[name = tensor<string, []>("op_6828_cast")];
            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_6828_cast)[name = tensor<string, []>("denom_45_cast")];
            tensor<fp16, [2, 1280, 1, 256]> out_45_cast = mul(x = zero_mean_45_cast, y = denom_45_cast)[name = tensor<string, []>("out_45_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_6832_to_fp16 = const()[name = tensor<string, []>("op_6832_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(426540288)))];
            tensor<fp16, [2, 1280, 1, 256]> var_6833_cast = add(x = out_45_cast, y = var_6832_to_fp16)[name = tensor<string, []>("op_6833_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_6835_to_fp16 = const()[name = tensor<string, []>("op_6835_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(426542912)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_169_cast = mul(x = var_6833_cast, y = var_6835_to_fp16)[name = tensor<string, []>("hidden_states_169_cast")];
            tensor<int32, [2]> var_6842 = const()[name = tensor<string, []>("op_6842"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6844 = const()[name = tensor<string, []>("op_6844"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(426545536))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(427774400))), name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 256]> q_31_cast = conv(dilations = var_6844, groups = var_6522, pad = q_31_pad_0, pad_type = q_31_pad_type_0, strides = var_6842, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_169_cast)[name = tensor<string, []>("q_31_cast")];
            tensor<int32, [2]> var_6848 = const()[name = tensor<string, []>("op_6848"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6850 = const()[name = tensor<string, []>("op_6850"), 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, 768, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [737280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(427774592))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(428511936))), name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 768, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_61_cast = conv(dilations = var_6850, groups = var_6522, pad = k_61_pad_0, pad_type = k_61_pad_type_0, strides = var_6848, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_61_cast")];
            tensor<int32, [2]> var_6854 = const()[name = tensor<string, []>("op_6854"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6856 = const()[name = tensor<string, []>("op_6856"), 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, 768, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [737280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(428512128))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(429249472))), name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 768, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_31_cast = conv(dilations = var_6856, groups = var_6522, pad = v_31_pad_0, pad_type = v_31_pad_type_0, strides = var_6854, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_31_cast")];
            tensor<int32, [4]> var_6860_begin_0 = const()[name = tensor<string, []>("op_6860_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_6860_end_0 = const()[name = tensor<string, []>("op_6860_end_0"), val = tensor<int32, [4]>([2, 160, 1, 256])];
            tensor<bool, [4]> var_6860_end_mask_0 = const()[name = tensor<string, []>("op_6860_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_6860_cast = slice_by_index(begin = var_6860_begin_0, end = var_6860_end_0, end_mask = var_6860_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_6860_cast")];
            tensor<int32, [4]> var_6864_begin_0 = const()[name = tensor<string, []>("op_6864_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_6864_end_0 = const()[name = tensor<string, []>("op_6864_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_6864_end_mask_0 = const()[name = tensor<string, []>("op_6864_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_6864_cast = slice_by_index(begin = var_6864_begin_0, end = var_6864_end_0, end_mask = var_6864_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_6864_cast")];
            tensor<int32, [4]> var_6868_begin_0 = const()[name = tensor<string, []>("op_6868_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_6868_end_0 = const()[name = tensor<string, []>("op_6868_end_0"), val = tensor<int32, [4]>([2, 480, 1, 256])];
            tensor<bool, [4]> var_6868_end_mask_0 = const()[name = tensor<string, []>("op_6868_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_6868_cast = slice_by_index(begin = var_6868_begin_0, end = var_6868_end_0, end_mask = var_6868_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_6868_cast")];
            tensor<int32, [4]> var_6872_begin_0 = const()[name = tensor<string, []>("op_6872_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_6872_end_0 = const()[name = tensor<string, []>("op_6872_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_6872_end_mask_0 = const()[name = tensor<string, []>("op_6872_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_6872_cast = slice_by_index(begin = var_6872_begin_0, end = var_6872_end_0, end_mask = var_6872_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_6872_cast")];
            tensor<int32, [4]> var_6876_begin_0 = const()[name = tensor<string, []>("op_6876_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_6876_end_0 = const()[name = tensor<string, []>("op_6876_end_0"), val = tensor<int32, [4]>([2, 800, 1, 256])];
            tensor<bool, [4]> var_6876_end_mask_0 = const()[name = tensor<string, []>("op_6876_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_6876_cast = slice_by_index(begin = var_6876_begin_0, end = var_6876_end_0, end_mask = var_6876_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_6876_cast")];
            tensor<int32, [4]> var_6880_begin_0 = const()[name = tensor<string, []>("op_6880_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_6880_end_0 = const()[name = tensor<string, []>("op_6880_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_6880_end_mask_0 = const()[name = tensor<string, []>("op_6880_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_6880_cast = slice_by_index(begin = var_6880_begin_0, end = var_6880_end_0, end_mask = var_6880_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_6880_cast")];
            tensor<int32, [4]> var_6884_begin_0 = const()[name = tensor<string, []>("op_6884_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_6884_end_0 = const()[name = tensor<string, []>("op_6884_end_0"), val = tensor<int32, [4]>([2, 1120, 1, 256])];
            tensor<bool, [4]> var_6884_end_mask_0 = const()[name = tensor<string, []>("op_6884_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_6884_cast = slice_by_index(begin = var_6884_begin_0, end = var_6884_end_0, end_mask = var_6884_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_6884_cast")];
            tensor<int32, [4]> var_6888_begin_0 = const()[name = tensor<string, []>("op_6888_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_6888_end_0 = const()[name = tensor<string, []>("op_6888_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_6888_end_mask_0 = const()[name = tensor<string, []>("op_6888_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_6888_cast = slice_by_index(begin = var_6888_begin_0, end = var_6888_end_0, end_mask = var_6888_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_6888_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_6895_begin_0 = const()[name = tensor<string, []>("op_6895_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_6895_end_0 = const()[name = tensor<string, []>("op_6895_end_0"), val = tensor<int32, [4]>([2, 77, 1, 160])];
            tensor<bool, [4]> var_6895_end_mask_0 = const()[name = tensor<string, []>("op_6895_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)[name = tensor<string, []>("transpose_16")];
            tensor<fp16, [2, 77, 1, 160]> var_6895_cast = slice_by_index(begin = var_6895_begin_0, end = var_6895_end_0, end_mask = var_6895_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_6895_cast")];
            tensor<int32, [4]> var_6899_begin_0 = const()[name = tensor<string, []>("op_6899_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_6899_end_0 = const()[name = tensor<string, []>("op_6899_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_6899_end_mask_0 = const()[name = tensor<string, []>("op_6899_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_6899_cast = slice_by_index(begin = var_6899_begin_0, end = var_6899_end_0, end_mask = var_6899_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_6899_cast")];
            tensor<int32, [4]> var_6903_begin_0 = const()[name = tensor<string, []>("op_6903_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_6903_end_0 = const()[name = tensor<string, []>("op_6903_end_0"), val = tensor<int32, [4]>([2, 77, 1, 480])];
            tensor<bool, [4]> var_6903_end_mask_0 = const()[name = tensor<string, []>("op_6903_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_6903_cast = slice_by_index(begin = var_6903_begin_0, end = var_6903_end_0, end_mask = var_6903_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_6903_cast")];
            tensor<int32, [4]> var_6907_begin_0 = const()[name = tensor<string, []>("op_6907_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 480])];
            tensor<int32, [4]> var_6907_end_0 = const()[name = tensor<string, []>("op_6907_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_6907_end_mask_0 = const()[name = tensor<string, []>("op_6907_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_6907_cast = slice_by_index(begin = var_6907_begin_0, end = var_6907_end_0, end_mask = var_6907_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_6907_cast")];
            tensor<int32, [4]> var_6911_begin_0 = const()[name = tensor<string, []>("op_6911_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_6911_end_0 = const()[name = tensor<string, []>("op_6911_end_0"), val = tensor<int32, [4]>([2, 77, 1, 800])];
            tensor<bool, [4]> var_6911_end_mask_0 = const()[name = tensor<string, []>("op_6911_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_6911_cast = slice_by_index(begin = var_6911_begin_0, end = var_6911_end_0, end_mask = var_6911_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_6911_cast")];
            tensor<int32, [4]> var_6915_begin_0 = const()[name = tensor<string, []>("op_6915_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 800])];
            tensor<int32, [4]> var_6915_end_0 = const()[name = tensor<string, []>("op_6915_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_6915_end_mask_0 = const()[name = tensor<string, []>("op_6915_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_6915_cast = slice_by_index(begin = var_6915_begin_0, end = var_6915_end_0, end_mask = var_6915_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_6915_cast")];
            tensor<int32, [4]> var_6919_begin_0 = const()[name = tensor<string, []>("op_6919_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_6919_end_0 = const()[name = tensor<string, []>("op_6919_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1120])];
            tensor<bool, [4]> var_6919_end_mask_0 = const()[name = tensor<string, []>("op_6919_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_6919_cast = slice_by_index(begin = var_6919_begin_0, end = var_6919_end_0, end_mask = var_6919_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_6919_cast")];
            tensor<int32, [4]> var_6923_begin_0 = const()[name = tensor<string, []>("op_6923_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1120])];
            tensor<int32, [4]> var_6923_end_0 = const()[name = tensor<string, []>("op_6923_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_6923_end_mask_0 = const()[name = tensor<string, []>("op_6923_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_6923_cast = slice_by_index(begin = var_6923_begin_0, end = var_6923_end_0, end_mask = var_6923_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_6923_cast")];
            tensor<int32, [4]> var_6925_begin_0 = const()[name = tensor<string, []>("op_6925_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_6925_end_0 = const()[name = tensor<string, []>("op_6925_end_0"), val = tensor<int32, [4]>([2, 160, 1, 77])];
            tensor<bool, [4]> var_6925_end_mask_0 = const()[name = tensor<string, []>("op_6925_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_6925_cast = slice_by_index(begin = var_6925_begin_0, end = var_6925_end_0, end_mask = var_6925_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_6925_cast")];
            tensor<int32, [4]> var_6929_begin_0 = const()[name = tensor<string, []>("op_6929_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_6929_end_0 = const()[name = tensor<string, []>("op_6929_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_6929_end_mask_0 = const()[name = tensor<string, []>("op_6929_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_6929_cast = slice_by_index(begin = var_6929_begin_0, end = var_6929_end_0, end_mask = var_6929_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_6929_cast")];
            tensor<int32, [4]> var_6933_begin_0 = const()[name = tensor<string, []>("op_6933_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_6933_end_0 = const()[name = tensor<string, []>("op_6933_end_0"), val = tensor<int32, [4]>([2, 480, 1, 77])];
            tensor<bool, [4]> var_6933_end_mask_0 = const()[name = tensor<string, []>("op_6933_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_6933_cast = slice_by_index(begin = var_6933_begin_0, end = var_6933_end_0, end_mask = var_6933_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_6933_cast")];
            tensor<int32, [4]> var_6937_begin_0 = const()[name = tensor<string, []>("op_6937_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_6937_end_0 = const()[name = tensor<string, []>("op_6937_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_6937_end_mask_0 = const()[name = tensor<string, []>("op_6937_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_6937_cast = slice_by_index(begin = var_6937_begin_0, end = var_6937_end_0, end_mask = var_6937_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_6937_cast")];
            tensor<int32, [4]> var_6941_begin_0 = const()[name = tensor<string, []>("op_6941_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_6941_end_0 = const()[name = tensor<string, []>("op_6941_end_0"), val = tensor<int32, [4]>([2, 800, 1, 77])];
            tensor<bool, [4]> var_6941_end_mask_0 = const()[name = tensor<string, []>("op_6941_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_6941_cast = slice_by_index(begin = var_6941_begin_0, end = var_6941_end_0, end_mask = var_6941_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_6941_cast")];
            tensor<int32, [4]> var_6945_begin_0 = const()[name = tensor<string, []>("op_6945_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_6945_end_0 = const()[name = tensor<string, []>("op_6945_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_6945_end_mask_0 = const()[name = tensor<string, []>("op_6945_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_6945_cast = slice_by_index(begin = var_6945_begin_0, end = var_6945_end_0, end_mask = var_6945_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_6945_cast")];
            tensor<int32, [4]> var_6949_begin_0 = const()[name = tensor<string, []>("op_6949_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_6949_end_0 = const()[name = tensor<string, []>("op_6949_end_0"), val = tensor<int32, [4]>([2, 1120, 1, 77])];
            tensor<bool, [4]> var_6949_end_mask_0 = const()[name = tensor<string, []>("op_6949_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_6949_cast = slice_by_index(begin = var_6949_begin_0, end = var_6949_end_0, end_mask = var_6949_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_6949_cast")];
            tensor<int32, [4]> var_6953_begin_0 = const()[name = tensor<string, []>("op_6953_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_6953_end_0 = const()[name = tensor<string, []>("op_6953_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_6953_end_mask_0 = const()[name = tensor<string, []>("op_6953_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_6953_cast = slice_by_index(begin = var_6953_begin_0, end = var_6953_end_0, end_mask = var_6953_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_6953_cast")];
            tensor<string, []> var_6957_equation_0 = const()[name = tensor<string, []>("op_6957_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_6957_cast = einsum(equation = var_6957_equation_0, values = (var_6895_cast, var_6860_cast))[name = tensor<string, []>("op_6957_cast")];
            tensor<fp16, []> var_6958_to_fp16 = const()[name = tensor<string, []>("op_6958_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_113_cast = mul(x = var_6957_cast, y = var_6958_to_fp16)[name = tensor<string, []>("aw_113_cast")];
            tensor<string, []> var_6961_equation_0 = const()[name = tensor<string, []>("op_6961_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_6961_cast = einsum(equation = var_6961_equation_0, values = (var_6899_cast, var_6864_cast))[name = tensor<string, []>("op_6961_cast")];
            tensor<fp16, []> var_6962_to_fp16 = const()[name = tensor<string, []>("op_6962_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_115_cast = mul(x = var_6961_cast, y = var_6962_to_fp16)[name = tensor<string, []>("aw_115_cast")];
            tensor<string, []> var_6965_equation_0 = const()[name = tensor<string, []>("op_6965_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_6965_cast = einsum(equation = var_6965_equation_0, values = (var_6903_cast, var_6868_cast))[name = tensor<string, []>("op_6965_cast")];
            tensor<fp16, []> var_6966_to_fp16 = const()[name = tensor<string, []>("op_6966_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_117_cast = mul(x = var_6965_cast, y = var_6966_to_fp16)[name = tensor<string, []>("aw_117_cast")];
            tensor<string, []> var_6969_equation_0 = const()[name = tensor<string, []>("op_6969_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_6969_cast = einsum(equation = var_6969_equation_0, values = (var_6907_cast, var_6872_cast))[name = tensor<string, []>("op_6969_cast")];
            tensor<fp16, []> var_6970_to_fp16 = const()[name = tensor<string, []>("op_6970_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_119_cast = mul(x = var_6969_cast, y = var_6970_to_fp16)[name = tensor<string, []>("aw_119_cast")];
            tensor<string, []> var_6973_equation_0 = const()[name = tensor<string, []>("op_6973_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_6973_cast = einsum(equation = var_6973_equation_0, values = (var_6911_cast, var_6876_cast))[name = tensor<string, []>("op_6973_cast")];
            tensor<fp16, []> var_6974_to_fp16 = const()[name = tensor<string, []>("op_6974_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_121_cast = mul(x = var_6973_cast, y = var_6974_to_fp16)[name = tensor<string, []>("aw_121_cast")];
            tensor<string, []> var_6977_equation_0 = const()[name = tensor<string, []>("op_6977_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_6977_cast = einsum(equation = var_6977_equation_0, values = (var_6915_cast, var_6880_cast))[name = tensor<string, []>("op_6977_cast")];
            tensor<fp16, []> var_6978_to_fp16 = const()[name = tensor<string, []>("op_6978_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_123_cast = mul(x = var_6977_cast, y = var_6978_to_fp16)[name = tensor<string, []>("aw_123_cast")];
            tensor<string, []> var_6981_equation_0 = const()[name = tensor<string, []>("op_6981_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_6981_cast = einsum(equation = var_6981_equation_0, values = (var_6919_cast, var_6884_cast))[name = tensor<string, []>("op_6981_cast")];
            tensor<fp16, []> var_6982_to_fp16 = const()[name = tensor<string, []>("op_6982_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_125_cast = mul(x = var_6981_cast, y = var_6982_to_fp16)[name = tensor<string, []>("aw_125_cast")];
            tensor<string, []> var_6985_equation_0 = const()[name = tensor<string, []>("op_6985_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_6985_cast = einsum(equation = var_6985_equation_0, values = (var_6923_cast, var_6888_cast))[name = tensor<string, []>("op_6985_cast")];
            tensor<fp16, []> var_6986_to_fp16 = const()[name = tensor<string, []>("op_6986_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_127_cast = mul(x = var_6985_cast, y = var_6986_to_fp16)[name = tensor<string, []>("aw_127_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_6988_cast = softmax(axis = var_6522, x = aw_113_cast)[name = tensor<string, []>("op_6988_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_6989_cast = softmax(axis = var_6522, x = aw_115_cast)[name = tensor<string, []>("op_6989_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_6990_cast = softmax(axis = var_6522, x = aw_117_cast)[name = tensor<string, []>("op_6990_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_6991_cast = softmax(axis = var_6522, x = aw_119_cast)[name = tensor<string, []>("op_6991_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_6992_cast = softmax(axis = var_6522, x = aw_121_cast)[name = tensor<string, []>("op_6992_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_6993_cast = softmax(axis = var_6522, x = aw_123_cast)[name = tensor<string, []>("op_6993_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_6994_cast = softmax(axis = var_6522, x = aw_125_cast)[name = tensor<string, []>("op_6994_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_6995_cast = softmax(axis = var_6522, x = aw_127_cast)[name = tensor<string, []>("op_6995_cast")];
            tensor<string, []> var_6997_equation_0 = const()[name = tensor<string, []>("op_6997_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_6997_cast = einsum(equation = var_6997_equation_0, values = (var_6925_cast, var_6988_cast))[name = tensor<string, []>("op_6997_cast")];
            tensor<string, []> var_6999_equation_0 = const()[name = tensor<string, []>("op_6999_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_6999_cast = einsum(equation = var_6999_equation_0, values = (var_6929_cast, var_6989_cast))[name = tensor<string, []>("op_6999_cast")];
            tensor<string, []> var_7001_equation_0 = const()[name = tensor<string, []>("op_7001_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7001_cast = einsum(equation = var_7001_equation_0, values = (var_6933_cast, var_6990_cast))[name = tensor<string, []>("op_7001_cast")];
            tensor<string, []> var_7003_equation_0 = const()[name = tensor<string, []>("op_7003_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7003_cast = einsum(equation = var_7003_equation_0, values = (var_6937_cast, var_6991_cast))[name = tensor<string, []>("op_7003_cast")];
            tensor<string, []> var_7005_equation_0 = const()[name = tensor<string, []>("op_7005_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7005_cast = einsum(equation = var_7005_equation_0, values = (var_6941_cast, var_6992_cast))[name = tensor<string, []>("op_7005_cast")];
            tensor<string, []> var_7007_equation_0 = const()[name = tensor<string, []>("op_7007_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7007_cast = einsum(equation = var_7007_equation_0, values = (var_6945_cast, var_6993_cast))[name = tensor<string, []>("op_7007_cast")];
            tensor<string, []> var_7009_equation_0 = const()[name = tensor<string, []>("op_7009_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7009_cast = einsum(equation = var_7009_equation_0, values = (var_6949_cast, var_6994_cast))[name = tensor<string, []>("op_7009_cast")];
            tensor<string, []> var_7011_equation_0 = const()[name = tensor<string, []>("op_7011_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7011_cast = einsum(equation = var_7011_equation_0, values = (var_6953_cast, var_6995_cast))[name = tensor<string, []>("op_7011_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_6522, interleave = input_303_interleave_0, values = (var_6997_cast, var_6999_cast, var_7001_cast, var_7003_cast, var_7005_cast, var_7007_cast, var_7009_cast, var_7011_cast))[name = tensor<string, []>("input_303_cast")];
            tensor<int32, [2]> var_7017 = const()[name = tensor<string, []>("op_7017"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7019 = const()[name = tensor<string, []>("op_7019"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_7021_pad_type_0 = const()[name = tensor<string, []>("op_7021_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_7021_pad_0 = const()[name = tensor<string, []>("op_7021_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(429249664))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(430478528))), name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(430478720)))];
            tensor<fp16, [2, 1280, 1, 256]> var_7021_cast = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_7019, groups = var_6522, pad = var_7021_pad_0, pad_type = var_7021_pad_type_0, strides = var_7017, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_303_cast)[name = tensor<string, []>("op_7021_cast")];
            tensor<fp16, [2, 1280, 1, 256]> inputs_47_cast = add(x = var_7021_cast, y = inputs_45_cast)[name = tensor<string, []>("inputs_47_cast")];
            tensor<int32, [1]> var_7025 = const()[name = tensor<string, []>("op_7025"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_47_cast = reduce_mean(axes = var_7025, keep_dims = var_6517, x = inputs_47_cast)[name = tensor<string, []>("channels_mean_47_cast")];
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_47_cast = sub(x = inputs_47_cast, y = channels_mean_47_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_47_cast")];
            tensor<int32, [1]> var_7029 = const()[name = tensor<string, []>("op_7029"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_7030_cast = reduce_mean(axes = var_7029, keep_dims = var_6517, x = zero_mean_sq_47_cast)[name = tensor<string, []>("op_7030_cast")];
            tensor<fp16, []> var_7031_to_fp16 = const()[name = tensor<string, []>("op_7031_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_7032_cast = add(x = var_7030_cast, y = var_7031_to_fp16)[name = tensor<string, []>("op_7032_cast")];
            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_7032_cast)[name = tensor<string, []>("denom_47_cast")];
            tensor<fp16, [2, 1280, 1, 256]> out_47_cast = mul(x = zero_mean_47_cast, y = denom_47_cast)[name = tensor<string, []>("out_47_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_7036_to_fp16 = const()[name = tensor<string, []>("op_7036_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(430481344)))];
            tensor<fp16, [2, 1280, 1, 256]> var_7037_cast = add(x = out_47_cast, y = var_7036_to_fp16)[name = tensor<string, []>("op_7037_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_7039_to_fp16 = const()[name = tensor<string, []>("op_7039_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(430483968)))];
            tensor<fp16, [2, 1280, 1, 256]> input_305_cast = mul(x = var_7037_cast, y = var_7039_to_fp16)[name = tensor<string, []>("input_305_cast")];
            tensor<int32, [2]> var_7047 = const()[name = tensor<string, []>("op_7047"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7049 = const()[name = tensor<string, []>("op_7049"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_7051_pad_type_0 = const()[name = tensor<string, []>("op_7051_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_7051_pad_0 = const()[name = tensor<string, []>("op_7051_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(430486592))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(440317056))), name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [7680]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(440317248))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(440324992))), name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([10240])];
            tensor<fp16, [2, 10240, 1, 256]> var_7051_cast = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_7049, groups = var_6522, pad = var_7051_pad_0, pad_type = var_7051_pad_type_0, strides = var_7047, weight = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_305_cast)[name = tensor<string, []>("op_7051_cast")];
            tensor<int32, [2]> var_7052_split_sizes_0 = const()[name = tensor<string, []>("op_7052_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_7052_axis_0 = const()[name = tensor<string, []>("op_7052_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 256]> var_7052_cast_0, tensor<fp16, [2, 5120, 1, 256]> var_7052_cast_1 = split(axis = var_7052_axis_0, split_sizes = var_7052_split_sizes_0, x = var_7051_cast)[name = tensor<string, []>("op_7052_cast")];
            tensor<string, []> var_7054_mode_0 = const()[name = tensor<string, []>("op_7054_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 256]> var_7054_cast = gelu(mode = var_7054_mode_0, x = var_7052_cast_1)[name = tensor<string, []>("op_7054_cast")];
            tensor<fp16, [2, 5120, 1, 256]> input_307_cast = mul(x = var_7052_cast_0, y = var_7054_cast)[name = tensor<string, []>("input_307_cast")];
            tensor<int32, [2]> var_7058 = const()[name = tensor<string, []>("op_7058"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7060 = const()[name = tensor<string, []>("op_7060"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_7062_pad_type_0 = const()[name = tensor<string, []>("op_7062_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_7062_pad_0 = const()[name = tensor<string, []>("op_7062_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(440325184))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(445240448))), name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            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, []>(445240640)))];
            tensor<fp16, [2, 1280, 1, 256]> var_7062_cast = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_7060, groups = var_6522, pad = var_7062_pad_0, pad_type = var_7062_pad_type_0, strides = var_7058, weight = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_307_cast)[name = tensor<string, []>("op_7062_cast")];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_173_cast = add(x = var_7062_cast, y = inputs_47_cast)[name = tensor<string, []>("hidden_states_173_cast")];
            tensor<int32, [4]> var_7064 = const()[name = tensor<string, []>("op_7064"), val = tensor<int32, [4]>([2, 1280, 16, 16])];
            tensor<fp16, [2, 1280, 16, 16]> input_309_cast = reshape(shape = var_7064, x = hidden_states_173_cast)[name = tensor<string, []>("input_309_cast")];
            tensor<int32, [2]> var_7068 = const()[name = tensor<string, []>("op_7068"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7070 = const()[name = tensor<string, []>("op_7070"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(445243264))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(446472128))), name = tensor<string, []>("up_blocks_1_attentions_0_proj_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(446472320)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_175_cast = conv(bias = up_blocks_1_attentions_0_proj_out_bias_to_fp16, dilations = var_7070, groups = var_6522, pad = hidden_states_175_pad_0, pad_type = hidden_states_175_pad_type_0, strides = var_7068, weight = up_blocks_1_attentions_0_proj_out_weight_to_fp16_palettized, x = input_309_cast)[name = tensor<string, []>("hidden_states_175_cast")];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_177_cast = add(x = hidden_states_175_cast, y = hidden_states_163_cast)[name = tensor<string, []>("hidden_states_177_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_6522, interleave = input_311_interleave_0, values = (hidden_states_177_cast, input_143_cast))[name = tensor<string, []>("input_311_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)[name = tensor<string, []>("reshape_144_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)[name = tensor<string, []>("reduce_mean_108_cast")];
            tensor<fp16, [2, 32, 80, 16, 16]> sub_72_cast = sub(x = reshape_144_cast, y = reduce_mean_108_cast)[name = tensor<string, []>("sub_72_cast")];
            tensor<fp16, [2, 32, 80, 16, 16]> square_36_cast = square(x = sub_72_cast)[name = tensor<string, []>("square_36_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)[name = tensor<string, []>("reduce_mean_110_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)[name = tensor<string, []>("add_72_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_36_cast = sqrt(x = add_72_cast)[name = tensor<string, []>("sqrt_36_cast")];
            tensor<fp16, [2, 32, 80, 16, 16]> real_div_36_cast = real_div(x = sub_72_cast, y = sqrt_36_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_145_cast")];
            tensor<fp16, [2560]> add_73_gamma_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(446474944))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(446476928))), name = tensor<string, []>("add_73_gamma_0_to_fp16_palettized"), shape = tensor<uint32, [1]>([2560])];
            tensor<fp16, [2560]> add_73_beta_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(446477120))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(446479104))), name = tensor<string, []>("add_73_beta_0_to_fp16_palettized"), shape = tensor<uint32, [1]>([2560])];
            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_palettized, epsilon = add_73_epsilon_0_to_fp16, gamma = add_73_gamma_0_to_fp16_palettized, mean = add_55_mean_0_to_fp16_palettized, variance = add_55_variance_0_to_fp16_palettized, x = reshape_145_cast)[name = tensor<string, []>("add_73_cast")];
            tensor<fp16, [2, 2560, 16, 16]> input_315_cast = silu(x = add_73_cast)[name = tensor<string, []>("input_315_cast")];
            tensor<int32, [2]> var_7088 = const()[name = tensor<string, []>("op_7088"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7090 = const()[name = tensor<string, []>("op_7090"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [22118400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(446479296))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(468597760))), name = tensor<string, []>("up_blocks_1_resnets_1_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2560, 3, 3])];
            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, []>(468597952)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_179_cast = conv(bias = up_blocks_1_resnets_1_conv1_bias_to_fp16, dilations = var_7090, groups = var_6522, pad = hidden_states_179_pad_0, pad_type = hidden_states_179_pad_type_0, strides = var_7088, weight = up_blocks_1_resnets_1_conv1_weight_to_fp16_palettized, x = input_315_cast)[name = tensor<string, []>("hidden_states_179_cast")];
            tensor<int32, [2]> var_7096 = const()[name = tensor<string, []>("op_7096"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7098 = const()[name = tensor<string, []>("op_7098"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(468600576))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(469829440))), name = tensor<string, []>("up_blocks_1_resnets_1_time_emb_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(469829632)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_29_cast = conv(bias = up_blocks_1_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_7098, groups = var_6522, pad = temb_29_pad_0, pad_type = temb_29_pad_type_0, strides = var_7096, weight = up_blocks_1_resnets_1_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor<string, []>("temb_29_cast")];
            tensor<fp16, [2, 1280, 16, 16]> input_319_cast = add(x = hidden_states_179_cast, y = temb_29_cast)[name = tensor<string, []>("input_319_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)[name = tensor<string, []>("reshape_148_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)[name = tensor<string, []>("reduce_mean_111_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> sub_74_cast = sub(x = reshape_148_cast, y = reduce_mean_111_cast)[name = tensor<string, []>("sub_74_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> square_37_cast = square(x = sub_74_cast)[name = tensor<string, []>("square_37_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)[name = tensor<string, []>("reduce_mean_113_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)[name = tensor<string, []>("add_74_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_37_cast = sqrt(x = add_74_cast)[name = tensor<string, []>("sqrt_37_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> real_div_37_cast = real_div(x = sub_74_cast, y = sqrt_37_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_149_cast")];
            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, []>(469832256)))];
            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, []>(469834880)))];
            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_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_149_cast)[name = tensor<string, []>("add_75_cast")];
            tensor<fp16, [2, 1280, 16, 16]> input_323_cast = silu(x = add_75_cast)[name = tensor<string, []>("input_323_cast")];
            tensor<int32, [2]> var_7108 = const()[name = tensor<string, []>("op_7108"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7110 = const()[name = tensor<string, []>("op_7110"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [11059200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(469837504))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(480896768))), name = tensor<string, []>("up_blocks_1_resnets_1_conv2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 3, 3])];
            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, []>(480896960)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_181_cast = conv(bias = up_blocks_1_resnets_1_conv2_bias_to_fp16, dilations = var_7110, groups = var_6522, pad = hidden_states_181_pad_0, pad_type = hidden_states_181_pad_type_0, strides = var_7108, weight = up_blocks_1_resnets_1_conv2_weight_to_fp16_palettized, x = input_323_cast)[name = tensor<string, []>("hidden_states_181_cast")];
            tensor<int32, [2]> var_7115 = const()[name = tensor<string, []>("op_7115"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7117 = const()[name = tensor<string, []>("op_7117"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2457600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(480899584))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(483357248))), name = tensor<string, []>("up_blocks_1_resnets_1_conv_shortcut_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 2560, 1, 1])];
            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, []>(483357440)))];
            tensor<fp16, [2, 1280, 16, 16]> x_13_cast = conv(bias = up_blocks_1_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_7117, groups = var_6522, pad = x_13_pad_0, pad_type = x_13_pad_type_0, strides = var_7115, weight = up_blocks_1_resnets_1_conv_shortcut_weight_to_fp16_palettized, x = input_311_cast)[name = tensor<string, []>("x_13_cast")];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_183_cast = add(x = x_13_cast, y = hidden_states_181_cast)[name = tensor<string, []>("hidden_states_183_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)[name = tensor<string, []>("reshape_152_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)[name = tensor<string, []>("reduce_mean_114_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> sub_76_cast = sub(x = reshape_152_cast, y = reduce_mean_114_cast)[name = tensor<string, []>("sub_76_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> square_38_cast = square(x = sub_76_cast)[name = tensor<string, []>("square_38_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)[name = tensor<string, []>("reduce_mean_116_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)[name = tensor<string, []>("add_76_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_38_cast = sqrt(x = add_76_cast)[name = tensor<string, []>("sqrt_38_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> real_div_38_cast = real_div(x = sub_76_cast, y = sqrt_38_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_153_cast")];
            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, []>(483360064)))];
            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, []>(483362688)))];
            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_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_153_cast)[name = tensor<string, []>("add_77_cast")];
            tensor<int32, [2]> var_7137 = const()[name = tensor<string, []>("op_7137"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7139 = const()[name = tensor<string, []>("op_7139"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(483365312))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(484594176))), name = tensor<string, []>("up_blocks_1_attentions_1_proj_in_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(484594368)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_185_cast = conv(bias = up_blocks_1_attentions_1_proj_in_bias_to_fp16, dilations = var_7139, groups = var_6522, pad = hidden_states_185_pad_0, pad_type = hidden_states_185_pad_type_0, strides = var_7137, weight = up_blocks_1_attentions_1_proj_in_weight_to_fp16_palettized, x = add_77_cast)[name = tensor<string, []>("hidden_states_185_cast")];
            tensor<int32, [4]> var_7144 = const()[name = tensor<string, []>("op_7144"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<fp16, [2, 1280, 1, 256]> inputs_49_cast = reshape(shape = var_7144, x = hidden_states_185_cast)[name = tensor<string, []>("inputs_49_cast")];
            tensor<int32, [1]> var_7154 = const()[name = tensor<string, []>("op_7154"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_49_cast = reduce_mean(axes = var_7154, keep_dims = var_6517, x = inputs_49_cast)[name = tensor<string, []>("channels_mean_49_cast")];
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_49_cast = sub(x = inputs_49_cast, y = channels_mean_49_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_49_cast")];
            tensor<int32, [1]> var_7158 = const()[name = tensor<string, []>("op_7158"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_7159_cast = reduce_mean(axes = var_7158, keep_dims = var_6517, x = zero_mean_sq_49_cast)[name = tensor<string, []>("op_7159_cast")];
            tensor<fp16, []> var_7160_to_fp16 = const()[name = tensor<string, []>("op_7160_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_7161_cast = add(x = var_7159_cast, y = var_7160_to_fp16)[name = tensor<string, []>("op_7161_cast")];
            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_7161_cast)[name = tensor<string, []>("denom_49_cast")];
            tensor<fp16, [2, 1280, 1, 256]> out_49_cast = mul(x = zero_mean_49_cast, y = denom_49_cast)[name = tensor<string, []>("out_49_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_7165_to_fp16 = const()[name = tensor<string, []>("op_7165_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(484596992)))];
            tensor<fp16, [2, 1280, 1, 256]> var_7166_cast = add(x = out_49_cast, y = var_7165_to_fp16)[name = tensor<string, []>("op_7166_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_7168_to_fp16 = const()[name = tensor<string, []>("op_7168_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(484599616)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_187_cast = mul(x = var_7166_cast, y = var_7168_to_fp16)[name = tensor<string, []>("hidden_states_187_cast")];
            tensor<int32, [2]> var_7175 = const()[name = tensor<string, []>("op_7175"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7177 = const()[name = tensor<string, []>("op_7177"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(484602240))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(485831104))), name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 256]> q_33_cast = conv(dilations = var_7177, groups = var_6522, pad = q_33_pad_0, pad_type = q_33_pad_type_0, strides = var_7175, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_187_cast)[name = tensor<string, []>("q_33_cast")];
            tensor<int32, [2]> var_7181 = const()[name = tensor<string, []>("op_7181"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7183 = const()[name = tensor<string, []>("op_7183"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(485831296))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(487060160))), name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 256]> k_65_cast = conv(dilations = var_7183, groups = var_6522, pad = k_65_pad_0, pad_type = k_65_pad_type_0, strides = var_7181, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_187_cast)[name = tensor<string, []>("k_65_cast")];
            tensor<int32, [2]> var_7187 = const()[name = tensor<string, []>("op_7187"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7189 = const()[name = tensor<string, []>("op_7189"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(487060352))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(488289216))), name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 256]> v_33_cast = conv(dilations = var_7189, groups = var_6522, pad = v_33_pad_0, pad_type = v_33_pad_type_0, strides = var_7187, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_187_cast)[name = tensor<string, []>("v_33_cast")];
            tensor<int32, [4]> var_7193_begin_0 = const()[name = tensor<string, []>("op_7193_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7193_end_0 = const()[name = tensor<string, []>("op_7193_end_0"), val = tensor<int32, [4]>([2, 160, 1, 256])];
            tensor<bool, [4]> var_7193_end_mask_0 = const()[name = tensor<string, []>("op_7193_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7193_cast = slice_by_index(begin = var_7193_begin_0, end = var_7193_end_0, end_mask = var_7193_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_7193_cast")];
            tensor<int32, [4]> var_7197_begin_0 = const()[name = tensor<string, []>("op_7197_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_7197_end_0 = const()[name = tensor<string, []>("op_7197_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_7197_end_mask_0 = const()[name = tensor<string, []>("op_7197_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7197_cast = slice_by_index(begin = var_7197_begin_0, end = var_7197_end_0, end_mask = var_7197_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_7197_cast")];
            tensor<int32, [4]> var_7201_begin_0 = const()[name = tensor<string, []>("op_7201_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_7201_end_0 = const()[name = tensor<string, []>("op_7201_end_0"), val = tensor<int32, [4]>([2, 480, 1, 256])];
            tensor<bool, [4]> var_7201_end_mask_0 = const()[name = tensor<string, []>("op_7201_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7201_cast = slice_by_index(begin = var_7201_begin_0, end = var_7201_end_0, end_mask = var_7201_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_7201_cast")];
            tensor<int32, [4]> var_7205_begin_0 = const()[name = tensor<string, []>("op_7205_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_7205_end_0 = const()[name = tensor<string, []>("op_7205_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_7205_end_mask_0 = const()[name = tensor<string, []>("op_7205_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7205_cast = slice_by_index(begin = var_7205_begin_0, end = var_7205_end_0, end_mask = var_7205_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_7205_cast")];
            tensor<int32, [4]> var_7209_begin_0 = const()[name = tensor<string, []>("op_7209_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_7209_end_0 = const()[name = tensor<string, []>("op_7209_end_0"), val = tensor<int32, [4]>([2, 800, 1, 256])];
            tensor<bool, [4]> var_7209_end_mask_0 = const()[name = tensor<string, []>("op_7209_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7209_cast = slice_by_index(begin = var_7209_begin_0, end = var_7209_end_0, end_mask = var_7209_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_7209_cast")];
            tensor<int32, [4]> var_7213_begin_0 = const()[name = tensor<string, []>("op_7213_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_7213_end_0 = const()[name = tensor<string, []>("op_7213_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_7213_end_mask_0 = const()[name = tensor<string, []>("op_7213_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7213_cast = slice_by_index(begin = var_7213_begin_0, end = var_7213_end_0, end_mask = var_7213_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_7213_cast")];
            tensor<int32, [4]> var_7217_begin_0 = const()[name = tensor<string, []>("op_7217_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_7217_end_0 = const()[name = tensor<string, []>("op_7217_end_0"), val = tensor<int32, [4]>([2, 1120, 1, 256])];
            tensor<bool, [4]> var_7217_end_mask_0 = const()[name = tensor<string, []>("op_7217_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7217_cast = slice_by_index(begin = var_7217_begin_0, end = var_7217_end_0, end_mask = var_7217_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_7217_cast")];
            tensor<int32, [4]> var_7221_begin_0 = const()[name = tensor<string, []>("op_7221_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_7221_end_0 = const()[name = tensor<string, []>("op_7221_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_7221_end_mask_0 = const()[name = tensor<string, []>("op_7221_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7221_cast = slice_by_index(begin = var_7221_begin_0, end = var_7221_end_0, end_mask = var_7221_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_7221_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_7228_begin_0 = const()[name = tensor<string, []>("op_7228_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7228_end_0 = const()[name = tensor<string, []>("op_7228_end_0"), val = tensor<int32, [4]>([2, 256, 1, 160])];
            tensor<bool, [4]> var_7228_end_mask_0 = const()[name = tensor<string, []>("op_7228_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)[name = tensor<string, []>("transpose_15")];
            tensor<fp16, [2, 256, 1, 160]> var_7228_cast = slice_by_index(begin = var_7228_begin_0, end = var_7228_end_0, end_mask = var_7228_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_7228_cast")];
            tensor<int32, [4]> var_7232_begin_0 = const()[name = tensor<string, []>("op_7232_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_7232_end_0 = const()[name = tensor<string, []>("op_7232_end_0"), val = tensor<int32, [4]>([2, 256, 1, 320])];
            tensor<bool, [4]> var_7232_end_mask_0 = const()[name = tensor<string, []>("op_7232_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_7232_cast = slice_by_index(begin = var_7232_begin_0, end = var_7232_end_0, end_mask = var_7232_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_7232_cast")];
            tensor<int32, [4]> var_7236_begin_0 = const()[name = tensor<string, []>("op_7236_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_7236_end_0 = const()[name = tensor<string, []>("op_7236_end_0"), val = tensor<int32, [4]>([2, 256, 1, 480])];
            tensor<bool, [4]> var_7236_end_mask_0 = const()[name = tensor<string, []>("op_7236_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_7236_cast = slice_by_index(begin = var_7236_begin_0, end = var_7236_end_0, end_mask = var_7236_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_7236_cast")];
            tensor<int32, [4]> var_7240_begin_0 = const()[name = tensor<string, []>("op_7240_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 480])];
            tensor<int32, [4]> var_7240_end_0 = const()[name = tensor<string, []>("op_7240_end_0"), val = tensor<int32, [4]>([2, 256, 1, 640])];
            tensor<bool, [4]> var_7240_end_mask_0 = const()[name = tensor<string, []>("op_7240_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_7240_cast = slice_by_index(begin = var_7240_begin_0, end = var_7240_end_0, end_mask = var_7240_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_7240_cast")];
            tensor<int32, [4]> var_7244_begin_0 = const()[name = tensor<string, []>("op_7244_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_7244_end_0 = const()[name = tensor<string, []>("op_7244_end_0"), val = tensor<int32, [4]>([2, 256, 1, 800])];
            tensor<bool, [4]> var_7244_end_mask_0 = const()[name = tensor<string, []>("op_7244_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_7244_cast = slice_by_index(begin = var_7244_begin_0, end = var_7244_end_0, end_mask = var_7244_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_7244_cast")];
            tensor<int32, [4]> var_7248_begin_0 = const()[name = tensor<string, []>("op_7248_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 800])];
            tensor<int32, [4]> var_7248_end_0 = const()[name = tensor<string, []>("op_7248_end_0"), val = tensor<int32, [4]>([2, 256, 1, 960])];
            tensor<bool, [4]> var_7248_end_mask_0 = const()[name = tensor<string, []>("op_7248_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_7248_cast = slice_by_index(begin = var_7248_begin_0, end = var_7248_end_0, end_mask = var_7248_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_7248_cast")];
            tensor<int32, [4]> var_7252_begin_0 = const()[name = tensor<string, []>("op_7252_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_7252_end_0 = const()[name = tensor<string, []>("op_7252_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1120])];
            tensor<bool, [4]> var_7252_end_mask_0 = const()[name = tensor<string, []>("op_7252_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_7252_cast = slice_by_index(begin = var_7252_begin_0, end = var_7252_end_0, end_mask = var_7252_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_7252_cast")];
            tensor<int32, [4]> var_7256_begin_0 = const()[name = tensor<string, []>("op_7256_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1120])];
            tensor<int32, [4]> var_7256_end_0 = const()[name = tensor<string, []>("op_7256_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1280])];
            tensor<bool, [4]> var_7256_end_mask_0 = const()[name = tensor<string, []>("op_7256_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_7256_cast = slice_by_index(begin = var_7256_begin_0, end = var_7256_end_0, end_mask = var_7256_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_7256_cast")];
            tensor<int32, [4]> var_7258_begin_0 = const()[name = tensor<string, []>("op_7258_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7258_end_0 = const()[name = tensor<string, []>("op_7258_end_0"), val = tensor<int32, [4]>([2, 160, 1, 256])];
            tensor<bool, [4]> var_7258_end_mask_0 = const()[name = tensor<string, []>("op_7258_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7258_cast = slice_by_index(begin = var_7258_begin_0, end = var_7258_end_0, end_mask = var_7258_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_7258_cast")];
            tensor<int32, [4]> var_7262_begin_0 = const()[name = tensor<string, []>("op_7262_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_7262_end_0 = const()[name = tensor<string, []>("op_7262_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_7262_end_mask_0 = const()[name = tensor<string, []>("op_7262_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7262_cast = slice_by_index(begin = var_7262_begin_0, end = var_7262_end_0, end_mask = var_7262_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_7262_cast")];
            tensor<int32, [4]> var_7266_begin_0 = const()[name = tensor<string, []>("op_7266_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_7266_end_0 = const()[name = tensor<string, []>("op_7266_end_0"), val = tensor<int32, [4]>([2, 480, 1, 256])];
            tensor<bool, [4]> var_7266_end_mask_0 = const()[name = tensor<string, []>("op_7266_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7266_cast = slice_by_index(begin = var_7266_begin_0, end = var_7266_end_0, end_mask = var_7266_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_7266_cast")];
            tensor<int32, [4]> var_7270_begin_0 = const()[name = tensor<string, []>("op_7270_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_7270_end_0 = const()[name = tensor<string, []>("op_7270_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_7270_end_mask_0 = const()[name = tensor<string, []>("op_7270_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7270_cast = slice_by_index(begin = var_7270_begin_0, end = var_7270_end_0, end_mask = var_7270_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_7270_cast")];
            tensor<int32, [4]> var_7274_begin_0 = const()[name = tensor<string, []>("op_7274_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_7274_end_0 = const()[name = tensor<string, []>("op_7274_end_0"), val = tensor<int32, [4]>([2, 800, 1, 256])];
            tensor<bool, [4]> var_7274_end_mask_0 = const()[name = tensor<string, []>("op_7274_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7274_cast = slice_by_index(begin = var_7274_begin_0, end = var_7274_end_0, end_mask = var_7274_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_7274_cast")];
            tensor<int32, [4]> var_7278_begin_0 = const()[name = tensor<string, []>("op_7278_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_7278_end_0 = const()[name = tensor<string, []>("op_7278_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_7278_end_mask_0 = const()[name = tensor<string, []>("op_7278_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7278_cast = slice_by_index(begin = var_7278_begin_0, end = var_7278_end_0, end_mask = var_7278_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_7278_cast")];
            tensor<int32, [4]> var_7282_begin_0 = const()[name = tensor<string, []>("op_7282_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_7282_end_0 = const()[name = tensor<string, []>("op_7282_end_0"), val = tensor<int32, [4]>([2, 1120, 1, 256])];
            tensor<bool, [4]> var_7282_end_mask_0 = const()[name = tensor<string, []>("op_7282_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7282_cast = slice_by_index(begin = var_7282_begin_0, end = var_7282_end_0, end_mask = var_7282_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_7282_cast")];
            tensor<int32, [4]> var_7286_begin_0 = const()[name = tensor<string, []>("op_7286_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_7286_end_0 = const()[name = tensor<string, []>("op_7286_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_7286_end_mask_0 = const()[name = tensor<string, []>("op_7286_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7286_cast = slice_by_index(begin = var_7286_begin_0, end = var_7286_end_0, end_mask = var_7286_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_7286_cast")];
            tensor<string, []> var_7290_equation_0 = const()[name = tensor<string, []>("op_7290_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7290_cast = einsum(equation = var_7290_equation_0, values = (var_7228_cast, var_7193_cast))[name = tensor<string, []>("op_7290_cast")];
            tensor<fp16, []> var_7291_to_fp16 = const()[name = tensor<string, []>("op_7291_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_129_cast = mul(x = var_7290_cast, y = var_7291_to_fp16)[name = tensor<string, []>("aw_129_cast")];
            tensor<string, []> var_7294_equation_0 = const()[name = tensor<string, []>("op_7294_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7294_cast = einsum(equation = var_7294_equation_0, values = (var_7232_cast, var_7197_cast))[name = tensor<string, []>("op_7294_cast")];
            tensor<fp16, []> var_7295_to_fp16 = const()[name = tensor<string, []>("op_7295_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_131_cast = mul(x = var_7294_cast, y = var_7295_to_fp16)[name = tensor<string, []>("aw_131_cast")];
            tensor<string, []> var_7298_equation_0 = const()[name = tensor<string, []>("op_7298_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7298_cast = einsum(equation = var_7298_equation_0, values = (var_7236_cast, var_7201_cast))[name = tensor<string, []>("op_7298_cast")];
            tensor<fp16, []> var_7299_to_fp16 = const()[name = tensor<string, []>("op_7299_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_133_cast = mul(x = var_7298_cast, y = var_7299_to_fp16)[name = tensor<string, []>("aw_133_cast")];
            tensor<string, []> var_7302_equation_0 = const()[name = tensor<string, []>("op_7302_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7302_cast = einsum(equation = var_7302_equation_0, values = (var_7240_cast, var_7205_cast))[name = tensor<string, []>("op_7302_cast")];
            tensor<fp16, []> var_7303_to_fp16 = const()[name = tensor<string, []>("op_7303_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_135_cast = mul(x = var_7302_cast, y = var_7303_to_fp16)[name = tensor<string, []>("aw_135_cast")];
            tensor<string, []> var_7306_equation_0 = const()[name = tensor<string, []>("op_7306_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7306_cast = einsum(equation = var_7306_equation_0, values = (var_7244_cast, var_7209_cast))[name = tensor<string, []>("op_7306_cast")];
            tensor<fp16, []> var_7307_to_fp16 = const()[name = tensor<string, []>("op_7307_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_137_cast = mul(x = var_7306_cast, y = var_7307_to_fp16)[name = tensor<string, []>("aw_137_cast")];
            tensor<string, []> var_7310_equation_0 = const()[name = tensor<string, []>("op_7310_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7310_cast = einsum(equation = var_7310_equation_0, values = (var_7248_cast, var_7213_cast))[name = tensor<string, []>("op_7310_cast")];
            tensor<fp16, []> var_7311_to_fp16 = const()[name = tensor<string, []>("op_7311_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_139_cast = mul(x = var_7310_cast, y = var_7311_to_fp16)[name = tensor<string, []>("aw_139_cast")];
            tensor<string, []> var_7314_equation_0 = const()[name = tensor<string, []>("op_7314_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7314_cast = einsum(equation = var_7314_equation_0, values = (var_7252_cast, var_7217_cast))[name = tensor<string, []>("op_7314_cast")];
            tensor<fp16, []> var_7315_to_fp16 = const()[name = tensor<string, []>("op_7315_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_141_cast = mul(x = var_7314_cast, y = var_7315_to_fp16)[name = tensor<string, []>("aw_141_cast")];
            tensor<string, []> var_7318_equation_0 = const()[name = tensor<string, []>("op_7318_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7318_cast = einsum(equation = var_7318_equation_0, values = (var_7256_cast, var_7221_cast))[name = tensor<string, []>("op_7318_cast")];
            tensor<fp16, []> var_7319_to_fp16 = const()[name = tensor<string, []>("op_7319_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_143_cast = mul(x = var_7318_cast, y = var_7319_to_fp16)[name = tensor<string, []>("aw_143_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7321_cast = softmax(axis = var_6522, x = aw_129_cast)[name = tensor<string, []>("op_7321_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7322_cast = softmax(axis = var_6522, x = aw_131_cast)[name = tensor<string, []>("op_7322_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7323_cast = softmax(axis = var_6522, x = aw_133_cast)[name = tensor<string, []>("op_7323_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7324_cast = softmax(axis = var_6522, x = aw_135_cast)[name = tensor<string, []>("op_7324_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7325_cast = softmax(axis = var_6522, x = aw_137_cast)[name = tensor<string, []>("op_7325_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7326_cast = softmax(axis = var_6522, x = aw_139_cast)[name = tensor<string, []>("op_7326_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7327_cast = softmax(axis = var_6522, x = aw_141_cast)[name = tensor<string, []>("op_7327_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7328_cast = softmax(axis = var_6522, x = aw_143_cast)[name = tensor<string, []>("op_7328_cast")];
            tensor<string, []> var_7330_equation_0 = const()[name = tensor<string, []>("op_7330_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7330_cast = einsum(equation = var_7330_equation_0, values = (var_7258_cast, var_7321_cast))[name = tensor<string, []>("op_7330_cast")];
            tensor<string, []> var_7332_equation_0 = const()[name = tensor<string, []>("op_7332_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7332_cast = einsum(equation = var_7332_equation_0, values = (var_7262_cast, var_7322_cast))[name = tensor<string, []>("op_7332_cast")];
            tensor<string, []> var_7334_equation_0 = const()[name = tensor<string, []>("op_7334_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7334_cast = einsum(equation = var_7334_equation_0, values = (var_7266_cast, var_7323_cast))[name = tensor<string, []>("op_7334_cast")];
            tensor<string, []> var_7336_equation_0 = const()[name = tensor<string, []>("op_7336_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7336_cast = einsum(equation = var_7336_equation_0, values = (var_7270_cast, var_7324_cast))[name = tensor<string, []>("op_7336_cast")];
            tensor<string, []> var_7338_equation_0 = const()[name = tensor<string, []>("op_7338_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7338_cast = einsum(equation = var_7338_equation_0, values = (var_7274_cast, var_7325_cast))[name = tensor<string, []>("op_7338_cast")];
            tensor<string, []> var_7340_equation_0 = const()[name = tensor<string, []>("op_7340_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7340_cast = einsum(equation = var_7340_equation_0, values = (var_7278_cast, var_7326_cast))[name = tensor<string, []>("op_7340_cast")];
            tensor<string, []> var_7342_equation_0 = const()[name = tensor<string, []>("op_7342_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7342_cast = einsum(equation = var_7342_equation_0, values = (var_7282_cast, var_7327_cast))[name = tensor<string, []>("op_7342_cast")];
            tensor<string, []> var_7344_equation_0 = const()[name = tensor<string, []>("op_7344_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7344_cast = einsum(equation = var_7344_equation_0, values = (var_7286_cast, var_7328_cast))[name = tensor<string, []>("op_7344_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_6522, interleave = input_327_interleave_0, values = (var_7330_cast, var_7332_cast, var_7334_cast, var_7336_cast, var_7338_cast, var_7340_cast, var_7342_cast, var_7344_cast))[name = tensor<string, []>("input_327_cast")];
            tensor<int32, [2]> var_7350 = const()[name = tensor<string, []>("op_7350"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7352 = const()[name = tensor<string, []>("op_7352"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_7354_pad_type_0 = const()[name = tensor<string, []>("op_7354_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_7354_pad_0 = const()[name = tensor<string, []>("op_7354_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(488289408))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(489518272))), name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(489518464)))];
            tensor<fp16, [2, 1280, 1, 256]> var_7354_cast = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_7352, groups = var_6522, pad = var_7354_pad_0, pad_type = var_7354_pad_type_0, strides = var_7350, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_327_cast)[name = tensor<string, []>("op_7354_cast")];
            tensor<fp16, [2, 1280, 1, 256]> inputs_51_cast = add(x = var_7354_cast, y = inputs_49_cast)[name = tensor<string, []>("inputs_51_cast")];
            tensor<int32, [1]> var_7358 = const()[name = tensor<string, []>("op_7358"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_51_cast = reduce_mean(axes = var_7358, keep_dims = var_6517, x = inputs_51_cast)[name = tensor<string, []>("channels_mean_51_cast")];
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_51_cast = sub(x = inputs_51_cast, y = channels_mean_51_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_51_cast")];
            tensor<int32, [1]> var_7362 = const()[name = tensor<string, []>("op_7362"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_7363_cast = reduce_mean(axes = var_7362, keep_dims = var_6517, x = zero_mean_sq_51_cast)[name = tensor<string, []>("op_7363_cast")];
            tensor<fp16, []> var_7364_to_fp16 = const()[name = tensor<string, []>("op_7364_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_7365_cast = add(x = var_7363_cast, y = var_7364_to_fp16)[name = tensor<string, []>("op_7365_cast")];
            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_7365_cast)[name = tensor<string, []>("denom_51_cast")];
            tensor<fp16, [2, 1280, 1, 256]> out_51_cast = mul(x = zero_mean_51_cast, y = denom_51_cast)[name = tensor<string, []>("out_51_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_7369_to_fp16 = const()[name = tensor<string, []>("op_7369_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(489521088)))];
            tensor<fp16, [2, 1280, 1, 256]> var_7370_cast = add(x = out_51_cast, y = var_7369_to_fp16)[name = tensor<string, []>("op_7370_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_7372_to_fp16 = const()[name = tensor<string, []>("op_7372_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(489523712)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_189_cast = mul(x = var_7370_cast, y = var_7372_to_fp16)[name = tensor<string, []>("hidden_states_189_cast")];
            tensor<int32, [2]> var_7379 = const()[name = tensor<string, []>("op_7379"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7381 = const()[name = tensor<string, []>("op_7381"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(489526336))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(490755200))), name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 256]> q_35_cast = conv(dilations = var_7381, groups = var_6522, pad = q_35_pad_0, pad_type = q_35_pad_type_0, strides = var_7379, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_189_cast)[name = tensor<string, []>("q_35_cast")];
            tensor<int32, [2]> var_7385 = const()[name = tensor<string, []>("op_7385"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7387 = const()[name = tensor<string, []>("op_7387"), 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, 768, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [737280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(490755392))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(491492736))), name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 768, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_69_cast = conv(dilations = var_7387, groups = var_6522, pad = k_69_pad_0, pad_type = k_69_pad_type_0, strides = var_7385, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_69_cast")];
            tensor<int32, [2]> var_7391 = const()[name = tensor<string, []>("op_7391"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7393 = const()[name = tensor<string, []>("op_7393"), 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, 768, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [737280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(491492928))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(492230272))), name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 768, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_35_cast = conv(dilations = var_7393, groups = var_6522, pad = v_35_pad_0, pad_type = v_35_pad_type_0, strides = var_7391, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_35_cast")];
            tensor<int32, [4]> var_7397_begin_0 = const()[name = tensor<string, []>("op_7397_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7397_end_0 = const()[name = tensor<string, []>("op_7397_end_0"), val = tensor<int32, [4]>([2, 160, 1, 256])];
            tensor<bool, [4]> var_7397_end_mask_0 = const()[name = tensor<string, []>("op_7397_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7397_cast = slice_by_index(begin = var_7397_begin_0, end = var_7397_end_0, end_mask = var_7397_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_7397_cast")];
            tensor<int32, [4]> var_7401_begin_0 = const()[name = tensor<string, []>("op_7401_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_7401_end_0 = const()[name = tensor<string, []>("op_7401_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_7401_end_mask_0 = const()[name = tensor<string, []>("op_7401_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7401_cast = slice_by_index(begin = var_7401_begin_0, end = var_7401_end_0, end_mask = var_7401_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_7401_cast")];
            tensor<int32, [4]> var_7405_begin_0 = const()[name = tensor<string, []>("op_7405_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_7405_end_0 = const()[name = tensor<string, []>("op_7405_end_0"), val = tensor<int32, [4]>([2, 480, 1, 256])];
            tensor<bool, [4]> var_7405_end_mask_0 = const()[name = tensor<string, []>("op_7405_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7405_cast = slice_by_index(begin = var_7405_begin_0, end = var_7405_end_0, end_mask = var_7405_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_7405_cast")];
            tensor<int32, [4]> var_7409_begin_0 = const()[name = tensor<string, []>("op_7409_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_7409_end_0 = const()[name = tensor<string, []>("op_7409_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_7409_end_mask_0 = const()[name = tensor<string, []>("op_7409_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7409_cast = slice_by_index(begin = var_7409_begin_0, end = var_7409_end_0, end_mask = var_7409_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_7409_cast")];
            tensor<int32, [4]> var_7413_begin_0 = const()[name = tensor<string, []>("op_7413_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_7413_end_0 = const()[name = tensor<string, []>("op_7413_end_0"), val = tensor<int32, [4]>([2, 800, 1, 256])];
            tensor<bool, [4]> var_7413_end_mask_0 = const()[name = tensor<string, []>("op_7413_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7413_cast = slice_by_index(begin = var_7413_begin_0, end = var_7413_end_0, end_mask = var_7413_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_7413_cast")];
            tensor<int32, [4]> var_7417_begin_0 = const()[name = tensor<string, []>("op_7417_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_7417_end_0 = const()[name = tensor<string, []>("op_7417_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_7417_end_mask_0 = const()[name = tensor<string, []>("op_7417_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7417_cast = slice_by_index(begin = var_7417_begin_0, end = var_7417_end_0, end_mask = var_7417_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_7417_cast")];
            tensor<int32, [4]> var_7421_begin_0 = const()[name = tensor<string, []>("op_7421_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_7421_end_0 = const()[name = tensor<string, []>("op_7421_end_0"), val = tensor<int32, [4]>([2, 1120, 1, 256])];
            tensor<bool, [4]> var_7421_end_mask_0 = const()[name = tensor<string, []>("op_7421_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7421_cast = slice_by_index(begin = var_7421_begin_0, end = var_7421_end_0, end_mask = var_7421_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_7421_cast")];
            tensor<int32, [4]> var_7425_begin_0 = const()[name = tensor<string, []>("op_7425_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_7425_end_0 = const()[name = tensor<string, []>("op_7425_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_7425_end_mask_0 = const()[name = tensor<string, []>("op_7425_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7425_cast = slice_by_index(begin = var_7425_begin_0, end = var_7425_end_0, end_mask = var_7425_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_7425_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_7432_begin_0 = const()[name = tensor<string, []>("op_7432_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7432_end_0 = const()[name = tensor<string, []>("op_7432_end_0"), val = tensor<int32, [4]>([2, 77, 1, 160])];
            tensor<bool, [4]> var_7432_end_mask_0 = const()[name = tensor<string, []>("op_7432_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)[name = tensor<string, []>("transpose_14")];
            tensor<fp16, [2, 77, 1, 160]> var_7432_cast = slice_by_index(begin = var_7432_begin_0, end = var_7432_end_0, end_mask = var_7432_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_7432_cast")];
            tensor<int32, [4]> var_7436_begin_0 = const()[name = tensor<string, []>("op_7436_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_7436_end_0 = const()[name = tensor<string, []>("op_7436_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_7436_end_mask_0 = const()[name = tensor<string, []>("op_7436_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_7436_cast = slice_by_index(begin = var_7436_begin_0, end = var_7436_end_0, end_mask = var_7436_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_7436_cast")];
            tensor<int32, [4]> var_7440_begin_0 = const()[name = tensor<string, []>("op_7440_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_7440_end_0 = const()[name = tensor<string, []>("op_7440_end_0"), val = tensor<int32, [4]>([2, 77, 1, 480])];
            tensor<bool, [4]> var_7440_end_mask_0 = const()[name = tensor<string, []>("op_7440_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_7440_cast = slice_by_index(begin = var_7440_begin_0, end = var_7440_end_0, end_mask = var_7440_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_7440_cast")];
            tensor<int32, [4]> var_7444_begin_0 = const()[name = tensor<string, []>("op_7444_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 480])];
            tensor<int32, [4]> var_7444_end_0 = const()[name = tensor<string, []>("op_7444_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_7444_end_mask_0 = const()[name = tensor<string, []>("op_7444_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_7444_cast = slice_by_index(begin = var_7444_begin_0, end = var_7444_end_0, end_mask = var_7444_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_7444_cast")];
            tensor<int32, [4]> var_7448_begin_0 = const()[name = tensor<string, []>("op_7448_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_7448_end_0 = const()[name = tensor<string, []>("op_7448_end_0"), val = tensor<int32, [4]>([2, 77, 1, 800])];
            tensor<bool, [4]> var_7448_end_mask_0 = const()[name = tensor<string, []>("op_7448_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_7448_cast = slice_by_index(begin = var_7448_begin_0, end = var_7448_end_0, end_mask = var_7448_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_7448_cast")];
            tensor<int32, [4]> var_7452_begin_0 = const()[name = tensor<string, []>("op_7452_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 800])];
            tensor<int32, [4]> var_7452_end_0 = const()[name = tensor<string, []>("op_7452_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_7452_end_mask_0 = const()[name = tensor<string, []>("op_7452_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_7452_cast = slice_by_index(begin = var_7452_begin_0, end = var_7452_end_0, end_mask = var_7452_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_7452_cast")];
            tensor<int32, [4]> var_7456_begin_0 = const()[name = tensor<string, []>("op_7456_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_7456_end_0 = const()[name = tensor<string, []>("op_7456_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1120])];
            tensor<bool, [4]> var_7456_end_mask_0 = const()[name = tensor<string, []>("op_7456_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_7456_cast = slice_by_index(begin = var_7456_begin_0, end = var_7456_end_0, end_mask = var_7456_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_7456_cast")];
            tensor<int32, [4]> var_7460_begin_0 = const()[name = tensor<string, []>("op_7460_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1120])];
            tensor<int32, [4]> var_7460_end_0 = const()[name = tensor<string, []>("op_7460_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_7460_end_mask_0 = const()[name = tensor<string, []>("op_7460_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_7460_cast = slice_by_index(begin = var_7460_begin_0, end = var_7460_end_0, end_mask = var_7460_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_7460_cast")];
            tensor<int32, [4]> var_7462_begin_0 = const()[name = tensor<string, []>("op_7462_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7462_end_0 = const()[name = tensor<string, []>("op_7462_end_0"), val = tensor<int32, [4]>([2, 160, 1, 77])];
            tensor<bool, [4]> var_7462_end_mask_0 = const()[name = tensor<string, []>("op_7462_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_7462_cast = slice_by_index(begin = var_7462_begin_0, end = var_7462_end_0, end_mask = var_7462_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_7462_cast")];
            tensor<int32, [4]> var_7466_begin_0 = const()[name = tensor<string, []>("op_7466_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_7466_end_0 = const()[name = tensor<string, []>("op_7466_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_7466_end_mask_0 = const()[name = tensor<string, []>("op_7466_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_7466_cast = slice_by_index(begin = var_7466_begin_0, end = var_7466_end_0, end_mask = var_7466_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_7466_cast")];
            tensor<int32, [4]> var_7470_begin_0 = const()[name = tensor<string, []>("op_7470_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_7470_end_0 = const()[name = tensor<string, []>("op_7470_end_0"), val = tensor<int32, [4]>([2, 480, 1, 77])];
            tensor<bool, [4]> var_7470_end_mask_0 = const()[name = tensor<string, []>("op_7470_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_7470_cast = slice_by_index(begin = var_7470_begin_0, end = var_7470_end_0, end_mask = var_7470_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_7470_cast")];
            tensor<int32, [4]> var_7474_begin_0 = const()[name = tensor<string, []>("op_7474_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_7474_end_0 = const()[name = tensor<string, []>("op_7474_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_7474_end_mask_0 = const()[name = tensor<string, []>("op_7474_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_7474_cast = slice_by_index(begin = var_7474_begin_0, end = var_7474_end_0, end_mask = var_7474_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_7474_cast")];
            tensor<int32, [4]> var_7478_begin_0 = const()[name = tensor<string, []>("op_7478_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_7478_end_0 = const()[name = tensor<string, []>("op_7478_end_0"), val = tensor<int32, [4]>([2, 800, 1, 77])];
            tensor<bool, [4]> var_7478_end_mask_0 = const()[name = tensor<string, []>("op_7478_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_7478_cast = slice_by_index(begin = var_7478_begin_0, end = var_7478_end_0, end_mask = var_7478_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_7478_cast")];
            tensor<int32, [4]> var_7482_begin_0 = const()[name = tensor<string, []>("op_7482_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_7482_end_0 = const()[name = tensor<string, []>("op_7482_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_7482_end_mask_0 = const()[name = tensor<string, []>("op_7482_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_7482_cast = slice_by_index(begin = var_7482_begin_0, end = var_7482_end_0, end_mask = var_7482_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_7482_cast")];
            tensor<int32, [4]> var_7486_begin_0 = const()[name = tensor<string, []>("op_7486_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_7486_end_0 = const()[name = tensor<string, []>("op_7486_end_0"), val = tensor<int32, [4]>([2, 1120, 1, 77])];
            tensor<bool, [4]> var_7486_end_mask_0 = const()[name = tensor<string, []>("op_7486_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_7486_cast = slice_by_index(begin = var_7486_begin_0, end = var_7486_end_0, end_mask = var_7486_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_7486_cast")];
            tensor<int32, [4]> var_7490_begin_0 = const()[name = tensor<string, []>("op_7490_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_7490_end_0 = const()[name = tensor<string, []>("op_7490_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_7490_end_mask_0 = const()[name = tensor<string, []>("op_7490_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_7490_cast = slice_by_index(begin = var_7490_begin_0, end = var_7490_end_0, end_mask = var_7490_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_7490_cast")];
            tensor<string, []> var_7494_equation_0 = const()[name = tensor<string, []>("op_7494_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_7494_cast = einsum(equation = var_7494_equation_0, values = (var_7432_cast, var_7397_cast))[name = tensor<string, []>("op_7494_cast")];
            tensor<fp16, []> var_7495_to_fp16 = const()[name = tensor<string, []>("op_7495_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_145_cast = mul(x = var_7494_cast, y = var_7495_to_fp16)[name = tensor<string, []>("aw_145_cast")];
            tensor<string, []> var_7498_equation_0 = const()[name = tensor<string, []>("op_7498_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_7498_cast = einsum(equation = var_7498_equation_0, values = (var_7436_cast, var_7401_cast))[name = tensor<string, []>("op_7498_cast")];
            tensor<fp16, []> var_7499_to_fp16 = const()[name = tensor<string, []>("op_7499_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_147_cast = mul(x = var_7498_cast, y = var_7499_to_fp16)[name = tensor<string, []>("aw_147_cast")];
            tensor<string, []> var_7502_equation_0 = const()[name = tensor<string, []>("op_7502_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_7502_cast = einsum(equation = var_7502_equation_0, values = (var_7440_cast, var_7405_cast))[name = tensor<string, []>("op_7502_cast")];
            tensor<fp16, []> var_7503_to_fp16 = const()[name = tensor<string, []>("op_7503_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_149_cast = mul(x = var_7502_cast, y = var_7503_to_fp16)[name = tensor<string, []>("aw_149_cast")];
            tensor<string, []> var_7506_equation_0 = const()[name = tensor<string, []>("op_7506_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_7506_cast = einsum(equation = var_7506_equation_0, values = (var_7444_cast, var_7409_cast))[name = tensor<string, []>("op_7506_cast")];
            tensor<fp16, []> var_7507_to_fp16 = const()[name = tensor<string, []>("op_7507_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_151_cast = mul(x = var_7506_cast, y = var_7507_to_fp16)[name = tensor<string, []>("aw_151_cast")];
            tensor<string, []> var_7510_equation_0 = const()[name = tensor<string, []>("op_7510_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_7510_cast = einsum(equation = var_7510_equation_0, values = (var_7448_cast, var_7413_cast))[name = tensor<string, []>("op_7510_cast")];
            tensor<fp16, []> var_7511_to_fp16 = const()[name = tensor<string, []>("op_7511_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_153_cast = mul(x = var_7510_cast, y = var_7511_to_fp16)[name = tensor<string, []>("aw_153_cast")];
            tensor<string, []> var_7514_equation_0 = const()[name = tensor<string, []>("op_7514_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_7514_cast = einsum(equation = var_7514_equation_0, values = (var_7452_cast, var_7417_cast))[name = tensor<string, []>("op_7514_cast")];
            tensor<fp16, []> var_7515_to_fp16 = const()[name = tensor<string, []>("op_7515_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_155_cast = mul(x = var_7514_cast, y = var_7515_to_fp16)[name = tensor<string, []>("aw_155_cast")];
            tensor<string, []> var_7518_equation_0 = const()[name = tensor<string, []>("op_7518_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_7518_cast = einsum(equation = var_7518_equation_0, values = (var_7456_cast, var_7421_cast))[name = tensor<string, []>("op_7518_cast")];
            tensor<fp16, []> var_7519_to_fp16 = const()[name = tensor<string, []>("op_7519_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_157_cast = mul(x = var_7518_cast, y = var_7519_to_fp16)[name = tensor<string, []>("aw_157_cast")];
            tensor<string, []> var_7522_equation_0 = const()[name = tensor<string, []>("op_7522_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_7522_cast = einsum(equation = var_7522_equation_0, values = (var_7460_cast, var_7425_cast))[name = tensor<string, []>("op_7522_cast")];
            tensor<fp16, []> var_7523_to_fp16 = const()[name = tensor<string, []>("op_7523_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_159_cast = mul(x = var_7522_cast, y = var_7523_to_fp16)[name = tensor<string, []>("aw_159_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_7525_cast = softmax(axis = var_6522, x = aw_145_cast)[name = tensor<string, []>("op_7525_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_7526_cast = softmax(axis = var_6522, x = aw_147_cast)[name = tensor<string, []>("op_7526_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_7527_cast = softmax(axis = var_6522, x = aw_149_cast)[name = tensor<string, []>("op_7527_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_7528_cast = softmax(axis = var_6522, x = aw_151_cast)[name = tensor<string, []>("op_7528_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_7529_cast = softmax(axis = var_6522, x = aw_153_cast)[name = tensor<string, []>("op_7529_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_7530_cast = softmax(axis = var_6522, x = aw_155_cast)[name = tensor<string, []>("op_7530_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_7531_cast = softmax(axis = var_6522, x = aw_157_cast)[name = tensor<string, []>("op_7531_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_7532_cast = softmax(axis = var_6522, x = aw_159_cast)[name = tensor<string, []>("op_7532_cast")];
            tensor<string, []> var_7534_equation_0 = const()[name = tensor<string, []>("op_7534_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7534_cast = einsum(equation = var_7534_equation_0, values = (var_7462_cast, var_7525_cast))[name = tensor<string, []>("op_7534_cast")];
            tensor<string, []> var_7536_equation_0 = const()[name = tensor<string, []>("op_7536_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7536_cast = einsum(equation = var_7536_equation_0, values = (var_7466_cast, var_7526_cast))[name = tensor<string, []>("op_7536_cast")];
            tensor<string, []> var_7538_equation_0 = const()[name = tensor<string, []>("op_7538_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7538_cast = einsum(equation = var_7538_equation_0, values = (var_7470_cast, var_7527_cast))[name = tensor<string, []>("op_7538_cast")];
            tensor<string, []> var_7540_equation_0 = const()[name = tensor<string, []>("op_7540_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7540_cast = einsum(equation = var_7540_equation_0, values = (var_7474_cast, var_7528_cast))[name = tensor<string, []>("op_7540_cast")];
            tensor<string, []> var_7542_equation_0 = const()[name = tensor<string, []>("op_7542_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7542_cast = einsum(equation = var_7542_equation_0, values = (var_7478_cast, var_7529_cast))[name = tensor<string, []>("op_7542_cast")];
            tensor<string, []> var_7544_equation_0 = const()[name = tensor<string, []>("op_7544_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7544_cast = einsum(equation = var_7544_equation_0, values = (var_7482_cast, var_7530_cast))[name = tensor<string, []>("op_7544_cast")];
            tensor<string, []> var_7546_equation_0 = const()[name = tensor<string, []>("op_7546_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7546_cast = einsum(equation = var_7546_equation_0, values = (var_7486_cast, var_7531_cast))[name = tensor<string, []>("op_7546_cast")];
            tensor<string, []> var_7548_equation_0 = const()[name = tensor<string, []>("op_7548_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7548_cast = einsum(equation = var_7548_equation_0, values = (var_7490_cast, var_7532_cast))[name = tensor<string, []>("op_7548_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_6522, interleave = input_329_interleave_0, values = (var_7534_cast, var_7536_cast, var_7538_cast, var_7540_cast, var_7542_cast, var_7544_cast, var_7546_cast, var_7548_cast))[name = tensor<string, []>("input_329_cast")];
            tensor<int32, [2]> var_7554 = const()[name = tensor<string, []>("op_7554"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7556 = const()[name = tensor<string, []>("op_7556"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_7558_pad_type_0 = const()[name = tensor<string, []>("op_7558_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_7558_pad_0 = const()[name = tensor<string, []>("op_7558_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(492230464))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(493459328))), name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(493459520)))];
            tensor<fp16, [2, 1280, 1, 256]> var_7558_cast = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_7556, groups = var_6522, pad = var_7558_pad_0, pad_type = var_7558_pad_type_0, strides = var_7554, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_329_cast)[name = tensor<string, []>("op_7558_cast")];
            tensor<fp16, [2, 1280, 1, 256]> inputs_53_cast = add(x = var_7558_cast, y = inputs_51_cast)[name = tensor<string, []>("inputs_53_cast")];
            tensor<int32, [1]> var_7562 = const()[name = tensor<string, []>("op_7562"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_53_cast = reduce_mean(axes = var_7562, keep_dims = var_6517, x = inputs_53_cast)[name = tensor<string, []>("channels_mean_53_cast")];
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_53_cast = sub(x = inputs_53_cast, y = channels_mean_53_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_53_cast")];
            tensor<int32, [1]> var_7566 = const()[name = tensor<string, []>("op_7566"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_7567_cast = reduce_mean(axes = var_7566, keep_dims = var_6517, x = zero_mean_sq_53_cast)[name = tensor<string, []>("op_7567_cast")];
            tensor<fp16, []> var_7568_to_fp16 = const()[name = tensor<string, []>("op_7568_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_7569_cast = add(x = var_7567_cast, y = var_7568_to_fp16)[name = tensor<string, []>("op_7569_cast")];
            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_7569_cast)[name = tensor<string, []>("denom_53_cast")];
            tensor<fp16, [2, 1280, 1, 256]> out_53_cast = mul(x = zero_mean_53_cast, y = denom_53_cast)[name = tensor<string, []>("out_53_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_7573_to_fp16 = const()[name = tensor<string, []>("op_7573_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(493462144)))];
            tensor<fp16, [2, 1280, 1, 256]> var_7574_cast = add(x = out_53_cast, y = var_7573_to_fp16)[name = tensor<string, []>("op_7574_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_7576_to_fp16 = const()[name = tensor<string, []>("op_7576_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(493464768)))];
            tensor<fp16, [2, 1280, 1, 256]> input_331_cast = mul(x = var_7574_cast, y = var_7576_to_fp16)[name = tensor<string, []>("input_331_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, [10240, 1280, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(493467392))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(503297856))), name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [7680]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(503298048))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(503305792))), name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([10240])];
            tensor<fp16, [2, 10240, 1, 256]> var_7588_cast = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_7586, groups = var_6522, 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_0_proj_weight_to_fp16_palettized, x = input_331_cast)[name = tensor<string, []>("op_7588_cast")];
            tensor<int32, [2]> var_7589_split_sizes_0 = const()[name = tensor<string, []>("op_7589_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_7589_axis_0 = const()[name = tensor<string, []>("op_7589_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 256]> var_7589_cast_0, tensor<fp16, [2, 5120, 1, 256]> var_7589_cast_1 = split(axis = var_7589_axis_0, split_sizes = var_7589_split_sizes_0, x = var_7588_cast)[name = tensor<string, []>("op_7589_cast")];
            tensor<string, []> var_7591_mode_0 = const()[name = tensor<string, []>("op_7591_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 256]> var_7591_cast = gelu(mode = var_7591_mode_0, x = var_7589_cast_1)[name = tensor<string, []>("op_7591_cast")];
            tensor<fp16, [2, 5120, 1, 256]> input_333_cast = mul(x = var_7589_cast_0, y = var_7591_cast)[name = tensor<string, []>("input_333_cast")];
            tensor<int32, [2]> var_7595 = const()[name = tensor<string, []>("op_7595"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7597 = const()[name = tensor<string, []>("op_7597"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_7599_pad_type_0 = const()[name = tensor<string, []>("op_7599_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_7599_pad_0 = const()[name = tensor<string, []>("op_7599_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(503305984))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(508221248))), name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            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, []>(508221440)))];
            tensor<fp16, [2, 1280, 1, 256]> var_7599_cast = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_7597, groups = var_6522, pad = var_7599_pad_0, pad_type = var_7599_pad_type_0, strides = var_7595, weight = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_333_cast)[name = tensor<string, []>("op_7599_cast")];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_193_cast = add(x = var_7599_cast, y = inputs_53_cast)[name = tensor<string, []>("hidden_states_193_cast")];
            tensor<int32, [4]> var_7601 = const()[name = tensor<string, []>("op_7601"), val = tensor<int32, [4]>([2, 1280, 16, 16])];
            tensor<fp16, [2, 1280, 16, 16]> input_335_cast = reshape(shape = var_7601, x = hidden_states_193_cast)[name = tensor<string, []>("input_335_cast")];
            tensor<int32, [2]> var_7605 = const()[name = tensor<string, []>("op_7605"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7607 = const()[name = tensor<string, []>("op_7607"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(508224064))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(509452928))), name = tensor<string, []>("up_blocks_1_attentions_1_proj_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(509453120)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_195_cast = conv(bias = up_blocks_1_attentions_1_proj_out_bias_to_fp16, dilations = var_7607, groups = var_6522, pad = hidden_states_195_pad_0, pad_type = hidden_states_195_pad_type_0, strides = var_7605, weight = up_blocks_1_attentions_1_proj_out_weight_to_fp16_palettized, x = input_335_cast)[name = tensor<string, []>("hidden_states_195_cast")];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_197_cast = add(x = hidden_states_195_cast, y = hidden_states_183_cast)[name = tensor<string, []>("hidden_states_197_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_6522, interleave = input_337_interleave_0, values = (hidden_states_197_cast, input_117_cast))[name = tensor<string, []>("input_337_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)[name = tensor<string, []>("reshape_156_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)[name = tensor<string, []>("reduce_mean_117_cast")];
            tensor<fp16, [2, 32, 60, 16, 16]> sub_78_cast = sub(x = reshape_156_cast, y = reduce_mean_117_cast)[name = tensor<string, []>("sub_78_cast")];
            tensor<fp16, [2, 32, 60, 16, 16]> square_39_cast = square(x = sub_78_cast)[name = tensor<string, []>("square_39_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)[name = tensor<string, []>("reduce_mean_119_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)[name = tensor<string, []>("add_78_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_39_cast = sqrt(x = add_78_cast)[name = tensor<string, []>("sqrt_39_cast")];
            tensor<fp16, [2, 32, 60, 16, 16]> real_div_39_cast = real_div(x = sub_78_cast, y = sqrt_39_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_157_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, []>(509455744)))];
            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, []>(509459648)))];
            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, []>(509463552)))];
            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, []>(509467456)))];
            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)[name = tensor<string, []>("add_79_cast")];
            tensor<fp16, [2, 1920, 16, 16]> input_341_cast = silu(x = add_79_cast)[name = tensor<string, []>("input_341_cast")];
            tensor<int32, [2]> var_7625 = const()[name = tensor<string, []>("op_7625"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7627 = const()[name = tensor<string, []>("op_7627"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [16588800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(509471360))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(526060224))), name = tensor<string, []>("up_blocks_1_resnets_2_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1920, 3, 3])];
            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, []>(526060416)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_199_cast = conv(bias = up_blocks_1_resnets_2_conv1_bias_to_fp16, dilations = var_7627, groups = var_6522, pad = hidden_states_199_pad_0, pad_type = hidden_states_199_pad_type_0, strides = var_7625, weight = up_blocks_1_resnets_2_conv1_weight_to_fp16_palettized, x = input_341_cast)[name = tensor<string, []>("hidden_states_199_cast")];
            tensor<int32, [2]> var_7633 = const()[name = tensor<string, []>("op_7633"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7635 = const()[name = tensor<string, []>("op_7635"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(526063040))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(527291904))), name = tensor<string, []>("up_blocks_1_resnets_2_time_emb_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(527292096)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_31_cast = conv(bias = up_blocks_1_resnets_2_time_emb_proj_bias_to_fp16, dilations = var_7635, groups = var_6522, pad = temb_31_pad_0, pad_type = temb_31_pad_type_0, strides = var_7633, weight = up_blocks_1_resnets_2_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor<string, []>("temb_31_cast")];
            tensor<fp16, [2, 1280, 16, 16]> input_345_cast = add(x = hidden_states_199_cast, y = temb_31_cast)[name = tensor<string, []>("input_345_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)[name = tensor<string, []>("reshape_160_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)[name = tensor<string, []>("reduce_mean_120_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> sub_80_cast = sub(x = reshape_160_cast, y = reduce_mean_120_cast)[name = tensor<string, []>("sub_80_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> square_40_cast = square(x = sub_80_cast)[name = tensor<string, []>("square_40_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)[name = tensor<string, []>("reduce_mean_122_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)[name = tensor<string, []>("add_80_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_40_cast = sqrt(x = add_80_cast)[name = tensor<string, []>("sqrt_40_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> real_div_40_cast = real_div(x = sub_80_cast, y = sqrt_40_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_161_cast")];
            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, []>(527294720)))];
            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, []>(527297344)))];
            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_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_161_cast)[name = tensor<string, []>("add_81_cast")];
            tensor<fp16, [2, 1280, 16, 16]> input_349_cast = silu(x = add_81_cast)[name = tensor<string, []>("input_349_cast")];
            tensor<int32, [2]> var_7645 = const()[name = tensor<string, []>("op_7645"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7647 = const()[name = tensor<string, []>("op_7647"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [11059200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(527299968))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(538359232))), name = tensor<string, []>("up_blocks_1_resnets_2_conv2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 3, 3])];
            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, []>(538359424)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_201_cast = conv(bias = up_blocks_1_resnets_2_conv2_bias_to_fp16, dilations = var_7647, groups = var_6522, pad = hidden_states_201_pad_0, pad_type = hidden_states_201_pad_type_0, strides = var_7645, weight = up_blocks_1_resnets_2_conv2_weight_to_fp16_palettized, x = input_349_cast)[name = tensor<string, []>("hidden_states_201_cast")];
            tensor<int32, [2]> var_7652 = const()[name = tensor<string, []>("op_7652"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7654 = const()[name = tensor<string, []>("op_7654"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1843200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(538362048))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(540205312))), name = tensor<string, []>("up_blocks_1_resnets_2_conv_shortcut_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1920, 1, 1])];
            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, []>(540205504)))];
            tensor<fp16, [2, 1280, 16, 16]> x_15_cast = conv(bias = up_blocks_1_resnets_2_conv_shortcut_bias_to_fp16, dilations = var_7654, groups = var_6522, pad = x_15_pad_0, pad_type = x_15_pad_type_0, strides = var_7652, weight = up_blocks_1_resnets_2_conv_shortcut_weight_to_fp16_palettized, x = input_337_cast)[name = tensor<string, []>("x_15_cast")];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_203_cast = add(x = x_15_cast, y = hidden_states_201_cast)[name = tensor<string, []>("hidden_states_203_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)[name = tensor<string, []>("reshape_164_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)[name = tensor<string, []>("reduce_mean_123_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> sub_82_cast = sub(x = reshape_164_cast, y = reduce_mean_123_cast)[name = tensor<string, []>("sub_82_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> square_41_cast = square(x = sub_82_cast)[name = tensor<string, []>("square_41_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)[name = tensor<string, []>("reduce_mean_125_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)[name = tensor<string, []>("add_82_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_41_cast = sqrt(x = add_82_cast)[name = tensor<string, []>("sqrt_41_cast")];
            tensor<fp16, [2, 32, 40, 16, 16]> real_div_41_cast = real_div(x = sub_82_cast, y = sqrt_41_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_165_cast")];
            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, []>(540208128)))];
            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, []>(540210752)))];
            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_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_165_cast)[name = tensor<string, []>("add_83_cast")];
            tensor<int32, [2]> var_7674 = const()[name = tensor<string, []>("op_7674"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7676 = const()[name = tensor<string, []>("op_7676"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(540213376))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(541442240))), name = tensor<string, []>("up_blocks_1_attentions_2_proj_in_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(541442432)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_205_cast = conv(bias = up_blocks_1_attentions_2_proj_in_bias_to_fp16, dilations = var_7676, groups = var_6522, pad = hidden_states_205_pad_0, pad_type = hidden_states_205_pad_type_0, strides = var_7674, weight = up_blocks_1_attentions_2_proj_in_weight_to_fp16_palettized, x = add_83_cast)[name = tensor<string, []>("hidden_states_205_cast")];
            tensor<int32, [4]> var_7681 = const()[name = tensor<string, []>("op_7681"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<fp16, [2, 1280, 1, 256]> inputs_55_cast = reshape(shape = var_7681, x = hidden_states_205_cast)[name = tensor<string, []>("inputs_55_cast")];
            tensor<int32, [1]> var_7691 = const()[name = tensor<string, []>("op_7691"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_55_cast = reduce_mean(axes = var_7691, keep_dims = var_6517, x = inputs_55_cast)[name = tensor<string, []>("channels_mean_55_cast")];
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_55_cast = sub(x = inputs_55_cast, y = channels_mean_55_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_55_cast")];
            tensor<int32, [1]> var_7695 = const()[name = tensor<string, []>("op_7695"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_7696_cast = reduce_mean(axes = var_7695, keep_dims = var_6517, x = zero_mean_sq_55_cast)[name = tensor<string, []>("op_7696_cast")];
            tensor<fp16, []> var_7697_to_fp16 = const()[name = tensor<string, []>("op_7697_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_7698_cast = add(x = var_7696_cast, y = var_7697_to_fp16)[name = tensor<string, []>("op_7698_cast")];
            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_7698_cast)[name = tensor<string, []>("denom_55_cast")];
            tensor<fp16, [2, 1280, 1, 256]> out_55_cast = mul(x = zero_mean_55_cast, y = denom_55_cast)[name = tensor<string, []>("out_55_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_7702_to_fp16 = const()[name = tensor<string, []>("op_7702_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(541445056)))];
            tensor<fp16, [2, 1280, 1, 256]> var_7703_cast = add(x = out_55_cast, y = var_7702_to_fp16)[name = tensor<string, []>("op_7703_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_7705_to_fp16 = const()[name = tensor<string, []>("op_7705_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(541447680)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_207_cast = mul(x = var_7703_cast, y = var_7705_to_fp16)[name = tensor<string, []>("hidden_states_207_cast")];
            tensor<int32, [2]> var_7712 = const()[name = tensor<string, []>("op_7712"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7714 = const()[name = tensor<string, []>("op_7714"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(541450304))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(542679168))), name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 256]> q_37_cast = conv(dilations = var_7714, groups = var_6522, pad = q_37_pad_0, pad_type = q_37_pad_type_0, strides = var_7712, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_207_cast)[name = tensor<string, []>("q_37_cast")];
            tensor<int32, [2]> var_7718 = const()[name = tensor<string, []>("op_7718"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7720 = const()[name = tensor<string, []>("op_7720"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(542679360))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(543908224))), name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 256]> k_73_cast = conv(dilations = var_7720, groups = var_6522, pad = k_73_pad_0, pad_type = k_73_pad_type_0, strides = var_7718, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_207_cast)[name = tensor<string, []>("k_73_cast")];
            tensor<int32, [2]> var_7724 = const()[name = tensor<string, []>("op_7724"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7726 = const()[name = tensor<string, []>("op_7726"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(543908416))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(545137280))), name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 256]> v_37_cast = conv(dilations = var_7726, groups = var_6522, pad = v_37_pad_0, pad_type = v_37_pad_type_0, strides = var_7724, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_207_cast)[name = tensor<string, []>("v_37_cast")];
            tensor<int32, [4]> var_7730_begin_0 = const()[name = tensor<string, []>("op_7730_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7730_end_0 = const()[name = tensor<string, []>("op_7730_end_0"), val = tensor<int32, [4]>([2, 160, 1, 256])];
            tensor<bool, [4]> var_7730_end_mask_0 = const()[name = tensor<string, []>("op_7730_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7730_cast = slice_by_index(begin = var_7730_begin_0, end = var_7730_end_0, end_mask = var_7730_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_7730_cast")];
            tensor<int32, [4]> var_7734_begin_0 = const()[name = tensor<string, []>("op_7734_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_7734_end_0 = const()[name = tensor<string, []>("op_7734_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_7734_end_mask_0 = const()[name = tensor<string, []>("op_7734_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7734_cast = slice_by_index(begin = var_7734_begin_0, end = var_7734_end_0, end_mask = var_7734_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_7734_cast")];
            tensor<int32, [4]> var_7738_begin_0 = const()[name = tensor<string, []>("op_7738_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_7738_end_0 = const()[name = tensor<string, []>("op_7738_end_0"), val = tensor<int32, [4]>([2, 480, 1, 256])];
            tensor<bool, [4]> var_7738_end_mask_0 = const()[name = tensor<string, []>("op_7738_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7738_cast = slice_by_index(begin = var_7738_begin_0, end = var_7738_end_0, end_mask = var_7738_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_7738_cast")];
            tensor<int32, [4]> var_7742_begin_0 = const()[name = tensor<string, []>("op_7742_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_7742_end_0 = const()[name = tensor<string, []>("op_7742_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_7742_end_mask_0 = const()[name = tensor<string, []>("op_7742_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7742_cast = slice_by_index(begin = var_7742_begin_0, end = var_7742_end_0, end_mask = var_7742_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_7742_cast")];
            tensor<int32, [4]> var_7746_begin_0 = const()[name = tensor<string, []>("op_7746_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_7746_end_0 = const()[name = tensor<string, []>("op_7746_end_0"), val = tensor<int32, [4]>([2, 800, 1, 256])];
            tensor<bool, [4]> var_7746_end_mask_0 = const()[name = tensor<string, []>("op_7746_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7746_cast = slice_by_index(begin = var_7746_begin_0, end = var_7746_end_0, end_mask = var_7746_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_7746_cast")];
            tensor<int32, [4]> var_7750_begin_0 = const()[name = tensor<string, []>("op_7750_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_7750_end_0 = const()[name = tensor<string, []>("op_7750_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_7750_end_mask_0 = const()[name = tensor<string, []>("op_7750_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7750_cast = slice_by_index(begin = var_7750_begin_0, end = var_7750_end_0, end_mask = var_7750_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_7750_cast")];
            tensor<int32, [4]> var_7754_begin_0 = const()[name = tensor<string, []>("op_7754_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_7754_end_0 = const()[name = tensor<string, []>("op_7754_end_0"), val = tensor<int32, [4]>([2, 1120, 1, 256])];
            tensor<bool, [4]> var_7754_end_mask_0 = const()[name = tensor<string, []>("op_7754_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7754_cast = slice_by_index(begin = var_7754_begin_0, end = var_7754_end_0, end_mask = var_7754_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_7754_cast")];
            tensor<int32, [4]> var_7758_begin_0 = const()[name = tensor<string, []>("op_7758_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_7758_end_0 = const()[name = tensor<string, []>("op_7758_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_7758_end_mask_0 = const()[name = tensor<string, []>("op_7758_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7758_cast = slice_by_index(begin = var_7758_begin_0, end = var_7758_end_0, end_mask = var_7758_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_7758_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_7765_begin_0 = const()[name = tensor<string, []>("op_7765_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7765_end_0 = const()[name = tensor<string, []>("op_7765_end_0"), val = tensor<int32, [4]>([2, 256, 1, 160])];
            tensor<bool, [4]> var_7765_end_mask_0 = const()[name = tensor<string, []>("op_7765_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)[name = tensor<string, []>("transpose_13")];
            tensor<fp16, [2, 256, 1, 160]> var_7765_cast = slice_by_index(begin = var_7765_begin_0, end = var_7765_end_0, end_mask = var_7765_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_7765_cast")];
            tensor<int32, [4]> var_7769_begin_0 = const()[name = tensor<string, []>("op_7769_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_7769_end_0 = const()[name = tensor<string, []>("op_7769_end_0"), val = tensor<int32, [4]>([2, 256, 1, 320])];
            tensor<bool, [4]> var_7769_end_mask_0 = const()[name = tensor<string, []>("op_7769_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_7769_cast = slice_by_index(begin = var_7769_begin_0, end = var_7769_end_0, end_mask = var_7769_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_7769_cast")];
            tensor<int32, [4]> var_7773_begin_0 = const()[name = tensor<string, []>("op_7773_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_7773_end_0 = const()[name = tensor<string, []>("op_7773_end_0"), val = tensor<int32, [4]>([2, 256, 1, 480])];
            tensor<bool, [4]> var_7773_end_mask_0 = const()[name = tensor<string, []>("op_7773_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_7773_cast = slice_by_index(begin = var_7773_begin_0, end = var_7773_end_0, end_mask = var_7773_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_7773_cast")];
            tensor<int32, [4]> var_7777_begin_0 = const()[name = tensor<string, []>("op_7777_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 480])];
            tensor<int32, [4]> var_7777_end_0 = const()[name = tensor<string, []>("op_7777_end_0"), val = tensor<int32, [4]>([2, 256, 1, 640])];
            tensor<bool, [4]> var_7777_end_mask_0 = const()[name = tensor<string, []>("op_7777_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_7777_cast = slice_by_index(begin = var_7777_begin_0, end = var_7777_end_0, end_mask = var_7777_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_7777_cast")];
            tensor<int32, [4]> var_7781_begin_0 = const()[name = tensor<string, []>("op_7781_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_7781_end_0 = const()[name = tensor<string, []>("op_7781_end_0"), val = tensor<int32, [4]>([2, 256, 1, 800])];
            tensor<bool, [4]> var_7781_end_mask_0 = const()[name = tensor<string, []>("op_7781_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_7781_cast = slice_by_index(begin = var_7781_begin_0, end = var_7781_end_0, end_mask = var_7781_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_7781_cast")];
            tensor<int32, [4]> var_7785_begin_0 = const()[name = tensor<string, []>("op_7785_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 800])];
            tensor<int32, [4]> var_7785_end_0 = const()[name = tensor<string, []>("op_7785_end_0"), val = tensor<int32, [4]>([2, 256, 1, 960])];
            tensor<bool, [4]> var_7785_end_mask_0 = const()[name = tensor<string, []>("op_7785_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_7785_cast = slice_by_index(begin = var_7785_begin_0, end = var_7785_end_0, end_mask = var_7785_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_7785_cast")];
            tensor<int32, [4]> var_7789_begin_0 = const()[name = tensor<string, []>("op_7789_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_7789_end_0 = const()[name = tensor<string, []>("op_7789_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1120])];
            tensor<bool, [4]> var_7789_end_mask_0 = const()[name = tensor<string, []>("op_7789_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_7789_cast = slice_by_index(begin = var_7789_begin_0, end = var_7789_end_0, end_mask = var_7789_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_7789_cast")];
            tensor<int32, [4]> var_7793_begin_0 = const()[name = tensor<string, []>("op_7793_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1120])];
            tensor<int32, [4]> var_7793_end_0 = const()[name = tensor<string, []>("op_7793_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1280])];
            tensor<bool, [4]> var_7793_end_mask_0 = const()[name = tensor<string, []>("op_7793_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 160]> var_7793_cast = slice_by_index(begin = var_7793_begin_0, end = var_7793_end_0, end_mask = var_7793_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_7793_cast")];
            tensor<int32, [4]> var_7795_begin_0 = const()[name = tensor<string, []>("op_7795_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7795_end_0 = const()[name = tensor<string, []>("op_7795_end_0"), val = tensor<int32, [4]>([2, 160, 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, 160, 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 = v_37_cast)[name = tensor<string, []>("op_7795_cast")];
            tensor<int32, [4]> var_7799_begin_0 = const()[name = tensor<string, []>("op_7799_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_7799_end_0 = const()[name = tensor<string, []>("op_7799_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_7799_end_mask_0 = const()[name = tensor<string, []>("op_7799_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7799_cast = slice_by_index(begin = var_7799_begin_0, end = var_7799_end_0, end_mask = var_7799_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_7799_cast")];
            tensor<int32, [4]> var_7803_begin_0 = const()[name = tensor<string, []>("op_7803_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_7803_end_0 = const()[name = tensor<string, []>("op_7803_end_0"), val = tensor<int32, [4]>([2, 480, 1, 256])];
            tensor<bool, [4]> var_7803_end_mask_0 = const()[name = tensor<string, []>("op_7803_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7803_cast = slice_by_index(begin = var_7803_begin_0, end = var_7803_end_0, end_mask = var_7803_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_7803_cast")];
            tensor<int32, [4]> var_7807_begin_0 = const()[name = tensor<string, []>("op_7807_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_7807_end_0 = const()[name = tensor<string, []>("op_7807_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_7807_end_mask_0 = const()[name = tensor<string, []>("op_7807_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7807_cast = slice_by_index(begin = var_7807_begin_0, end = var_7807_end_0, end_mask = var_7807_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_7807_cast")];
            tensor<int32, [4]> var_7811_begin_0 = const()[name = tensor<string, []>("op_7811_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_7811_end_0 = const()[name = tensor<string, []>("op_7811_end_0"), val = tensor<int32, [4]>([2, 800, 1, 256])];
            tensor<bool, [4]> var_7811_end_mask_0 = const()[name = tensor<string, []>("op_7811_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7811_cast = slice_by_index(begin = var_7811_begin_0, end = var_7811_end_0, end_mask = var_7811_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_7811_cast")];
            tensor<int32, [4]> var_7815_begin_0 = const()[name = tensor<string, []>("op_7815_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_7815_end_0 = const()[name = tensor<string, []>("op_7815_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_7815_end_mask_0 = const()[name = tensor<string, []>("op_7815_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7815_cast = slice_by_index(begin = var_7815_begin_0, end = var_7815_end_0, end_mask = var_7815_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_7815_cast")];
            tensor<int32, [4]> var_7819_begin_0 = const()[name = tensor<string, []>("op_7819_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_7819_end_0 = const()[name = tensor<string, []>("op_7819_end_0"), val = tensor<int32, [4]>([2, 1120, 1, 256])];
            tensor<bool, [4]> var_7819_end_mask_0 = const()[name = tensor<string, []>("op_7819_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7819_cast = slice_by_index(begin = var_7819_begin_0, end = var_7819_end_0, end_mask = var_7819_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_7819_cast")];
            tensor<int32, [4]> var_7823_begin_0 = const()[name = tensor<string, []>("op_7823_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_7823_end_0 = const()[name = tensor<string, []>("op_7823_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_7823_end_mask_0 = const()[name = tensor<string, []>("op_7823_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7823_cast = slice_by_index(begin = var_7823_begin_0, end = var_7823_end_0, end_mask = var_7823_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_7823_cast")];
            tensor<string, []> var_7827_equation_0 = const()[name = tensor<string, []>("op_7827_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7827_cast = einsum(equation = var_7827_equation_0, values = (var_7765_cast, var_7730_cast))[name = tensor<string, []>("op_7827_cast")];
            tensor<fp16, []> var_7828_to_fp16 = const()[name = tensor<string, []>("op_7828_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_161_cast = mul(x = var_7827_cast, y = var_7828_to_fp16)[name = tensor<string, []>("aw_161_cast")];
            tensor<string, []> var_7831_equation_0 = const()[name = tensor<string, []>("op_7831_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7831_cast = einsum(equation = var_7831_equation_0, values = (var_7769_cast, var_7734_cast))[name = tensor<string, []>("op_7831_cast")];
            tensor<fp16, []> var_7832_to_fp16 = const()[name = tensor<string, []>("op_7832_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_163_cast = mul(x = var_7831_cast, y = var_7832_to_fp16)[name = tensor<string, []>("aw_163_cast")];
            tensor<string, []> var_7835_equation_0 = const()[name = tensor<string, []>("op_7835_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7835_cast = einsum(equation = var_7835_equation_0, values = (var_7773_cast, var_7738_cast))[name = tensor<string, []>("op_7835_cast")];
            tensor<fp16, []> var_7836_to_fp16 = const()[name = tensor<string, []>("op_7836_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_165_cast = mul(x = var_7835_cast, y = var_7836_to_fp16)[name = tensor<string, []>("aw_165_cast")];
            tensor<string, []> var_7839_equation_0 = const()[name = tensor<string, []>("op_7839_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7839_cast = einsum(equation = var_7839_equation_0, values = (var_7777_cast, var_7742_cast))[name = tensor<string, []>("op_7839_cast")];
            tensor<fp16, []> var_7840_to_fp16 = const()[name = tensor<string, []>("op_7840_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_167_cast = mul(x = var_7839_cast, y = var_7840_to_fp16)[name = tensor<string, []>("aw_167_cast")];
            tensor<string, []> var_7843_equation_0 = const()[name = tensor<string, []>("op_7843_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7843_cast = einsum(equation = var_7843_equation_0, values = (var_7781_cast, var_7746_cast))[name = tensor<string, []>("op_7843_cast")];
            tensor<fp16, []> var_7844_to_fp16 = const()[name = tensor<string, []>("op_7844_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_169_cast = mul(x = var_7843_cast, y = var_7844_to_fp16)[name = tensor<string, []>("aw_169_cast")];
            tensor<string, []> var_7847_equation_0 = const()[name = tensor<string, []>("op_7847_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7847_cast = einsum(equation = var_7847_equation_0, values = (var_7785_cast, var_7750_cast))[name = tensor<string, []>("op_7847_cast")];
            tensor<fp16, []> var_7848_to_fp16 = const()[name = tensor<string, []>("op_7848_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_171_cast = mul(x = var_7847_cast, y = var_7848_to_fp16)[name = tensor<string, []>("aw_171_cast")];
            tensor<string, []> var_7851_equation_0 = const()[name = tensor<string, []>("op_7851_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7851_cast = einsum(equation = var_7851_equation_0, values = (var_7789_cast, var_7754_cast))[name = tensor<string, []>("op_7851_cast")];
            tensor<fp16, []> var_7852_to_fp16 = const()[name = tensor<string, []>("op_7852_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_173_cast = mul(x = var_7851_cast, y = var_7852_to_fp16)[name = tensor<string, []>("aw_173_cast")];
            tensor<string, []> var_7855_equation_0 = const()[name = tensor<string, []>("op_7855_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7855_cast = einsum(equation = var_7855_equation_0, values = (var_7793_cast, var_7758_cast))[name = tensor<string, []>("op_7855_cast")];
            tensor<fp16, []> var_7856_to_fp16 = const()[name = tensor<string, []>("op_7856_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 256, 1, 256]> aw_175_cast = mul(x = var_7855_cast, y = var_7856_to_fp16)[name = tensor<string, []>("aw_175_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7858_cast = softmax(axis = var_6522, x = aw_161_cast)[name = tensor<string, []>("op_7858_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7859_cast = softmax(axis = var_6522, x = aw_163_cast)[name = tensor<string, []>("op_7859_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7860_cast = softmax(axis = var_6522, x = aw_165_cast)[name = tensor<string, []>("op_7860_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7861_cast = softmax(axis = var_6522, x = aw_167_cast)[name = tensor<string, []>("op_7861_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7862_cast = softmax(axis = var_6522, x = aw_169_cast)[name = tensor<string, []>("op_7862_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7863_cast = softmax(axis = var_6522, x = aw_171_cast)[name = tensor<string, []>("op_7863_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7864_cast = softmax(axis = var_6522, x = aw_173_cast)[name = tensor<string, []>("op_7864_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7865_cast = softmax(axis = var_6522, x = aw_175_cast)[name = tensor<string, []>("op_7865_cast")];
            tensor<string, []> var_7867_equation_0 = const()[name = tensor<string, []>("op_7867_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7867_cast = einsum(equation = var_7867_equation_0, values = (var_7795_cast, var_7858_cast))[name = tensor<string, []>("op_7867_cast")];
            tensor<string, []> var_7869_equation_0 = const()[name = tensor<string, []>("op_7869_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7869_cast = einsum(equation = var_7869_equation_0, values = (var_7799_cast, var_7859_cast))[name = tensor<string, []>("op_7869_cast")];
            tensor<string, []> var_7871_equation_0 = const()[name = tensor<string, []>("op_7871_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7871_cast = einsum(equation = var_7871_equation_0, values = (var_7803_cast, var_7860_cast))[name = tensor<string, []>("op_7871_cast")];
            tensor<string, []> var_7873_equation_0 = const()[name = tensor<string, []>("op_7873_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7873_cast = einsum(equation = var_7873_equation_0, values = (var_7807_cast, var_7861_cast))[name = tensor<string, []>("op_7873_cast")];
            tensor<string, []> var_7875_equation_0 = const()[name = tensor<string, []>("op_7875_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7875_cast = einsum(equation = var_7875_equation_0, values = (var_7811_cast, var_7862_cast))[name = tensor<string, []>("op_7875_cast")];
            tensor<string, []> var_7877_equation_0 = const()[name = tensor<string, []>("op_7877_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7877_cast = einsum(equation = var_7877_equation_0, values = (var_7815_cast, var_7863_cast))[name = tensor<string, []>("op_7877_cast")];
            tensor<string, []> var_7879_equation_0 = const()[name = tensor<string, []>("op_7879_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7879_cast = einsum(equation = var_7879_equation_0, values = (var_7819_cast, var_7864_cast))[name = tensor<string, []>("op_7879_cast")];
            tensor<string, []> var_7881_equation_0 = const()[name = tensor<string, []>("op_7881_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_7881_cast = einsum(equation = var_7881_equation_0, values = (var_7823_cast, var_7865_cast))[name = tensor<string, []>("op_7881_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_6522, interleave = input_353_interleave_0, values = (var_7867_cast, var_7869_cast, var_7871_cast, var_7873_cast, var_7875_cast, var_7877_cast, var_7879_cast, var_7881_cast))[name = tensor<string, []>("input_353_cast")];
            tensor<int32, [2]> var_7887 = const()[name = tensor<string, []>("op_7887"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7889 = const()[name = tensor<string, []>("op_7889"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_7891_pad_type_0 = const()[name = tensor<string, []>("op_7891_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_7891_pad_0 = const()[name = tensor<string, []>("op_7891_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(545137472))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(546366336))), name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(546366528)))];
            tensor<fp16, [2, 1280, 1, 256]> var_7891_cast = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_7889, groups = var_6522, pad = var_7891_pad_0, pad_type = var_7891_pad_type_0, strides = var_7887, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_353_cast)[name = tensor<string, []>("op_7891_cast")];
            tensor<fp16, [2, 1280, 1, 256]> inputs_57_cast = add(x = var_7891_cast, y = inputs_55_cast)[name = tensor<string, []>("inputs_57_cast")];
            tensor<int32, [1]> var_7895 = const()[name = tensor<string, []>("op_7895"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_57_cast = reduce_mean(axes = var_7895, keep_dims = var_6517, x = inputs_57_cast)[name = tensor<string, []>("channels_mean_57_cast")];
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_57_cast = sub(x = inputs_57_cast, y = channels_mean_57_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_57_cast")];
            tensor<int32, [1]> var_7899 = const()[name = tensor<string, []>("op_7899"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_7900_cast = reduce_mean(axes = var_7899, keep_dims = var_6517, x = zero_mean_sq_57_cast)[name = tensor<string, []>("op_7900_cast")];
            tensor<fp16, []> var_7901_to_fp16 = const()[name = tensor<string, []>("op_7901_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_7902_cast = add(x = var_7900_cast, y = var_7901_to_fp16)[name = tensor<string, []>("op_7902_cast")];
            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_7902_cast)[name = tensor<string, []>("denom_57_cast")];
            tensor<fp16, [2, 1280, 1, 256]> out_57_cast = mul(x = zero_mean_57_cast, y = denom_57_cast)[name = tensor<string, []>("out_57_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_7906_to_fp16 = const()[name = tensor<string, []>("op_7906_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(546369152)))];
            tensor<fp16, [2, 1280, 1, 256]> var_7907_cast = add(x = out_57_cast, y = var_7906_to_fp16)[name = tensor<string, []>("op_7907_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_7909_to_fp16 = const()[name = tensor<string, []>("op_7909_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(546371776)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_209_cast = mul(x = var_7907_cast, y = var_7909_to_fp16)[name = tensor<string, []>("hidden_states_209_cast")];
            tensor<int32, [2]> var_7916 = const()[name = tensor<string, []>("op_7916"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7918 = const()[name = tensor<string, []>("op_7918"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(546374400))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(547603264))), name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            tensor<fp16, [2, 1280, 1, 256]> q_39_cast = conv(dilations = var_7918, groups = var_6522, pad = q_39_pad_0, pad_type = q_39_pad_type_0, strides = var_7916, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_209_cast)[name = tensor<string, []>("q_39_cast")];
            tensor<int32, [2]> var_7922 = const()[name = tensor<string, []>("op_7922"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7924 = const()[name = tensor<string, []>("op_7924"), 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, 768, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [737280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(547603456))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(548340800))), name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 768, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_77_cast = conv(dilations = var_7924, groups = var_6522, pad = k_77_pad_0, pad_type = k_77_pad_type_0, strides = var_7922, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_77_cast")];
            tensor<int32, [2]> var_7928 = const()[name = tensor<string, []>("op_7928"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7930 = const()[name = tensor<string, []>("op_7930"), 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, 768, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [737280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(548340992))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(549078336))), name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 768, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_39_cast = conv(dilations = var_7930, groups = var_6522, pad = v_39_pad_0, pad_type = v_39_pad_type_0, strides = var_7928, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_39_cast")];
            tensor<int32, [4]> var_7934_begin_0 = const()[name = tensor<string, []>("op_7934_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7934_end_0 = const()[name = tensor<string, []>("op_7934_end_0"), val = tensor<int32, [4]>([2, 160, 1, 256])];
            tensor<bool, [4]> var_7934_end_mask_0 = const()[name = tensor<string, []>("op_7934_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7934_cast = slice_by_index(begin = var_7934_begin_0, end = var_7934_end_0, end_mask = var_7934_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_7934_cast")];
            tensor<int32, [4]> var_7938_begin_0 = const()[name = tensor<string, []>("op_7938_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_7938_end_0 = const()[name = tensor<string, []>("op_7938_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_7938_end_mask_0 = const()[name = tensor<string, []>("op_7938_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7938_cast = slice_by_index(begin = var_7938_begin_0, end = var_7938_end_0, end_mask = var_7938_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_7938_cast")];
            tensor<int32, [4]> var_7942_begin_0 = const()[name = tensor<string, []>("op_7942_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_7942_end_0 = const()[name = tensor<string, []>("op_7942_end_0"), val = tensor<int32, [4]>([2, 480, 1, 256])];
            tensor<bool, [4]> var_7942_end_mask_0 = const()[name = tensor<string, []>("op_7942_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7942_cast = slice_by_index(begin = var_7942_begin_0, end = var_7942_end_0, end_mask = var_7942_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_7942_cast")];
            tensor<int32, [4]> var_7946_begin_0 = const()[name = tensor<string, []>("op_7946_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_7946_end_0 = const()[name = tensor<string, []>("op_7946_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_7946_end_mask_0 = const()[name = tensor<string, []>("op_7946_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7946_cast = slice_by_index(begin = var_7946_begin_0, end = var_7946_end_0, end_mask = var_7946_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_7946_cast")];
            tensor<int32, [4]> var_7950_begin_0 = const()[name = tensor<string, []>("op_7950_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_7950_end_0 = const()[name = tensor<string, []>("op_7950_end_0"), val = tensor<int32, [4]>([2, 800, 1, 256])];
            tensor<bool, [4]> var_7950_end_mask_0 = const()[name = tensor<string, []>("op_7950_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7950_cast = slice_by_index(begin = var_7950_begin_0, end = var_7950_end_0, end_mask = var_7950_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_7950_cast")];
            tensor<int32, [4]> var_7954_begin_0 = const()[name = tensor<string, []>("op_7954_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_7954_end_0 = const()[name = tensor<string, []>("op_7954_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_7954_end_mask_0 = const()[name = tensor<string, []>("op_7954_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7954_cast = slice_by_index(begin = var_7954_begin_0, end = var_7954_end_0, end_mask = var_7954_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_7954_cast")];
            tensor<int32, [4]> var_7958_begin_0 = const()[name = tensor<string, []>("op_7958_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_7958_end_0 = const()[name = tensor<string, []>("op_7958_end_0"), val = tensor<int32, [4]>([2, 1120, 1, 256])];
            tensor<bool, [4]> var_7958_end_mask_0 = const()[name = tensor<string, []>("op_7958_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7958_cast = slice_by_index(begin = var_7958_begin_0, end = var_7958_end_0, end_mask = var_7958_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_7958_cast")];
            tensor<int32, [4]> var_7962_begin_0 = const()[name = tensor<string, []>("op_7962_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_7962_end_0 = const()[name = tensor<string, []>("op_7962_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_7962_end_mask_0 = const()[name = tensor<string, []>("op_7962_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 256]> var_7962_cast = slice_by_index(begin = var_7962_begin_0, end = var_7962_end_0, end_mask = var_7962_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_7962_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_7969_begin_0 = const()[name = tensor<string, []>("op_7969_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7969_end_0 = const()[name = tensor<string, []>("op_7969_end_0"), val = tensor<int32, [4]>([2, 77, 1, 160])];
            tensor<bool, [4]> var_7969_end_mask_0 = const()[name = tensor<string, []>("op_7969_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)[name = tensor<string, []>("transpose_12")];
            tensor<fp16, [2, 77, 1, 160]> var_7969_cast = slice_by_index(begin = var_7969_begin_0, end = var_7969_end_0, end_mask = var_7969_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_7969_cast")];
            tensor<int32, [4]> var_7973_begin_0 = const()[name = tensor<string, []>("op_7973_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_7973_end_0 = const()[name = tensor<string, []>("op_7973_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_7973_end_mask_0 = const()[name = tensor<string, []>("op_7973_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_7973_cast = slice_by_index(begin = var_7973_begin_0, end = var_7973_end_0, end_mask = var_7973_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_7973_cast")];
            tensor<int32, [4]> var_7977_begin_0 = const()[name = tensor<string, []>("op_7977_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_7977_end_0 = const()[name = tensor<string, []>("op_7977_end_0"), val = tensor<int32, [4]>([2, 77, 1, 480])];
            tensor<bool, [4]> var_7977_end_mask_0 = const()[name = tensor<string, []>("op_7977_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_7977_cast = slice_by_index(begin = var_7977_begin_0, end = var_7977_end_0, end_mask = var_7977_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_7977_cast")];
            tensor<int32, [4]> var_7981_begin_0 = const()[name = tensor<string, []>("op_7981_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 480])];
            tensor<int32, [4]> var_7981_end_0 = const()[name = tensor<string, []>("op_7981_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_7981_end_mask_0 = const()[name = tensor<string, []>("op_7981_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_7981_cast = slice_by_index(begin = var_7981_begin_0, end = var_7981_end_0, end_mask = var_7981_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_7981_cast")];
            tensor<int32, [4]> var_7985_begin_0 = const()[name = tensor<string, []>("op_7985_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_7985_end_0 = const()[name = tensor<string, []>("op_7985_end_0"), val = tensor<int32, [4]>([2, 77, 1, 800])];
            tensor<bool, [4]> var_7985_end_mask_0 = const()[name = tensor<string, []>("op_7985_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_7985_cast = slice_by_index(begin = var_7985_begin_0, end = var_7985_end_0, end_mask = var_7985_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_7985_cast")];
            tensor<int32, [4]> var_7989_begin_0 = const()[name = tensor<string, []>("op_7989_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 800])];
            tensor<int32, [4]> var_7989_end_0 = const()[name = tensor<string, []>("op_7989_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_7989_end_mask_0 = const()[name = tensor<string, []>("op_7989_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_7989_cast = slice_by_index(begin = var_7989_begin_0, end = var_7989_end_0, end_mask = var_7989_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_7989_cast")];
            tensor<int32, [4]> var_7993_begin_0 = const()[name = tensor<string, []>("op_7993_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_7993_end_0 = const()[name = tensor<string, []>("op_7993_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1120])];
            tensor<bool, [4]> var_7993_end_mask_0 = const()[name = tensor<string, []>("op_7993_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_7993_cast = slice_by_index(begin = var_7993_begin_0, end = var_7993_end_0, end_mask = var_7993_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_7993_cast")];
            tensor<int32, [4]> var_7997_begin_0 = const()[name = tensor<string, []>("op_7997_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1120])];
            tensor<int32, [4]> var_7997_end_0 = const()[name = tensor<string, []>("op_7997_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_7997_end_mask_0 = const()[name = tensor<string, []>("op_7997_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 160]> var_7997_cast = slice_by_index(begin = var_7997_begin_0, end = var_7997_end_0, end_mask = var_7997_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_7997_cast")];
            tensor<int32, [4]> var_7999_begin_0 = const()[name = tensor<string, []>("op_7999_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7999_end_0 = const()[name = tensor<string, []>("op_7999_end_0"), val = tensor<int32, [4]>([2, 160, 1, 77])];
            tensor<bool, [4]> var_7999_end_mask_0 = const()[name = tensor<string, []>("op_7999_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_7999_cast = slice_by_index(begin = var_7999_begin_0, end = var_7999_end_0, end_mask = var_7999_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_7999_cast")];
            tensor<int32, [4]> var_8003_begin_0 = const()[name = tensor<string, []>("op_8003_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_8003_end_0 = const()[name = tensor<string, []>("op_8003_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_8003_end_mask_0 = const()[name = tensor<string, []>("op_8003_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_8003_cast = slice_by_index(begin = var_8003_begin_0, end = var_8003_end_0, end_mask = var_8003_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_8003_cast")];
            tensor<int32, [4]> var_8007_begin_0 = const()[name = tensor<string, []>("op_8007_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_8007_end_0 = const()[name = tensor<string, []>("op_8007_end_0"), val = tensor<int32, [4]>([2, 480, 1, 77])];
            tensor<bool, [4]> var_8007_end_mask_0 = const()[name = tensor<string, []>("op_8007_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_8007_cast = slice_by_index(begin = var_8007_begin_0, end = var_8007_end_0, end_mask = var_8007_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_8007_cast")];
            tensor<int32, [4]> var_8011_begin_0 = const()[name = tensor<string, []>("op_8011_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_8011_end_0 = const()[name = tensor<string, []>("op_8011_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_8011_end_mask_0 = const()[name = tensor<string, []>("op_8011_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_8011_cast = slice_by_index(begin = var_8011_begin_0, end = var_8011_end_0, end_mask = var_8011_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_8011_cast")];
            tensor<int32, [4]> var_8015_begin_0 = const()[name = tensor<string, []>("op_8015_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_8015_end_0 = const()[name = tensor<string, []>("op_8015_end_0"), val = tensor<int32, [4]>([2, 800, 1, 77])];
            tensor<bool, [4]> var_8015_end_mask_0 = const()[name = tensor<string, []>("op_8015_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_8015_cast = slice_by_index(begin = var_8015_begin_0, end = var_8015_end_0, end_mask = var_8015_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_8015_cast")];
            tensor<int32, [4]> var_8019_begin_0 = const()[name = tensor<string, []>("op_8019_begin_0"), val = tensor<int32, [4]>([0, 800, 0, 0])];
            tensor<int32, [4]> var_8019_end_0 = const()[name = tensor<string, []>("op_8019_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_8019_end_mask_0 = const()[name = tensor<string, []>("op_8019_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_8019_cast = slice_by_index(begin = var_8019_begin_0, end = var_8019_end_0, end_mask = var_8019_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_8019_cast")];
            tensor<int32, [4]> var_8023_begin_0 = const()[name = tensor<string, []>("op_8023_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_8023_end_0 = const()[name = tensor<string, []>("op_8023_end_0"), val = tensor<int32, [4]>([2, 1120, 1, 77])];
            tensor<bool, [4]> var_8023_end_mask_0 = const()[name = tensor<string, []>("op_8023_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_8023_cast = slice_by_index(begin = var_8023_begin_0, end = var_8023_end_0, end_mask = var_8023_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_8023_cast")];
            tensor<int32, [4]> var_8027_begin_0 = const()[name = tensor<string, []>("op_8027_begin_0"), val = tensor<int32, [4]>([0, 1120, 0, 0])];
            tensor<int32, [4]> var_8027_end_0 = const()[name = tensor<string, []>("op_8027_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_8027_end_mask_0 = const()[name = tensor<string, []>("op_8027_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 160, 1, 77]> var_8027_cast = slice_by_index(begin = var_8027_begin_0, end = var_8027_end_0, end_mask = var_8027_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_8027_cast")];
            tensor<string, []> var_8031_equation_0 = const()[name = tensor<string, []>("op_8031_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8031_cast = einsum(equation = var_8031_equation_0, values = (var_7969_cast, var_7934_cast))[name = tensor<string, []>("op_8031_cast")];
            tensor<fp16, []> var_8032_to_fp16 = const()[name = tensor<string, []>("op_8032_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_177_cast = mul(x = var_8031_cast, y = var_8032_to_fp16)[name = tensor<string, []>("aw_177_cast")];
            tensor<string, []> var_8035_equation_0 = const()[name = tensor<string, []>("op_8035_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8035_cast = einsum(equation = var_8035_equation_0, values = (var_7973_cast, var_7938_cast))[name = tensor<string, []>("op_8035_cast")];
            tensor<fp16, []> var_8036_to_fp16 = const()[name = tensor<string, []>("op_8036_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_179_cast = mul(x = var_8035_cast, y = var_8036_to_fp16)[name = tensor<string, []>("aw_179_cast")];
            tensor<string, []> var_8039_equation_0 = const()[name = tensor<string, []>("op_8039_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8039_cast = einsum(equation = var_8039_equation_0, values = (var_7977_cast, var_7942_cast))[name = tensor<string, []>("op_8039_cast")];
            tensor<fp16, []> var_8040_to_fp16 = const()[name = tensor<string, []>("op_8040_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_181_cast = mul(x = var_8039_cast, y = var_8040_to_fp16)[name = tensor<string, []>("aw_181_cast")];
            tensor<string, []> var_8043_equation_0 = const()[name = tensor<string, []>("op_8043_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8043_cast = einsum(equation = var_8043_equation_0, values = (var_7981_cast, var_7946_cast))[name = tensor<string, []>("op_8043_cast")];
            tensor<fp16, []> var_8044_to_fp16 = const()[name = tensor<string, []>("op_8044_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_183_cast = mul(x = var_8043_cast, y = var_8044_to_fp16)[name = tensor<string, []>("aw_183_cast")];
            tensor<string, []> var_8047_equation_0 = const()[name = tensor<string, []>("op_8047_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8047_cast = einsum(equation = var_8047_equation_0, values = (var_7985_cast, var_7950_cast))[name = tensor<string, []>("op_8047_cast")];
            tensor<fp16, []> var_8048_to_fp16 = const()[name = tensor<string, []>("op_8048_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_185_cast = mul(x = var_8047_cast, y = var_8048_to_fp16)[name = tensor<string, []>("aw_185_cast")];
            tensor<string, []> var_8051_equation_0 = const()[name = tensor<string, []>("op_8051_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8051_cast = einsum(equation = var_8051_equation_0, values = (var_7989_cast, var_7954_cast))[name = tensor<string, []>("op_8051_cast")];
            tensor<fp16, []> var_8052_to_fp16 = const()[name = tensor<string, []>("op_8052_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_187_cast = mul(x = var_8051_cast, y = var_8052_to_fp16)[name = tensor<string, []>("aw_187_cast")];
            tensor<string, []> var_8055_equation_0 = const()[name = tensor<string, []>("op_8055_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8055_cast = einsum(equation = var_8055_equation_0, values = (var_7993_cast, var_7958_cast))[name = tensor<string, []>("op_8055_cast")];
            tensor<fp16, []> var_8056_to_fp16 = const()[name = tensor<string, []>("op_8056_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_189_cast = mul(x = var_8055_cast, y = var_8056_to_fp16)[name = tensor<string, []>("aw_189_cast")];
            tensor<string, []> var_8059_equation_0 = const()[name = tensor<string, []>("op_8059_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8059_cast = einsum(equation = var_8059_equation_0, values = (var_7997_cast, var_7962_cast))[name = tensor<string, []>("op_8059_cast")];
            tensor<fp16, []> var_8060_to_fp16 = const()[name = tensor<string, []>("op_8060_to_fp16"), val = tensor<fp16, []>(0x1.43cp-4)];
            tensor<fp16, [2, 77, 1, 256]> aw_cast = mul(x = var_8059_cast, y = var_8060_to_fp16)[name = tensor<string, []>("aw_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_8062_cast = softmax(axis = var_6522, x = aw_177_cast)[name = tensor<string, []>("op_8062_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_8063_cast = softmax(axis = var_6522, x = aw_179_cast)[name = tensor<string, []>("op_8063_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_8064_cast = softmax(axis = var_6522, x = aw_181_cast)[name = tensor<string, []>("op_8064_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_8065_cast = softmax(axis = var_6522, x = aw_183_cast)[name = tensor<string, []>("op_8065_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_8066_cast = softmax(axis = var_6522, x = aw_185_cast)[name = tensor<string, []>("op_8066_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_8067_cast = softmax(axis = var_6522, x = aw_187_cast)[name = tensor<string, []>("op_8067_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_8068_cast = softmax(axis = var_6522, x = aw_189_cast)[name = tensor<string, []>("op_8068_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_8069_cast = softmax(axis = var_6522, x = aw_cast)[name = tensor<string, []>("op_8069_cast")];
            tensor<string, []> var_8071_equation_0 = const()[name = tensor<string, []>("op_8071_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_8071_cast = einsum(equation = var_8071_equation_0, values = (var_7999_cast, var_8062_cast))[name = tensor<string, []>("op_8071_cast")];
            tensor<string, []> var_8073_equation_0 = const()[name = tensor<string, []>("op_8073_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_8073_cast = einsum(equation = var_8073_equation_0, values = (var_8003_cast, var_8063_cast))[name = tensor<string, []>("op_8073_cast")];
            tensor<string, []> var_8075_equation_0 = const()[name = tensor<string, []>("op_8075_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_8075_cast = einsum(equation = var_8075_equation_0, values = (var_8007_cast, var_8064_cast))[name = tensor<string, []>("op_8075_cast")];
            tensor<string, []> var_8077_equation_0 = const()[name = tensor<string, []>("op_8077_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_8077_cast = einsum(equation = var_8077_equation_0, values = (var_8011_cast, var_8065_cast))[name = tensor<string, []>("op_8077_cast")];
            tensor<string, []> var_8079_equation_0 = const()[name = tensor<string, []>("op_8079_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_8079_cast = einsum(equation = var_8079_equation_0, values = (var_8015_cast, var_8066_cast))[name = tensor<string, []>("op_8079_cast")];
            tensor<string, []> var_8081_equation_0 = const()[name = tensor<string, []>("op_8081_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_8081_cast = einsum(equation = var_8081_equation_0, values = (var_8019_cast, var_8067_cast))[name = tensor<string, []>("op_8081_cast")];
            tensor<string, []> var_8083_equation_0 = const()[name = tensor<string, []>("op_8083_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_8083_cast = einsum(equation = var_8083_equation_0, values = (var_8023_cast, var_8068_cast))[name = tensor<string, []>("op_8083_cast")];
            tensor<string, []> var_8085_equation_0 = const()[name = tensor<string, []>("op_8085_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 160, 1, 256]> var_8085_cast = einsum(equation = var_8085_equation_0, values = (var_8027_cast, var_8069_cast))[name = tensor<string, []>("op_8085_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_6522, interleave = input_355_interleave_0, values = (var_8071_cast, var_8073_cast, var_8075_cast, var_8077_cast, var_8079_cast, var_8081_cast, var_8083_cast, var_8085_cast))[name = tensor<string, []>("input_355_cast")];
            tensor<int32, [2]> var_8091 = const()[name = tensor<string, []>("op_8091"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8093 = const()[name = tensor<string, []>("op_8093"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_8095_pad_type_0 = const()[name = tensor<string, []>("op_8095_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_8095_pad_0 = const()[name = tensor<string, []>("op_8095_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(549078528))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(550307392))), name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(550307584)))];
            tensor<fp16, [2, 1280, 1, 256]> var_8095_cast = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_8093, groups = var_6522, pad = var_8095_pad_0, pad_type = var_8095_pad_type_0, strides = var_8091, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_355_cast)[name = tensor<string, []>("op_8095_cast")];
            tensor<fp16, [2, 1280, 1, 256]> inputs_59_cast = add(x = var_8095_cast, y = inputs_57_cast)[name = tensor<string, []>("inputs_59_cast")];
            tensor<int32, [1]> var_8099 = const()[name = tensor<string, []>("op_8099"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_59_cast = reduce_mean(axes = var_8099, keep_dims = var_6517, x = inputs_59_cast)[name = tensor<string, []>("channels_mean_59_cast")];
            tensor<fp16, [2, 1280, 1, 256]> zero_mean_59_cast = sub(x = inputs_59_cast, y = channels_mean_59_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_59_cast")];
            tensor<int32, [1]> var_8103 = const()[name = tensor<string, []>("op_8103"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_8104_cast = reduce_mean(axes = var_8103, keep_dims = var_6517, x = zero_mean_sq_59_cast)[name = tensor<string, []>("op_8104_cast")];
            tensor<fp16, []> var_8105_to_fp16 = const()[name = tensor<string, []>("op_8105_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_8106_cast = add(x = var_8104_cast, y = var_8105_to_fp16)[name = tensor<string, []>("op_8106_cast")];
            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_8106_cast)[name = tensor<string, []>("denom_59_cast")];
            tensor<fp16, [2, 1280, 1, 256]> out_59_cast = mul(x = zero_mean_59_cast, y = denom_59_cast)[name = tensor<string, []>("out_59_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_8110_to_fp16 = const()[name = tensor<string, []>("op_8110_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(550310208)))];
            tensor<fp16, [2, 1280, 1, 256]> var_8111_cast = add(x = out_59_cast, y = var_8110_to_fp16)[name = tensor<string, []>("op_8111_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_8113_to_fp16 = const()[name = tensor<string, []>("op_8113_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(550312832)))];
            tensor<fp16, [2, 1280, 1, 256]> input_357_cast = mul(x = var_8111_cast, y = var_8113_to_fp16)[name = tensor<string, []>("input_357_cast")];
            tensor<int32, [2]> var_8121 = const()[name = tensor<string, []>("op_8121"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8123 = const()[name = tensor<string, []>("op_8123"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_8125_pad_type_0 = const()[name = tensor<string, []>("op_8125_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_8125_pad_0 = const()[name = tensor<string, []>("op_8125_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [9830400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(550315456))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(560145920))), name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([10240, 1280, 1, 1])];
            tensor<fp16, [10240]> up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [7680]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(560146112))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(560153856))), name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([10240])];
            tensor<fp16, [2, 10240, 1, 256]> var_8125_cast = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_8123, groups = var_6522, pad = var_8125_pad_0, pad_type = var_8125_pad_type_0, strides = var_8121, weight = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_357_cast)[name = tensor<string, []>("op_8125_cast")];
            tensor<int32, [2]> var_8126_split_sizes_0 = const()[name = tensor<string, []>("op_8126_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_8126_axis_0 = const()[name = tensor<string, []>("op_8126_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 256]> var_8126_cast_0, tensor<fp16, [2, 5120, 1, 256]> var_8126_cast_1 = split(axis = var_8126_axis_0, split_sizes = var_8126_split_sizes_0, x = var_8125_cast)[name = tensor<string, []>("op_8126_cast")];
            tensor<string, []> var_8128_mode_0 = const()[name = tensor<string, []>("op_8128_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 256]> var_8128_cast = gelu(mode = var_8128_mode_0, x = var_8126_cast_1)[name = tensor<string, []>("op_8128_cast")];
            tensor<fp16, [2, 5120, 1, 256]> input_359_cast = mul(x = var_8126_cast_0, y = var_8128_cast)[name = tensor<string, []>("input_359_cast")];
            tensor<int32, [2]> var_8132 = const()[name = tensor<string, []>("op_8132"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8134 = const()[name = tensor<string, []>("op_8134"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_8136_pad_type_0 = const()[name = tensor<string, []>("op_8136_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_8136_pad_0 = const()[name = tensor<string, []>("op_8136_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4915200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(560154048))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(565069312))), name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 5120, 1, 1])];
            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, []>(565069504)))];
            tensor<fp16, [2, 1280, 1, 256]> var_8136_cast = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_8134, groups = var_6522, pad = var_8136_pad_0, pad_type = var_8136_pad_type_0, strides = var_8132, weight = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_359_cast)[name = tensor<string, []>("op_8136_cast")];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_213_cast = add(x = var_8136_cast, y = inputs_59_cast)[name = tensor<string, []>("hidden_states_213_cast")];
            tensor<int32, [4]> var_8138 = const()[name = tensor<string, []>("op_8138"), val = tensor<int32, [4]>([2, 1280, 16, 16])];
            tensor<fp16, [2, 1280, 16, 16]> input_361_cast = reshape(shape = var_8138, x = hidden_states_213_cast)[name = tensor<string, []>("input_361_cast")];
            tensor<int32, [2]> var_8142 = const()[name = tensor<string, []>("op_8142"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8144 = const()[name = tensor<string, []>("op_8144"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(565072128))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(566300992))), name = tensor<string, []>("up_blocks_1_attentions_2_proj_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 1, 1])];
            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, []>(566301184)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_215_cast = conv(bias = up_blocks_1_attentions_2_proj_out_bias_to_fp16, dilations = var_8144, groups = var_6522, pad = hidden_states_215_pad_0, pad_type = hidden_states_215_pad_type_0, strides = var_8142, weight = up_blocks_1_attentions_2_proj_out_weight_to_fp16_palettized, x = input_361_cast)[name = tensor<string, []>("hidden_states_215_cast")];
            tensor<fp16, [2, 1280, 16, 16]> input_363_cast = add(x = hidden_states_215_cast, y = hidden_states_203_cast)[name = tensor<string, []>("input_363_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)[name = tensor<string, []>("input_365_cast")];
            tensor<int32, [2]> var_8153 = const()[name = tensor<string, []>("op_8153"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8155 = const()[name = tensor<string, []>("op_8155"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [11059200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(566303808))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(577363072))), name = tensor<string, []>("up_blocks_1_upsamplers_0_conv_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1280, 3, 3])];
            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, []>(577363264)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_217_cast = conv(bias = up_blocks_1_upsamplers_0_conv_bias_to_fp16, dilations = var_8155, groups = var_6522, pad = hidden_states_217_pad_0, pad_type = hidden_states_217_pad_type_0, strides = var_8153, weight = up_blocks_1_upsamplers_0_conv_weight_to_fp16_palettized, x = input_365_cast)[name = tensor<string, []>("hidden_states_217_cast")];
            tensor<int32, []> var_8160 = const()[name = tensor<string, []>("op_8160"), val = tensor<int32, []>(3)];
            tensor<bool, []> var_8177 = const()[name = tensor<string, []>("op_8177"), val = tensor<bool, []>(true)];
            tensor<int32, []> var_8182 = const()[name = tensor<string, []>("op_8182"), 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_8182, interleave = input_367_interleave_0, values = (hidden_states_217_cast, input_115_cast))[name = tensor<string, []>("input_367_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)[name = tensor<string, []>("reshape_168_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)[name = tensor<string, []>("reduce_mean_126_cast")];
            tensor<fp16, [2, 32, 60, 32, 32]> sub_84_cast = sub(x = reshape_168_cast, y = reduce_mean_126_cast)[name = tensor<string, []>("sub_84_cast")];
            tensor<fp16, [2, 32, 60, 32, 32]> square_42_cast = square(x = sub_84_cast)[name = tensor<string, []>("square_42_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)[name = tensor<string, []>("reduce_mean_128_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)[name = tensor<string, []>("add_84_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_42_cast = sqrt(x = add_84_cast)[name = tensor<string, []>("sqrt_42_cast")];
            tensor<fp16, [2, 32, 60, 32, 32]> real_div_42_cast = real_div(x = sub_84_cast, y = sqrt_42_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_169_cast")];
            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, []>(577365888)))];
            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, []>(577369792)))];
            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_79_mean_0_to_fp16, variance = add_79_variance_0_to_fp16, x = reshape_169_cast)[name = tensor<string, []>("add_85_cast")];
            tensor<fp16, [2, 1920, 32, 32]> input_371_cast = silu(x = add_85_cast)[name = tensor<string, []>("input_371_cast")];
            tensor<int32, [2]> var_8211 = const()[name = tensor<string, []>("op_8211"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8213 = const()[name = tensor<string, []>("op_8213"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [8294400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(577373696))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(585668160))), name = tensor<string, []>("up_blocks_2_resnets_0_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 1920, 3, 3])];
            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, []>(585668352)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_219_cast = conv(bias = up_blocks_2_resnets_0_conv1_bias_to_fp16, dilations = var_8213, groups = var_8182, pad = hidden_states_219_pad_0, pad_type = hidden_states_219_pad_type_0, strides = var_8211, weight = up_blocks_2_resnets_0_conv1_weight_to_fp16_palettized, x = input_371_cast)[name = tensor<string, []>("hidden_states_219_cast")];
            tensor<int32, [2]> var_8219 = const()[name = tensor<string, []>("op_8219"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8221 = const()[name = tensor<string, []>("op_8221"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [614400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(585669696))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(586284160))), name = tensor<string, []>("up_blocks_2_resnets_0_time_emb_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 1280, 1, 1])];
            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, []>(586284352)))];
            tensor<fp16, [2, 640, 1, 1]> temb_33_cast = conv(bias = up_blocks_2_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_8221, groups = var_8182, pad = temb_33_pad_0, pad_type = temb_33_pad_type_0, strides = var_8219, weight = up_blocks_2_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor<string, []>("temb_33_cast")];
            tensor<fp16, [2, 640, 32, 32]> input_375_cast = add(x = hidden_states_219_cast, y = temb_33_cast)[name = tensor<string, []>("input_375_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)[name = tensor<string, []>("reshape_172_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)[name = tensor<string, []>("reduce_mean_129_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> sub_86_cast = sub(x = reshape_172_cast, y = reduce_mean_129_cast)[name = tensor<string, []>("sub_86_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> square_43_cast = square(x = sub_86_cast)[name = tensor<string, []>("square_43_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)[name = tensor<string, []>("reduce_mean_131_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)[name = tensor<string, []>("add_86_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_43_cast = sqrt(x = add_86_cast)[name = tensor<string, []>("sqrt_43_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> real_div_43_cast = real_div(x = sub_86_cast, y = sqrt_43_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_173_cast")];
            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, []>(586285696)))];
            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, []>(586287040)))];
            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_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_173_cast)[name = tensor<string, []>("add_87_cast")];
            tensor<fp16, [2, 640, 32, 32]> input_379_cast = silu(x = add_87_cast)[name = tensor<string, []>("input_379_cast")];
            tensor<int32, [2]> var_8231 = const()[name = tensor<string, []>("op_8231"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8233 = const()[name = tensor<string, []>("op_8233"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2764800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(586288384))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(589053248))), name = tensor<string, []>("up_blocks_2_resnets_0_conv2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 3, 3])];
            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, []>(589053440)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_221_cast = conv(bias = up_blocks_2_resnets_0_conv2_bias_to_fp16, dilations = var_8233, groups = var_8182, pad = hidden_states_221_pad_0, pad_type = hidden_states_221_pad_type_0, strides = var_8231, weight = up_blocks_2_resnets_0_conv2_weight_to_fp16_palettized, x = input_379_cast)[name = tensor<string, []>("hidden_states_221_cast")];
            tensor<int32, [2]> var_8238 = const()[name = tensor<string, []>("op_8238"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8240 = const()[name = tensor<string, []>("op_8240"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [921600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(589054784))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(589976448))), name = tensor<string, []>("up_blocks_2_resnets_0_conv_shortcut_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 1920, 1, 1])];
            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, []>(589976640)))];
            tensor<fp16, [2, 640, 32, 32]> x_17_cast = conv(bias = up_blocks_2_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_8240, groups = var_8182, pad = x_17_pad_0, pad_type = x_17_pad_type_0, strides = var_8238, weight = up_blocks_2_resnets_0_conv_shortcut_weight_to_fp16_palettized, x = input_367_cast)[name = tensor<string, []>("x_17_cast")];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_223_cast = add(x = x_17_cast, y = hidden_states_221_cast)[name = tensor<string, []>("hidden_states_223_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)[name = tensor<string, []>("reshape_176_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)[name = tensor<string, []>("reduce_mean_132_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> sub_88_cast = sub(x = reshape_176_cast, y = reduce_mean_132_cast)[name = tensor<string, []>("sub_88_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> square_44_cast = square(x = sub_88_cast)[name = tensor<string, []>("square_44_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)[name = tensor<string, []>("reduce_mean_134_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)[name = tensor<string, []>("add_88_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_44_cast = sqrt(x = add_88_cast)[name = tensor<string, []>("sqrt_44_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> real_div_44_cast = real_div(x = sub_88_cast, y = sqrt_44_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_177_cast")];
            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, []>(589977984)))];
            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, []>(589979328)))];
            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_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_177_cast)[name = tensor<string, []>("add_89_cast")];
            tensor<int32, [2]> var_8260 = const()[name = tensor<string, []>("op_8260"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8262 = const()[name = tensor<string, []>("op_8262"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(589980672))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(590287936))), name = tensor<string, []>("up_blocks_2_attentions_0_proj_in_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            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, []>(590288128)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_225_cast = conv(bias = up_blocks_2_attentions_0_proj_in_bias_to_fp16, dilations = var_8262, groups = var_8182, pad = hidden_states_225_pad_0, pad_type = hidden_states_225_pad_type_0, strides = var_8260, weight = up_blocks_2_attentions_0_proj_in_weight_to_fp16_palettized, x = add_89_cast)[name = tensor<string, []>("hidden_states_225_cast")];
            tensor<int32, [4]> var_8267 = const()[name = tensor<string, []>("op_8267"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<fp16, [2, 640, 1, 1024]> inputs_61_cast = reshape(shape = var_8267, x = hidden_states_225_cast)[name = tensor<string, []>("inputs_61_cast")];
            tensor<int32, [1]> var_8277 = const()[name = tensor<string, []>("op_8277"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_61_cast = reduce_mean(axes = var_8277, keep_dims = var_8177, x = inputs_61_cast)[name = tensor<string, []>("channels_mean_61_cast")];
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_61_cast = sub(x = inputs_61_cast, y = channels_mean_61_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_61_cast")];
            tensor<int32, [1]> var_8281 = const()[name = tensor<string, []>("op_8281"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_8282_cast = reduce_mean(axes = var_8281, keep_dims = var_8177, x = zero_mean_sq_61_cast)[name = tensor<string, []>("op_8282_cast")];
            tensor<fp16, []> var_8283_to_fp16 = const()[name = tensor<string, []>("op_8283_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_8284_cast = add(x = var_8282_cast, y = var_8283_to_fp16)[name = tensor<string, []>("op_8284_cast")];
            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_8284_cast)[name = tensor<string, []>("denom_61_cast")];
            tensor<fp16, [2, 640, 1, 1024]> out_61_cast = mul(x = zero_mean_61_cast, y = denom_61_cast)[name = tensor<string, []>("out_61_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_8288_to_fp16 = const()[name = tensor<string, []>("op_8288_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(590289472)))];
            tensor<fp16, [2, 640, 1, 1024]> var_8289_cast = add(x = out_61_cast, y = var_8288_to_fp16)[name = tensor<string, []>("op_8289_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_8291_to_fp16 = const()[name = tensor<string, []>("op_8291_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(590290816)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_227_cast = mul(x = var_8289_cast, y = var_8291_to_fp16)[name = tensor<string, []>("hidden_states_227_cast")];
            tensor<int32, [2]> var_8298 = const()[name = tensor<string, []>("op_8298"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8300 = const()[name = tensor<string, []>("op_8300"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(590292160))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(590599424))), name = tensor<string, []>("up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 1024]> q_41_cast = conv(dilations = var_8300, groups = var_8182, pad = q_41_pad_0, pad_type = q_41_pad_type_0, strides = var_8298, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_227_cast)[name = tensor<string, []>("q_41_cast")];
            tensor<int32, [2]> var_8304 = const()[name = tensor<string, []>("op_8304"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8306 = const()[name = tensor<string, []>("op_8306"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(590599616))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(590906880))), name = tensor<string, []>("up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 1024]> k_81_cast = conv(dilations = var_8306, groups = var_8182, pad = k_81_pad_0, pad_type = k_81_pad_type_0, strides = var_8304, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_227_cast)[name = tensor<string, []>("k_81_cast")];
            tensor<int32, [2]> var_8310 = const()[name = tensor<string, []>("op_8310"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8312 = const()[name = tensor<string, []>("op_8312"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(590907072))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(591214336))), name = tensor<string, []>("up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 1024]> v_41_cast = conv(dilations = var_8312, groups = var_8182, pad = v_41_pad_0, pad_type = v_41_pad_type_0, strides = var_8310, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_227_cast)[name = tensor<string, []>("v_41_cast")];
            tensor<int32, [4]> var_8316_begin_0 = const()[name = tensor<string, []>("op_8316_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8316_end_0 = const()[name = tensor<string, []>("op_8316_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            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, 80, 1, 1024]> var_8316_cast = slice_by_index(begin = var_8316_begin_0, end = var_8316_end_0, end_mask = var_8316_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_8316_cast")];
            tensor<int32, [4]> var_8320_begin_0 = const()[name = tensor<string, []>("op_8320_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_8320_end_0 = const()[name = tensor<string, []>("op_8320_end_0"), val = tensor<int32, [4]>([2, 160, 1, 1024])];
            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, 80, 1, 1024]> var_8320_cast = slice_by_index(begin = var_8320_begin_0, end = var_8320_end_0, end_mask = var_8320_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_8320_cast")];
            tensor<int32, [4]> var_8324_begin_0 = const()[name = tensor<string, []>("op_8324_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_8324_end_0 = const()[name = tensor<string, []>("op_8324_end_0"), val = tensor<int32, [4]>([2, 240, 1, 1024])];
            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, 80, 1, 1024]> var_8324_cast = slice_by_index(begin = var_8324_begin_0, end = var_8324_end_0, end_mask = var_8324_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_8324_cast")];
            tensor<int32, [4]> var_8328_begin_0 = const()[name = tensor<string, []>("op_8328_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_8328_end_0 = const()[name = tensor<string, []>("op_8328_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            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, 80, 1, 1024]> var_8328_cast = slice_by_index(begin = var_8328_begin_0, end = var_8328_end_0, end_mask = var_8328_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_8328_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, 400, 1, 1024])];
            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, 80, 1, 1024]> var_8332_cast = slice_by_index(begin = var_8332_begin_0, end = var_8332_end_0, end_mask = var_8332_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_8332_cast")];
            tensor<int32, [4]> var_8336_begin_0 = const()[name = tensor<string, []>("op_8336_begin_0"), val = tensor<int32, [4]>([0, 400, 0, 0])];
            tensor<int32, [4]> var_8336_end_0 = const()[name = tensor<string, []>("op_8336_end_0"), val = tensor<int32, [4]>([2, 480, 1, 1024])];
            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, 80, 1, 1024]> var_8336_cast = slice_by_index(begin = var_8336_begin_0, end = var_8336_end_0, end_mask = var_8336_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_8336_cast")];
            tensor<int32, [4]> var_8340_begin_0 = const()[name = tensor<string, []>("op_8340_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_8340_end_0 = const()[name = tensor<string, []>("op_8340_end_0"), val = tensor<int32, [4]>([2, 560, 1, 1024])];
            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, 80, 1, 1024]> var_8340_cast = slice_by_index(begin = var_8340_begin_0, end = var_8340_end_0, end_mask = var_8340_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_8340_cast")];
            tensor<int32, [4]> var_8344_begin_0 = const()[name = tensor<string, []>("op_8344_begin_0"), val = tensor<int32, [4]>([0, 560, 0, 0])];
            tensor<int32, [4]> var_8344_end_0 = const()[name = tensor<string, []>("op_8344_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            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, 80, 1, 1024]> var_8344_cast = slice_by_index(begin = var_8344_begin_0, end = var_8344_end_0, end_mask = var_8344_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_8344_cast")];
            tensor<int32, [4]> var_8347_begin_0 = const()[name = tensor<string, []>("op_8347_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8347_end_0 = const()[name = tensor<string, []>("op_8347_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_8347_end_mask_0 = const()[name = tensor<string, []>("op_8347_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8347_cast = slice_by_index(begin = var_8347_begin_0, end = var_8347_end_0, end_mask = var_8347_end_mask_0, x = var_8316_cast)[name = tensor<string, []>("op_8347_cast")];
            tensor<int32, [4]> var_8348_begin_0 = const()[name = tensor<string, []>("op_8348_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_8348_end_0 = const()[name = tensor<string, []>("op_8348_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_8348_end_mask_0 = const()[name = tensor<string, []>("op_8348_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8348_cast = slice_by_index(begin = var_8348_begin_0, end = var_8348_end_0, end_mask = var_8348_end_mask_0, x = var_8316_cast)[name = tensor<string, []>("op_8348_cast")];
            tensor<int32, [4]> var_8349_begin_0 = const()[name = tensor<string, []>("op_8349_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8349_end_0 = const()[name = tensor<string, []>("op_8349_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_8349_end_mask_0 = const()[name = tensor<string, []>("op_8349_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8349_cast = slice_by_index(begin = var_8349_begin_0, end = var_8349_end_0, end_mask = var_8349_end_mask_0, x = var_8320_cast)[name = tensor<string, []>("op_8349_cast")];
            tensor<int32, [4]> var_8350_begin_0 = const()[name = tensor<string, []>("op_8350_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_8350_end_0 = const()[name = tensor<string, []>("op_8350_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_8350_end_mask_0 = const()[name = tensor<string, []>("op_8350_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8350_cast = slice_by_index(begin = var_8350_begin_0, end = var_8350_end_0, end_mask = var_8350_end_mask_0, x = var_8320_cast)[name = tensor<string, []>("op_8350_cast")];
            tensor<int32, [4]> var_8351_begin_0 = const()[name = tensor<string, []>("op_8351_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8351_end_0 = const()[name = tensor<string, []>("op_8351_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_8351_end_mask_0 = const()[name = tensor<string, []>("op_8351_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8351_cast = slice_by_index(begin = var_8351_begin_0, end = var_8351_end_0, end_mask = var_8351_end_mask_0, x = var_8324_cast)[name = tensor<string, []>("op_8351_cast")];
            tensor<int32, [4]> var_8352_begin_0 = const()[name = tensor<string, []>("op_8352_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_8352_end_0 = const()[name = tensor<string, []>("op_8352_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_8352_end_mask_0 = const()[name = tensor<string, []>("op_8352_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8352_cast = slice_by_index(begin = var_8352_begin_0, end = var_8352_end_0, end_mask = var_8352_end_mask_0, x = var_8324_cast)[name = tensor<string, []>("op_8352_cast")];
            tensor<int32, [4]> var_8353_begin_0 = const()[name = tensor<string, []>("op_8353_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8353_end_0 = const()[name = tensor<string, []>("op_8353_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_8353_end_mask_0 = const()[name = tensor<string, []>("op_8353_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8353_cast = slice_by_index(begin = var_8353_begin_0, end = var_8353_end_0, end_mask = var_8353_end_mask_0, x = var_8328_cast)[name = tensor<string, []>("op_8353_cast")];
            tensor<int32, [4]> var_8354_begin_0 = const()[name = tensor<string, []>("op_8354_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_8354_end_0 = const()[name = tensor<string, []>("op_8354_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_8354_end_mask_0 = const()[name = tensor<string, []>("op_8354_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8354_cast = slice_by_index(begin = var_8354_begin_0, end = var_8354_end_0, end_mask = var_8354_end_mask_0, x = var_8328_cast)[name = tensor<string, []>("op_8354_cast")];
            tensor<int32, [4]> var_8355_begin_0 = const()[name = tensor<string, []>("op_8355_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8355_end_0 = const()[name = tensor<string, []>("op_8355_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_8355_end_mask_0 = const()[name = tensor<string, []>("op_8355_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8355_cast = slice_by_index(begin = var_8355_begin_0, end = var_8355_end_0, end_mask = var_8355_end_mask_0, x = var_8332_cast)[name = tensor<string, []>("op_8355_cast")];
            tensor<int32, [4]> var_8356_begin_0 = const()[name = tensor<string, []>("op_8356_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_8356_end_0 = const()[name = tensor<string, []>("op_8356_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_8356_end_mask_0 = const()[name = tensor<string, []>("op_8356_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8356_cast = slice_by_index(begin = var_8356_begin_0, end = var_8356_end_0, end_mask = var_8356_end_mask_0, x = var_8332_cast)[name = tensor<string, []>("op_8356_cast")];
            tensor<int32, [4]> var_8357_begin_0 = const()[name = tensor<string, []>("op_8357_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8357_end_0 = const()[name = tensor<string, []>("op_8357_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_8357_end_mask_0 = const()[name = tensor<string, []>("op_8357_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8357_cast = slice_by_index(begin = var_8357_begin_0, end = var_8357_end_0, end_mask = var_8357_end_mask_0, x = var_8336_cast)[name = tensor<string, []>("op_8357_cast")];
            tensor<int32, [4]> var_8358_begin_0 = const()[name = tensor<string, []>("op_8358_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_8358_end_0 = const()[name = tensor<string, []>("op_8358_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_8358_end_mask_0 = const()[name = tensor<string, []>("op_8358_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8358_cast = slice_by_index(begin = var_8358_begin_0, end = var_8358_end_0, end_mask = var_8358_end_mask_0, x = var_8336_cast)[name = tensor<string, []>("op_8358_cast")];
            tensor<int32, [4]> var_8359_begin_0 = const()[name = tensor<string, []>("op_8359_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8359_end_0 = const()[name = tensor<string, []>("op_8359_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_8359_end_mask_0 = const()[name = tensor<string, []>("op_8359_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8359_cast = slice_by_index(begin = var_8359_begin_0, end = var_8359_end_0, end_mask = var_8359_end_mask_0, x = var_8340_cast)[name = tensor<string, []>("op_8359_cast")];
            tensor<int32, [4]> var_8360_begin_0 = const()[name = tensor<string, []>("op_8360_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_8360_end_0 = const()[name = tensor<string, []>("op_8360_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_8360_end_mask_0 = const()[name = tensor<string, []>("op_8360_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8360_cast = slice_by_index(begin = var_8360_begin_0, end = var_8360_end_0, end_mask = var_8360_end_mask_0, x = var_8340_cast)[name = tensor<string, []>("op_8360_cast")];
            tensor<int32, [4]> var_8361_begin_0 = const()[name = tensor<string, []>("op_8361_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8361_end_0 = const()[name = tensor<string, []>("op_8361_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_8361_end_mask_0 = const()[name = tensor<string, []>("op_8361_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8361_cast = slice_by_index(begin = var_8361_begin_0, end = var_8361_end_0, end_mask = var_8361_end_mask_0, x = var_8344_cast)[name = tensor<string, []>("op_8361_cast")];
            tensor<int32, [4]> var_8362_begin_0 = const()[name = tensor<string, []>("op_8362_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_8362_end_0 = const()[name = tensor<string, []>("op_8362_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_8362_end_mask_0 = const()[name = tensor<string, []>("op_8362_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8362_cast = slice_by_index(begin = var_8362_begin_0, end = var_8362_end_0, end_mask = var_8362_end_mask_0, x = var_8344_cast)[name = tensor<string, []>("op_8362_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_8367_begin_0 = const()[name = tensor<string, []>("op_8367_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8367_end_0 = const()[name = tensor<string, []>("op_8367_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 80])];
            tensor<bool, [4]> var_8367_end_mask_0 = const()[name = tensor<string, []>("op_8367_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)[name = tensor<string, []>("transpose_11")];
            tensor<fp16, [2, 1024, 1, 80]> var_8367_cast = slice_by_index(begin = var_8367_begin_0, end = var_8367_end_0, end_mask = var_8367_end_mask_0, x = transpose_11)[name = tensor<string, []>("op_8367_cast")];
            tensor<int32, [4]> var_8371_begin_0 = const()[name = tensor<string, []>("op_8371_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 80])];
            tensor<int32, [4]> var_8371_end_0 = const()[name = tensor<string, []>("op_8371_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 160])];
            tensor<bool, [4]> var_8371_end_mask_0 = const()[name = tensor<string, []>("op_8371_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_8371_cast = slice_by_index(begin = var_8371_begin_0, end = var_8371_end_0, end_mask = var_8371_end_mask_0, x = transpose_11)[name = tensor<string, []>("op_8371_cast")];
            tensor<int32, [4]> var_8375_begin_0 = const()[name = tensor<string, []>("op_8375_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_8375_end_0 = const()[name = tensor<string, []>("op_8375_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 240])];
            tensor<bool, [4]> var_8375_end_mask_0 = const()[name = tensor<string, []>("op_8375_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_8375_cast = slice_by_index(begin = var_8375_begin_0, end = var_8375_end_0, end_mask = var_8375_end_mask_0, x = transpose_11)[name = tensor<string, []>("op_8375_cast")];
            tensor<int32, [4]> var_8379_begin_0 = const()[name = tensor<string, []>("op_8379_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 240])];
            tensor<int32, [4]> var_8379_end_0 = const()[name = tensor<string, []>("op_8379_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_8379_end_mask_0 = const()[name = tensor<string, []>("op_8379_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_8379_cast = slice_by_index(begin = var_8379_begin_0, end = var_8379_end_0, end_mask = var_8379_end_mask_0, x = transpose_11)[name = tensor<string, []>("op_8379_cast")];
            tensor<int32, [4]> var_8383_begin_0 = const()[name = tensor<string, []>("op_8383_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_8383_end_0 = const()[name = tensor<string, []>("op_8383_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 400])];
            tensor<bool, [4]> var_8383_end_mask_0 = const()[name = tensor<string, []>("op_8383_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_8383_cast = slice_by_index(begin = var_8383_begin_0, end = var_8383_end_0, end_mask = var_8383_end_mask_0, x = transpose_11)[name = tensor<string, []>("op_8383_cast")];
            tensor<int32, [4]> var_8387_begin_0 = const()[name = tensor<string, []>("op_8387_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 400])];
            tensor<int32, [4]> var_8387_end_0 = const()[name = tensor<string, []>("op_8387_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 480])];
            tensor<bool, [4]> var_8387_end_mask_0 = const()[name = tensor<string, []>("op_8387_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_8387_cast = slice_by_index(begin = var_8387_begin_0, end = var_8387_end_0, end_mask = var_8387_end_mask_0, x = transpose_11)[name = tensor<string, []>("op_8387_cast")];
            tensor<int32, [4]> var_8391_begin_0 = const()[name = tensor<string, []>("op_8391_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 480])];
            tensor<int32, [4]> var_8391_end_0 = const()[name = tensor<string, []>("op_8391_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 560])];
            tensor<bool, [4]> var_8391_end_mask_0 = const()[name = tensor<string, []>("op_8391_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_8391_cast = slice_by_index(begin = var_8391_begin_0, end = var_8391_end_0, end_mask = var_8391_end_mask_0, x = transpose_11)[name = tensor<string, []>("op_8391_cast")];
            tensor<int32, [4]> var_8395_begin_0 = const()[name = tensor<string, []>("op_8395_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 560])];
            tensor<int32, [4]> var_8395_end_0 = const()[name = tensor<string, []>("op_8395_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_8395_end_mask_0 = const()[name = tensor<string, []>("op_8395_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_8395_cast = slice_by_index(begin = var_8395_begin_0, end = var_8395_end_0, end_mask = var_8395_end_mask_0, x = transpose_11)[name = tensor<string, []>("op_8395_cast")];
            tensor<int32, [4]> var_8397_begin_0 = const()[name = tensor<string, []>("op_8397_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8397_end_0 = const()[name = tensor<string, []>("op_8397_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_8397_end_mask_0 = const()[name = tensor<string, []>("op_8397_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_8397_cast = slice_by_index(begin = var_8397_begin_0, end = var_8397_end_0, end_mask = var_8397_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_8397_cast")];
            tensor<int32, [4]> var_8401_begin_0 = const()[name = tensor<string, []>("op_8401_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_8401_end_0 = const()[name = tensor<string, []>("op_8401_end_0"), val = tensor<int32, [4]>([2, 160, 1, 1024])];
            tensor<bool, [4]> var_8401_end_mask_0 = const()[name = tensor<string, []>("op_8401_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_8401_cast = slice_by_index(begin = var_8401_begin_0, end = var_8401_end_0, end_mask = var_8401_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_8401_cast")];
            tensor<int32, [4]> var_8405_begin_0 = const()[name = tensor<string, []>("op_8405_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_8405_end_0 = const()[name = tensor<string, []>("op_8405_end_0"), val = tensor<int32, [4]>([2, 240, 1, 1024])];
            tensor<bool, [4]> var_8405_end_mask_0 = const()[name = tensor<string, []>("op_8405_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_8405_cast = slice_by_index(begin = var_8405_begin_0, end = var_8405_end_0, end_mask = var_8405_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_8405_cast")];
            tensor<int32, [4]> var_8409_begin_0 = const()[name = tensor<string, []>("op_8409_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_8409_end_0 = const()[name = tensor<string, []>("op_8409_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_8409_end_mask_0 = const()[name = tensor<string, []>("op_8409_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_8409_cast = slice_by_index(begin = var_8409_begin_0, end = var_8409_end_0, end_mask = var_8409_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_8409_cast")];
            tensor<int32, [4]> var_8413_begin_0 = const()[name = tensor<string, []>("op_8413_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_8413_end_0 = const()[name = tensor<string, []>("op_8413_end_0"), val = tensor<int32, [4]>([2, 400, 1, 1024])];
            tensor<bool, [4]> var_8413_end_mask_0 = const()[name = tensor<string, []>("op_8413_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_8413_cast = slice_by_index(begin = var_8413_begin_0, end = var_8413_end_0, end_mask = var_8413_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_8413_cast")];
            tensor<int32, [4]> var_8417_begin_0 = const()[name = tensor<string, []>("op_8417_begin_0"), val = tensor<int32, [4]>([0, 400, 0, 0])];
            tensor<int32, [4]> var_8417_end_0 = const()[name = tensor<string, []>("op_8417_end_0"), val = tensor<int32, [4]>([2, 480, 1, 1024])];
            tensor<bool, [4]> var_8417_end_mask_0 = const()[name = tensor<string, []>("op_8417_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_8417_cast = slice_by_index(begin = var_8417_begin_0, end = var_8417_end_0, end_mask = var_8417_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_8417_cast")];
            tensor<int32, [4]> var_8421_begin_0 = const()[name = tensor<string, []>("op_8421_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_8421_end_0 = const()[name = tensor<string, []>("op_8421_end_0"), val = tensor<int32, [4]>([2, 560, 1, 1024])];
            tensor<bool, [4]> var_8421_end_mask_0 = const()[name = tensor<string, []>("op_8421_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_8421_cast = slice_by_index(begin = var_8421_begin_0, end = var_8421_end_0, end_mask = var_8421_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_8421_cast")];
            tensor<int32, [4]> var_8425_begin_0 = const()[name = tensor<string, []>("op_8425_begin_0"), val = tensor<int32, [4]>([0, 560, 0, 0])];
            tensor<int32, [4]> var_8425_end_0 = const()[name = tensor<string, []>("op_8425_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_8425_end_mask_0 = const()[name = tensor<string, []>("op_8425_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_8425_cast = slice_by_index(begin = var_8425_begin_0, end = var_8425_end_0, end_mask = var_8425_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_8425_cast")];
            tensor<string, []> var_8429_equation_0 = const()[name = tensor<string, []>("op_8429_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_8429_cast = einsum(equation = var_8429_equation_0, values = (var_8367_cast, var_8347_cast))[name = tensor<string, []>("op_8429_cast")];
            tensor<fp16, []> var_8430_to_fp16 = const()[name = tensor<string, []>("op_8430_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_641_cast = mul(x = var_8429_cast, y = var_8430_to_fp16)[name = tensor<string, []>("aw_chunk_641_cast")];
            tensor<string, []> var_8433_equation_0 = const()[name = tensor<string, []>("op_8433_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_8433_cast = einsum(equation = var_8433_equation_0, values = (var_8367_cast, var_8348_cast))[name = tensor<string, []>("op_8433_cast")];
            tensor<fp16, []> var_8434_to_fp16 = const()[name = tensor<string, []>("op_8434_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_643_cast = mul(x = var_8433_cast, y = var_8434_to_fp16)[name = tensor<string, []>("aw_chunk_643_cast")];
            tensor<string, []> var_8437_equation_0 = const()[name = tensor<string, []>("op_8437_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_8437_cast = einsum(equation = var_8437_equation_0, values = (var_8371_cast, var_8349_cast))[name = tensor<string, []>("op_8437_cast")];
            tensor<fp16, []> var_8438_to_fp16 = const()[name = tensor<string, []>("op_8438_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_645_cast = mul(x = var_8437_cast, y = var_8438_to_fp16)[name = tensor<string, []>("aw_chunk_645_cast")];
            tensor<string, []> var_8441_equation_0 = const()[name = tensor<string, []>("op_8441_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_8441_cast = einsum(equation = var_8441_equation_0, values = (var_8371_cast, var_8350_cast))[name = tensor<string, []>("op_8441_cast")];
            tensor<fp16, []> var_8442_to_fp16 = const()[name = tensor<string, []>("op_8442_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_647_cast = mul(x = var_8441_cast, y = var_8442_to_fp16)[name = tensor<string, []>("aw_chunk_647_cast")];
            tensor<string, []> var_8445_equation_0 = const()[name = tensor<string, []>("op_8445_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_8445_cast = einsum(equation = var_8445_equation_0, values = (var_8375_cast, var_8351_cast))[name = tensor<string, []>("op_8445_cast")];
            tensor<fp16, []> var_8446_to_fp16 = const()[name = tensor<string, []>("op_8446_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_649_cast = mul(x = var_8445_cast, y = var_8446_to_fp16)[name = tensor<string, []>("aw_chunk_649_cast")];
            tensor<string, []> var_8449_equation_0 = const()[name = tensor<string, []>("op_8449_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_8449_cast = einsum(equation = var_8449_equation_0, values = (var_8375_cast, var_8352_cast))[name = tensor<string, []>("op_8449_cast")];
            tensor<fp16, []> var_8450_to_fp16 = const()[name = tensor<string, []>("op_8450_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_651_cast = mul(x = var_8449_cast, y = var_8450_to_fp16)[name = tensor<string, []>("aw_chunk_651_cast")];
            tensor<string, []> var_8453_equation_0 = const()[name = tensor<string, []>("op_8453_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_8453_cast = einsum(equation = var_8453_equation_0, values = (var_8379_cast, var_8353_cast))[name = tensor<string, []>("op_8453_cast")];
            tensor<fp16, []> var_8454_to_fp16 = const()[name = tensor<string, []>("op_8454_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_653_cast = mul(x = var_8453_cast, y = var_8454_to_fp16)[name = tensor<string, []>("aw_chunk_653_cast")];
            tensor<string, []> var_8457_equation_0 = const()[name = tensor<string, []>("op_8457_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_8457_cast = einsum(equation = var_8457_equation_0, values = (var_8379_cast, var_8354_cast))[name = tensor<string, []>("op_8457_cast")];
            tensor<fp16, []> var_8458_to_fp16 = const()[name = tensor<string, []>("op_8458_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_655_cast = mul(x = var_8457_cast, y = var_8458_to_fp16)[name = tensor<string, []>("aw_chunk_655_cast")];
            tensor<string, []> var_8461_equation_0 = const()[name = tensor<string, []>("op_8461_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_8461_cast = einsum(equation = var_8461_equation_0, values = (var_8383_cast, var_8355_cast))[name = tensor<string, []>("op_8461_cast")];
            tensor<fp16, []> var_8462_to_fp16 = const()[name = tensor<string, []>("op_8462_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_657_cast = mul(x = var_8461_cast, y = var_8462_to_fp16)[name = tensor<string, []>("aw_chunk_657_cast")];
            tensor<string, []> var_8465_equation_0 = const()[name = tensor<string, []>("op_8465_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_8465_cast = einsum(equation = var_8465_equation_0, values = (var_8383_cast, var_8356_cast))[name = tensor<string, []>("op_8465_cast")];
            tensor<fp16, []> var_8466_to_fp16 = const()[name = tensor<string, []>("op_8466_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_659_cast = mul(x = var_8465_cast, y = var_8466_to_fp16)[name = tensor<string, []>("aw_chunk_659_cast")];
            tensor<string, []> var_8469_equation_0 = const()[name = tensor<string, []>("op_8469_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_8469_cast = einsum(equation = var_8469_equation_0, values = (var_8387_cast, var_8357_cast))[name = tensor<string, []>("op_8469_cast")];
            tensor<fp16, []> var_8470_to_fp16 = const()[name = tensor<string, []>("op_8470_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_661_cast = mul(x = var_8469_cast, y = var_8470_to_fp16)[name = tensor<string, []>("aw_chunk_661_cast")];
            tensor<string, []> var_8473_equation_0 = const()[name = tensor<string, []>("op_8473_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_8473_cast = einsum(equation = var_8473_equation_0, values = (var_8387_cast, var_8358_cast))[name = tensor<string, []>("op_8473_cast")];
            tensor<fp16, []> var_8474_to_fp16 = const()[name = tensor<string, []>("op_8474_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_663_cast = mul(x = var_8473_cast, y = var_8474_to_fp16)[name = tensor<string, []>("aw_chunk_663_cast")];
            tensor<string, []> var_8477_equation_0 = const()[name = tensor<string, []>("op_8477_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_8477_cast = einsum(equation = var_8477_equation_0, values = (var_8391_cast, var_8359_cast))[name = tensor<string, []>("op_8477_cast")];
            tensor<fp16, []> var_8478_to_fp16 = const()[name = tensor<string, []>("op_8478_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_665_cast = mul(x = var_8477_cast, y = var_8478_to_fp16)[name = tensor<string, []>("aw_chunk_665_cast")];
            tensor<string, []> var_8481_equation_0 = const()[name = tensor<string, []>("op_8481_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_8481_cast = einsum(equation = var_8481_equation_0, values = (var_8391_cast, var_8360_cast))[name = tensor<string, []>("op_8481_cast")];
            tensor<fp16, []> var_8482_to_fp16 = const()[name = tensor<string, []>("op_8482_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_667_cast = mul(x = var_8481_cast, y = var_8482_to_fp16)[name = tensor<string, []>("aw_chunk_667_cast")];
            tensor<string, []> var_8485_equation_0 = const()[name = tensor<string, []>("op_8485_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_8485_cast = einsum(equation = var_8485_equation_0, values = (var_8395_cast, var_8361_cast))[name = tensor<string, []>("op_8485_cast")];
            tensor<fp16, []> var_8486_to_fp16 = const()[name = tensor<string, []>("op_8486_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_669_cast = mul(x = var_8485_cast, y = var_8486_to_fp16)[name = tensor<string, []>("aw_chunk_669_cast")];
            tensor<string, []> var_8489_equation_0 = const()[name = tensor<string, []>("op_8489_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_8489_cast = einsum(equation = var_8489_equation_0, values = (var_8395_cast, var_8362_cast))[name = tensor<string, []>("op_8489_cast")];
            tensor<fp16, []> var_8490_to_fp16 = const()[name = tensor<string, []>("op_8490_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_671_cast = mul(x = var_8489_cast, y = var_8490_to_fp16)[name = tensor<string, []>("aw_chunk_671_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_8492_cast = softmax(axis = var_8182, x = aw_chunk_641_cast)[name = tensor<string, []>("op_8492_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_8493_cast = softmax(axis = var_8182, x = aw_chunk_643_cast)[name = tensor<string, []>("op_8493_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_8494_cast = softmax(axis = var_8182, x = aw_chunk_645_cast)[name = tensor<string, []>("op_8494_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_8495_cast = softmax(axis = var_8182, x = aw_chunk_647_cast)[name = tensor<string, []>("op_8495_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_8496_cast = softmax(axis = var_8182, x = aw_chunk_649_cast)[name = tensor<string, []>("op_8496_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_8497_cast = softmax(axis = var_8182, x = aw_chunk_651_cast)[name = tensor<string, []>("op_8497_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_8498_cast = softmax(axis = var_8182, x = aw_chunk_653_cast)[name = tensor<string, []>("op_8498_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_8499_cast = softmax(axis = var_8182, x = aw_chunk_655_cast)[name = tensor<string, []>("op_8499_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_8500_cast = softmax(axis = var_8182, x = aw_chunk_657_cast)[name = tensor<string, []>("op_8500_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_8501_cast = softmax(axis = var_8182, x = aw_chunk_659_cast)[name = tensor<string, []>("op_8501_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_8502_cast = softmax(axis = var_8182, x = aw_chunk_661_cast)[name = tensor<string, []>("op_8502_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_8503_cast = softmax(axis = var_8182, x = aw_chunk_663_cast)[name = tensor<string, []>("op_8503_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_8504_cast = softmax(axis = var_8182, x = aw_chunk_665_cast)[name = tensor<string, []>("op_8504_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_8505_cast = softmax(axis = var_8182, x = aw_chunk_667_cast)[name = tensor<string, []>("op_8505_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_8506_cast = softmax(axis = var_8182, x = aw_chunk_669_cast)[name = tensor<string, []>("op_8506_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_8507_cast = softmax(axis = var_8182, x = aw_chunk_671_cast)[name = tensor<string, []>("op_8507_cast")];
            tensor<string, []> var_8509_equation_0 = const()[name = tensor<string, []>("op_8509_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8509_cast = einsum(equation = var_8509_equation_0, values = (var_8397_cast, var_8492_cast))[name = tensor<string, []>("op_8509_cast")];
            tensor<string, []> var_8511_equation_0 = const()[name = tensor<string, []>("op_8511_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8511_cast = einsum(equation = var_8511_equation_0, values = (var_8397_cast, var_8493_cast))[name = tensor<string, []>("op_8511_cast")];
            tensor<string, []> var_8513_equation_0 = const()[name = tensor<string, []>("op_8513_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8513_cast = einsum(equation = var_8513_equation_0, values = (var_8401_cast, var_8494_cast))[name = tensor<string, []>("op_8513_cast")];
            tensor<string, []> var_8515_equation_0 = const()[name = tensor<string, []>("op_8515_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8515_cast = einsum(equation = var_8515_equation_0, values = (var_8401_cast, var_8495_cast))[name = tensor<string, []>("op_8515_cast")];
            tensor<string, []> var_8517_equation_0 = const()[name = tensor<string, []>("op_8517_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8517_cast = einsum(equation = var_8517_equation_0, values = (var_8405_cast, var_8496_cast))[name = tensor<string, []>("op_8517_cast")];
            tensor<string, []> var_8519_equation_0 = const()[name = tensor<string, []>("op_8519_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8519_cast = einsum(equation = var_8519_equation_0, values = (var_8405_cast, var_8497_cast))[name = tensor<string, []>("op_8519_cast")];
            tensor<string, []> var_8521_equation_0 = const()[name = tensor<string, []>("op_8521_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8521_cast = einsum(equation = var_8521_equation_0, values = (var_8409_cast, var_8498_cast))[name = tensor<string, []>("op_8521_cast")];
            tensor<string, []> var_8523_equation_0 = const()[name = tensor<string, []>("op_8523_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8523_cast = einsum(equation = var_8523_equation_0, values = (var_8409_cast, var_8499_cast))[name = tensor<string, []>("op_8523_cast")];
            tensor<string, []> var_8525_equation_0 = const()[name = tensor<string, []>("op_8525_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8525_cast = einsum(equation = var_8525_equation_0, values = (var_8413_cast, var_8500_cast))[name = tensor<string, []>("op_8525_cast")];
            tensor<string, []> var_8527_equation_0 = const()[name = tensor<string, []>("op_8527_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8527_cast = einsum(equation = var_8527_equation_0, values = (var_8413_cast, var_8501_cast))[name = tensor<string, []>("op_8527_cast")];
            tensor<string, []> var_8529_equation_0 = const()[name = tensor<string, []>("op_8529_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8529_cast = einsum(equation = var_8529_equation_0, values = (var_8417_cast, var_8502_cast))[name = tensor<string, []>("op_8529_cast")];
            tensor<string, []> var_8531_equation_0 = const()[name = tensor<string, []>("op_8531_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8531_cast = einsum(equation = var_8531_equation_0, values = (var_8417_cast, var_8503_cast))[name = tensor<string, []>("op_8531_cast")];
            tensor<string, []> var_8533_equation_0 = const()[name = tensor<string, []>("op_8533_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8533_cast = einsum(equation = var_8533_equation_0, values = (var_8421_cast, var_8504_cast))[name = tensor<string, []>("op_8533_cast")];
            tensor<string, []> var_8535_equation_0 = const()[name = tensor<string, []>("op_8535_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8535_cast = einsum(equation = var_8535_equation_0, values = (var_8421_cast, var_8505_cast))[name = tensor<string, []>("op_8535_cast")];
            tensor<string, []> var_8537_equation_0 = const()[name = tensor<string, []>("op_8537_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8537_cast = einsum(equation = var_8537_equation_0, values = (var_8425_cast, var_8506_cast))[name = tensor<string, []>("op_8537_cast")];
            tensor<string, []> var_8539_equation_0 = const()[name = tensor<string, []>("op_8539_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8539_cast = einsum(equation = var_8539_equation_0, values = (var_8425_cast, var_8507_cast))[name = tensor<string, []>("op_8539_cast")];
            tensor<bool, []> var_8541_interleave_0 = const()[name = tensor<string, []>("op_8541_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_8541_cast = concat(axis = var_8160, interleave = var_8541_interleave_0, values = (var_8509_cast, var_8511_cast))[name = tensor<string, []>("op_8541_cast")];
            tensor<bool, []> var_8543_interleave_0 = const()[name = tensor<string, []>("op_8543_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_8543_cast = concat(axis = var_8160, interleave = var_8543_interleave_0, values = (var_8513_cast, var_8515_cast))[name = tensor<string, []>("op_8543_cast")];
            tensor<bool, []> var_8545_interleave_0 = const()[name = tensor<string, []>("op_8545_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_8545_cast = concat(axis = var_8160, interleave = var_8545_interleave_0, values = (var_8517_cast, var_8519_cast))[name = tensor<string, []>("op_8545_cast")];
            tensor<bool, []> var_8547_interleave_0 = const()[name = tensor<string, []>("op_8547_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_8547_cast = concat(axis = var_8160, interleave = var_8547_interleave_0, values = (var_8521_cast, var_8523_cast))[name = tensor<string, []>("op_8547_cast")];
            tensor<bool, []> var_8549_interleave_0 = const()[name = tensor<string, []>("op_8549_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_8549_cast = concat(axis = var_8160, interleave = var_8549_interleave_0, values = (var_8525_cast, var_8527_cast))[name = tensor<string, []>("op_8549_cast")];
            tensor<bool, []> var_8551_interleave_0 = const()[name = tensor<string, []>("op_8551_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_8551_cast = concat(axis = var_8160, interleave = var_8551_interleave_0, values = (var_8529_cast, var_8531_cast))[name = tensor<string, []>("op_8551_cast")];
            tensor<bool, []> var_8553_interleave_0 = const()[name = tensor<string, []>("op_8553_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_8553_cast = concat(axis = var_8160, interleave = var_8553_interleave_0, values = (var_8533_cast, var_8535_cast))[name = tensor<string, []>("op_8553_cast")];
            tensor<bool, []> var_8555_interleave_0 = const()[name = tensor<string, []>("op_8555_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_8555_cast = concat(axis = var_8160, interleave = var_8555_interleave_0, values = (var_8537_cast, var_8539_cast))[name = tensor<string, []>("op_8555_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_8182, interleave = input_383_interleave_0, values = (var_8541_cast, var_8543_cast, var_8545_cast, var_8547_cast, var_8549_cast, var_8551_cast, var_8553_cast, var_8555_cast))[name = tensor<string, []>("input_383_cast")];
            tensor<int32, [2]> var_8561 = const()[name = tensor<string, []>("op_8561"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8563 = const()[name = tensor<string, []>("op_8563"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_8565_pad_type_0 = const()[name = tensor<string, []>("op_8565_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_8565_pad_0 = const()[name = tensor<string, []>("op_8565_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(591214528))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(591521792))), name = tensor<string, []>("up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            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, []>(591521984)))];
            tensor<fp16, [2, 640, 1, 1024]> var_8565_cast = conv(bias = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_8563, groups = var_8182, pad = var_8565_pad_0, pad_type = var_8565_pad_type_0, strides = var_8561, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_383_cast)[name = tensor<string, []>("op_8565_cast")];
            tensor<fp16, [2, 640, 1, 1024]> inputs_63_cast = add(x = var_8565_cast, y = inputs_61_cast)[name = tensor<string, []>("inputs_63_cast")];
            tensor<int32, [1]> var_8569 = const()[name = tensor<string, []>("op_8569"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_63_cast = reduce_mean(axes = var_8569, keep_dims = var_8177, x = inputs_63_cast)[name = tensor<string, []>("channels_mean_63_cast")];
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_63_cast = sub(x = inputs_63_cast, y = channels_mean_63_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_63_cast")];
            tensor<int32, [1]> var_8573 = const()[name = tensor<string, []>("op_8573"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_8574_cast = reduce_mean(axes = var_8573, keep_dims = var_8177, x = zero_mean_sq_63_cast)[name = tensor<string, []>("op_8574_cast")];
            tensor<fp16, []> var_8575_to_fp16 = const()[name = tensor<string, []>("op_8575_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_8576_cast = add(x = var_8574_cast, y = var_8575_to_fp16)[name = tensor<string, []>("op_8576_cast")];
            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_8576_cast)[name = tensor<string, []>("denom_63_cast")];
            tensor<fp16, [2, 640, 1, 1024]> out_63_cast = mul(x = zero_mean_63_cast, y = denom_63_cast)[name = tensor<string, []>("out_63_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_8580_to_fp16 = const()[name = tensor<string, []>("op_8580_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(591523328)))];
            tensor<fp16, [2, 640, 1, 1024]> var_8581_cast = add(x = out_63_cast, y = var_8580_to_fp16)[name = tensor<string, []>("op_8581_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_8583_to_fp16 = const()[name = tensor<string, []>("op_8583_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(591524672)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_229_cast = mul(x = var_8581_cast, y = var_8583_to_fp16)[name = tensor<string, []>("hidden_states_229_cast")];
            tensor<int32, [2]> var_8590 = const()[name = tensor<string, []>("op_8590"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8592 = const()[name = tensor<string, []>("op_8592"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(591526016))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(591833280))), name = tensor<string, []>("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 1024]> q_43_cast = conv(dilations = var_8592, groups = var_8182, pad = q_43_pad_0, pad_type = q_43_pad_type_0, strides = var_8590, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_229_cast)[name = tensor<string, []>("q_43_cast")];
            tensor<int32, [2]> var_8596 = const()[name = tensor<string, []>("op_8596"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8598 = const()[name = tensor<string, []>("op_8598"), 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, 768, 1, 1]> up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [368640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(591833472))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(592202176))), name = tensor<string, []>("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 768, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> k_85_cast = conv(dilations = var_8598, groups = var_8182, pad = k_85_pad_0, pad_type = k_85_pad_type_0, strides = var_8596, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_85_cast")];
            tensor<int32, [2]> var_8602 = const()[name = tensor<string, []>("op_8602"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8604 = const()[name = tensor<string, []>("op_8604"), 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, 768, 1, 1]> up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [368640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(592202368))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(592571072))), name = tensor<string, []>("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 768, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> v_43_cast = conv(dilations = var_8604, groups = var_8182, pad = v_43_pad_0, pad_type = v_43_pad_type_0, strides = var_8602, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_43_cast")];
            tensor<int32, [4]> var_8608_begin_0 = const()[name = tensor<string, []>("op_8608_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8608_end_0 = const()[name = tensor<string, []>("op_8608_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_8608_end_mask_0 = const()[name = tensor<string, []>("op_8608_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_8608_cast = slice_by_index(begin = var_8608_begin_0, end = var_8608_end_0, end_mask = var_8608_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_8608_cast")];
            tensor<int32, [4]> var_8612_begin_0 = const()[name = tensor<string, []>("op_8612_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_8612_end_0 = const()[name = tensor<string, []>("op_8612_end_0"), val = tensor<int32, [4]>([2, 160, 1, 1024])];
            tensor<bool, [4]> var_8612_end_mask_0 = const()[name = tensor<string, []>("op_8612_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_8612_cast = slice_by_index(begin = var_8612_begin_0, end = var_8612_end_0, end_mask = var_8612_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_8612_cast")];
            tensor<int32, [4]> var_8616_begin_0 = const()[name = tensor<string, []>("op_8616_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_8616_end_0 = const()[name = tensor<string, []>("op_8616_end_0"), val = tensor<int32, [4]>([2, 240, 1, 1024])];
            tensor<bool, [4]> var_8616_end_mask_0 = const()[name = tensor<string, []>("op_8616_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_8616_cast = slice_by_index(begin = var_8616_begin_0, end = var_8616_end_0, end_mask = var_8616_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_8616_cast")];
            tensor<int32, [4]> var_8620_begin_0 = const()[name = tensor<string, []>("op_8620_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_8620_end_0 = const()[name = tensor<string, []>("op_8620_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_8620_end_mask_0 = const()[name = tensor<string, []>("op_8620_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_8620_cast = slice_by_index(begin = var_8620_begin_0, end = var_8620_end_0, end_mask = var_8620_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_8620_cast")];
            tensor<int32, [4]> var_8624_begin_0 = const()[name = tensor<string, []>("op_8624_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_8624_end_0 = const()[name = tensor<string, []>("op_8624_end_0"), val = tensor<int32, [4]>([2, 400, 1, 1024])];
            tensor<bool, [4]> var_8624_end_mask_0 = const()[name = tensor<string, []>("op_8624_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_8624_cast = slice_by_index(begin = var_8624_begin_0, end = var_8624_end_0, end_mask = var_8624_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_8624_cast")];
            tensor<int32, [4]> var_8628_begin_0 = const()[name = tensor<string, []>("op_8628_begin_0"), val = tensor<int32, [4]>([0, 400, 0, 0])];
            tensor<int32, [4]> var_8628_end_0 = const()[name = tensor<string, []>("op_8628_end_0"), val = tensor<int32, [4]>([2, 480, 1, 1024])];
            tensor<bool, [4]> var_8628_end_mask_0 = const()[name = tensor<string, []>("op_8628_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_8628_cast = slice_by_index(begin = var_8628_begin_0, end = var_8628_end_0, end_mask = var_8628_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_8628_cast")];
            tensor<int32, [4]> var_8632_begin_0 = const()[name = tensor<string, []>("op_8632_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_8632_end_0 = const()[name = tensor<string, []>("op_8632_end_0"), val = tensor<int32, [4]>([2, 560, 1, 1024])];
            tensor<bool, [4]> var_8632_end_mask_0 = const()[name = tensor<string, []>("op_8632_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_8632_cast = slice_by_index(begin = var_8632_begin_0, end = var_8632_end_0, end_mask = var_8632_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_8632_cast")];
            tensor<int32, [4]> var_8636_begin_0 = const()[name = tensor<string, []>("op_8636_begin_0"), val = tensor<int32, [4]>([0, 560, 0, 0])];
            tensor<int32, [4]> var_8636_end_0 = const()[name = tensor<string, []>("op_8636_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_8636_end_mask_0 = const()[name = tensor<string, []>("op_8636_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_8636_cast = slice_by_index(begin = var_8636_begin_0, end = var_8636_end_0, end_mask = var_8636_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_8636_cast")];
            tensor<int32, [4]> var_8639_begin_0 = const()[name = tensor<string, []>("op_8639_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8639_end_0 = const()[name = tensor<string, []>("op_8639_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_8639_end_mask_0 = const()[name = tensor<string, []>("op_8639_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8639_cast = slice_by_index(begin = var_8639_begin_0, end = var_8639_end_0, end_mask = var_8639_end_mask_0, x = var_8608_cast)[name = tensor<string, []>("op_8639_cast")];
            tensor<int32, [4]> var_8640_begin_0 = const()[name = tensor<string, []>("op_8640_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_8640_end_0 = const()[name = tensor<string, []>("op_8640_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_8640_end_mask_0 = const()[name = tensor<string, []>("op_8640_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8640_cast = slice_by_index(begin = var_8640_begin_0, end = var_8640_end_0, end_mask = var_8640_end_mask_0, x = var_8608_cast)[name = tensor<string, []>("op_8640_cast")];
            tensor<int32, [4]> var_8641_begin_0 = const()[name = tensor<string, []>("op_8641_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8641_end_0 = const()[name = tensor<string, []>("op_8641_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_8641_end_mask_0 = const()[name = tensor<string, []>("op_8641_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8641_cast = slice_by_index(begin = var_8641_begin_0, end = var_8641_end_0, end_mask = var_8641_end_mask_0, x = var_8612_cast)[name = tensor<string, []>("op_8641_cast")];
            tensor<int32, [4]> var_8642_begin_0 = const()[name = tensor<string, []>("op_8642_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_8642_end_0 = const()[name = tensor<string, []>("op_8642_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_8642_end_mask_0 = const()[name = tensor<string, []>("op_8642_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8642_cast = slice_by_index(begin = var_8642_begin_0, end = var_8642_end_0, end_mask = var_8642_end_mask_0, x = var_8612_cast)[name = tensor<string, []>("op_8642_cast")];
            tensor<int32, [4]> var_8643_begin_0 = const()[name = tensor<string, []>("op_8643_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8643_end_0 = const()[name = tensor<string, []>("op_8643_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_8643_end_mask_0 = const()[name = tensor<string, []>("op_8643_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8643_cast = slice_by_index(begin = var_8643_begin_0, end = var_8643_end_0, end_mask = var_8643_end_mask_0, x = var_8616_cast)[name = tensor<string, []>("op_8643_cast")];
            tensor<int32, [4]> var_8644_begin_0 = const()[name = tensor<string, []>("op_8644_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_8644_end_0 = const()[name = tensor<string, []>("op_8644_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_8644_end_mask_0 = const()[name = tensor<string, []>("op_8644_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8644_cast = slice_by_index(begin = var_8644_begin_0, end = var_8644_end_0, end_mask = var_8644_end_mask_0, x = var_8616_cast)[name = tensor<string, []>("op_8644_cast")];
            tensor<int32, [4]> var_8645_begin_0 = const()[name = tensor<string, []>("op_8645_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8645_end_0 = const()[name = tensor<string, []>("op_8645_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_8645_end_mask_0 = const()[name = tensor<string, []>("op_8645_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8645_cast = slice_by_index(begin = var_8645_begin_0, end = var_8645_end_0, end_mask = var_8645_end_mask_0, x = var_8620_cast)[name = tensor<string, []>("op_8645_cast")];
            tensor<int32, [4]> var_8646_begin_0 = const()[name = tensor<string, []>("op_8646_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_8646_end_0 = const()[name = tensor<string, []>("op_8646_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_8646_end_mask_0 = const()[name = tensor<string, []>("op_8646_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8646_cast = slice_by_index(begin = var_8646_begin_0, end = var_8646_end_0, end_mask = var_8646_end_mask_0, x = var_8620_cast)[name = tensor<string, []>("op_8646_cast")];
            tensor<int32, [4]> var_8647_begin_0 = const()[name = tensor<string, []>("op_8647_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8647_end_0 = const()[name = tensor<string, []>("op_8647_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_8647_end_mask_0 = const()[name = tensor<string, []>("op_8647_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8647_cast = slice_by_index(begin = var_8647_begin_0, end = var_8647_end_0, end_mask = var_8647_end_mask_0, x = var_8624_cast)[name = tensor<string, []>("op_8647_cast")];
            tensor<int32, [4]> var_8648_begin_0 = const()[name = tensor<string, []>("op_8648_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_8648_end_0 = const()[name = tensor<string, []>("op_8648_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_8648_end_mask_0 = const()[name = tensor<string, []>("op_8648_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8648_cast = slice_by_index(begin = var_8648_begin_0, end = var_8648_end_0, end_mask = var_8648_end_mask_0, x = var_8624_cast)[name = tensor<string, []>("op_8648_cast")];
            tensor<int32, [4]> var_8649_begin_0 = const()[name = tensor<string, []>("op_8649_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8649_end_0 = const()[name = tensor<string, []>("op_8649_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_8649_end_mask_0 = const()[name = tensor<string, []>("op_8649_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8649_cast = slice_by_index(begin = var_8649_begin_0, end = var_8649_end_0, end_mask = var_8649_end_mask_0, x = var_8628_cast)[name = tensor<string, []>("op_8649_cast")];
            tensor<int32, [4]> var_8650_begin_0 = const()[name = tensor<string, []>("op_8650_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_8650_end_0 = const()[name = tensor<string, []>("op_8650_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_8650_end_mask_0 = const()[name = tensor<string, []>("op_8650_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8650_cast = slice_by_index(begin = var_8650_begin_0, end = var_8650_end_0, end_mask = var_8650_end_mask_0, x = var_8628_cast)[name = tensor<string, []>("op_8650_cast")];
            tensor<int32, [4]> var_8651_begin_0 = const()[name = tensor<string, []>("op_8651_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8651_end_0 = const()[name = tensor<string, []>("op_8651_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_8651_end_mask_0 = const()[name = tensor<string, []>("op_8651_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8651_cast = slice_by_index(begin = var_8651_begin_0, end = var_8651_end_0, end_mask = var_8651_end_mask_0, x = var_8632_cast)[name = tensor<string, []>("op_8651_cast")];
            tensor<int32, [4]> var_8652_begin_0 = const()[name = tensor<string, []>("op_8652_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_8652_end_0 = const()[name = tensor<string, []>("op_8652_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_8652_end_mask_0 = const()[name = tensor<string, []>("op_8652_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8652_cast = slice_by_index(begin = var_8652_begin_0, end = var_8652_end_0, end_mask = var_8652_end_mask_0, x = var_8632_cast)[name = tensor<string, []>("op_8652_cast")];
            tensor<int32, [4]> var_8653_begin_0 = const()[name = tensor<string, []>("op_8653_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8653_end_0 = const()[name = tensor<string, []>("op_8653_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_8653_end_mask_0 = const()[name = tensor<string, []>("op_8653_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8653_cast = slice_by_index(begin = var_8653_begin_0, end = var_8653_end_0, end_mask = var_8653_end_mask_0, x = var_8636_cast)[name = tensor<string, []>("op_8653_cast")];
            tensor<int32, [4]> var_8654_begin_0 = const()[name = tensor<string, []>("op_8654_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_8654_end_0 = const()[name = tensor<string, []>("op_8654_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_8654_end_mask_0 = const()[name = tensor<string, []>("op_8654_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_8654_cast = slice_by_index(begin = var_8654_begin_0, end = var_8654_end_0, end_mask = var_8654_end_mask_0, x = var_8636_cast)[name = tensor<string, []>("op_8654_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_8659_begin_0 = const()[name = tensor<string, []>("op_8659_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8659_end_0 = const()[name = tensor<string, []>("op_8659_end_0"), val = tensor<int32, [4]>([2, 77, 1, 80])];
            tensor<bool, [4]> var_8659_end_mask_0 = const()[name = tensor<string, []>("op_8659_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)[name = tensor<string, []>("transpose_10")];
            tensor<fp16, [2, 77, 1, 80]> var_8659_cast = slice_by_index(begin = var_8659_begin_0, end = var_8659_end_0, end_mask = var_8659_end_mask_0, x = transpose_10)[name = tensor<string, []>("op_8659_cast")];
            tensor<int32, [4]> var_8663_begin_0 = const()[name = tensor<string, []>("op_8663_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 80])];
            tensor<int32, [4]> var_8663_end_0 = const()[name = tensor<string, []>("op_8663_end_0"), val = tensor<int32, [4]>([2, 77, 1, 160])];
            tensor<bool, [4]> var_8663_end_mask_0 = const()[name = tensor<string, []>("op_8663_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_8663_cast = slice_by_index(begin = var_8663_begin_0, end = var_8663_end_0, end_mask = var_8663_end_mask_0, x = transpose_10)[name = tensor<string, []>("op_8663_cast")];
            tensor<int32, [4]> var_8667_begin_0 = const()[name = tensor<string, []>("op_8667_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_8667_end_0 = const()[name = tensor<string, []>("op_8667_end_0"), val = tensor<int32, [4]>([2, 77, 1, 240])];
            tensor<bool, [4]> var_8667_end_mask_0 = const()[name = tensor<string, []>("op_8667_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_8667_cast = slice_by_index(begin = var_8667_begin_0, end = var_8667_end_0, end_mask = var_8667_end_mask_0, x = transpose_10)[name = tensor<string, []>("op_8667_cast")];
            tensor<int32, [4]> var_8671_begin_0 = const()[name = tensor<string, []>("op_8671_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 240])];
            tensor<int32, [4]> var_8671_end_0 = const()[name = tensor<string, []>("op_8671_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_8671_end_mask_0 = const()[name = tensor<string, []>("op_8671_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_8671_cast = slice_by_index(begin = var_8671_begin_0, end = var_8671_end_0, end_mask = var_8671_end_mask_0, x = transpose_10)[name = tensor<string, []>("op_8671_cast")];
            tensor<int32, [4]> var_8675_begin_0 = const()[name = tensor<string, []>("op_8675_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_8675_end_0 = const()[name = tensor<string, []>("op_8675_end_0"), val = tensor<int32, [4]>([2, 77, 1, 400])];
            tensor<bool, [4]> var_8675_end_mask_0 = const()[name = tensor<string, []>("op_8675_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_8675_cast = slice_by_index(begin = var_8675_begin_0, end = var_8675_end_0, end_mask = var_8675_end_mask_0, x = transpose_10)[name = tensor<string, []>("op_8675_cast")];
            tensor<int32, [4]> var_8679_begin_0 = const()[name = tensor<string, []>("op_8679_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 400])];
            tensor<int32, [4]> var_8679_end_0 = const()[name = tensor<string, []>("op_8679_end_0"), val = tensor<int32, [4]>([2, 77, 1, 480])];
            tensor<bool, [4]> var_8679_end_mask_0 = const()[name = tensor<string, []>("op_8679_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_8679_cast = slice_by_index(begin = var_8679_begin_0, end = var_8679_end_0, end_mask = var_8679_end_mask_0, x = transpose_10)[name = tensor<string, []>("op_8679_cast")];
            tensor<int32, [4]> var_8683_begin_0 = const()[name = tensor<string, []>("op_8683_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 480])];
            tensor<int32, [4]> var_8683_end_0 = const()[name = tensor<string, []>("op_8683_end_0"), val = tensor<int32, [4]>([2, 77, 1, 560])];
            tensor<bool, [4]> var_8683_end_mask_0 = const()[name = tensor<string, []>("op_8683_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_8683_cast = slice_by_index(begin = var_8683_begin_0, end = var_8683_end_0, end_mask = var_8683_end_mask_0, x = transpose_10)[name = tensor<string, []>("op_8683_cast")];
            tensor<int32, [4]> var_8687_begin_0 = const()[name = tensor<string, []>("op_8687_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 560])];
            tensor<int32, [4]> var_8687_end_0 = const()[name = tensor<string, []>("op_8687_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_8687_end_mask_0 = const()[name = tensor<string, []>("op_8687_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_8687_cast = slice_by_index(begin = var_8687_begin_0, end = var_8687_end_0, end_mask = var_8687_end_mask_0, x = transpose_10)[name = tensor<string, []>("op_8687_cast")];
            tensor<int32, [4]> var_8689_begin_0 = const()[name = tensor<string, []>("op_8689_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8689_end_0 = const()[name = tensor<string, []>("op_8689_end_0"), val = tensor<int32, [4]>([2, 80, 1, 77])];
            tensor<bool, [4]> var_8689_end_mask_0 = const()[name = tensor<string, []>("op_8689_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_8689_cast = slice_by_index(begin = var_8689_begin_0, end = var_8689_end_0, end_mask = var_8689_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_8689_cast")];
            tensor<int32, [4]> var_8693_begin_0 = const()[name = tensor<string, []>("op_8693_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_8693_end_0 = const()[name = tensor<string, []>("op_8693_end_0"), val = tensor<int32, [4]>([2, 160, 1, 77])];
            tensor<bool, [4]> var_8693_end_mask_0 = const()[name = tensor<string, []>("op_8693_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_8693_cast = slice_by_index(begin = var_8693_begin_0, end = var_8693_end_0, end_mask = var_8693_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_8693_cast")];
            tensor<int32, [4]> var_8697_begin_0 = const()[name = tensor<string, []>("op_8697_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_8697_end_0 = const()[name = tensor<string, []>("op_8697_end_0"), val = tensor<int32, [4]>([2, 240, 1, 77])];
            tensor<bool, [4]> var_8697_end_mask_0 = const()[name = tensor<string, []>("op_8697_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_8697_cast = slice_by_index(begin = var_8697_begin_0, end = var_8697_end_0, end_mask = var_8697_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_8697_cast")];
            tensor<int32, [4]> var_8701_begin_0 = const()[name = tensor<string, []>("op_8701_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_8701_end_0 = const()[name = tensor<string, []>("op_8701_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_8701_end_mask_0 = const()[name = tensor<string, []>("op_8701_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_8701_cast = slice_by_index(begin = var_8701_begin_0, end = var_8701_end_0, end_mask = var_8701_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_8701_cast")];
            tensor<int32, [4]> var_8705_begin_0 = const()[name = tensor<string, []>("op_8705_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_8705_end_0 = const()[name = tensor<string, []>("op_8705_end_0"), val = tensor<int32, [4]>([2, 400, 1, 77])];
            tensor<bool, [4]> var_8705_end_mask_0 = const()[name = tensor<string, []>("op_8705_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_8705_cast = slice_by_index(begin = var_8705_begin_0, end = var_8705_end_0, end_mask = var_8705_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_8705_cast")];
            tensor<int32, [4]> var_8709_begin_0 = const()[name = tensor<string, []>("op_8709_begin_0"), val = tensor<int32, [4]>([0, 400, 0, 0])];
            tensor<int32, [4]> var_8709_end_0 = const()[name = tensor<string, []>("op_8709_end_0"), val = tensor<int32, [4]>([2, 480, 1, 77])];
            tensor<bool, [4]> var_8709_end_mask_0 = const()[name = tensor<string, []>("op_8709_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_8709_cast = slice_by_index(begin = var_8709_begin_0, end = var_8709_end_0, end_mask = var_8709_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_8709_cast")];
            tensor<int32, [4]> var_8713_begin_0 = const()[name = tensor<string, []>("op_8713_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_8713_end_0 = const()[name = tensor<string, []>("op_8713_end_0"), val = tensor<int32, [4]>([2, 560, 1, 77])];
            tensor<bool, [4]> var_8713_end_mask_0 = const()[name = tensor<string, []>("op_8713_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_8713_cast = slice_by_index(begin = var_8713_begin_0, end = var_8713_end_0, end_mask = var_8713_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_8713_cast")];
            tensor<int32, [4]> var_8717_begin_0 = const()[name = tensor<string, []>("op_8717_begin_0"), val = tensor<int32, [4]>([0, 560, 0, 0])];
            tensor<int32, [4]> var_8717_end_0 = const()[name = tensor<string, []>("op_8717_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_8717_end_mask_0 = const()[name = tensor<string, []>("op_8717_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_8717_cast = slice_by_index(begin = var_8717_begin_0, end = var_8717_end_0, end_mask = var_8717_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_8717_cast")];
            tensor<string, []> var_8721_equation_0 = const()[name = tensor<string, []>("op_8721_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_8721_cast = einsum(equation = var_8721_equation_0, values = (var_8659_cast, var_8639_cast))[name = tensor<string, []>("op_8721_cast")];
            tensor<fp16, []> var_8722_to_fp16 = const()[name = tensor<string, []>("op_8722_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_673_cast = mul(x = var_8721_cast, y = var_8722_to_fp16)[name = tensor<string, []>("aw_chunk_673_cast")];
            tensor<string, []> var_8725_equation_0 = const()[name = tensor<string, []>("op_8725_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_8725_cast = einsum(equation = var_8725_equation_0, values = (var_8659_cast, var_8640_cast))[name = tensor<string, []>("op_8725_cast")];
            tensor<fp16, []> var_8726_to_fp16 = const()[name = tensor<string, []>("op_8726_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_675_cast = mul(x = var_8725_cast, y = var_8726_to_fp16)[name = tensor<string, []>("aw_chunk_675_cast")];
            tensor<string, []> var_8729_equation_0 = const()[name = tensor<string, []>("op_8729_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_8729_cast = einsum(equation = var_8729_equation_0, values = (var_8663_cast, var_8641_cast))[name = tensor<string, []>("op_8729_cast")];
            tensor<fp16, []> var_8730_to_fp16 = const()[name = tensor<string, []>("op_8730_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_677_cast = mul(x = var_8729_cast, y = var_8730_to_fp16)[name = tensor<string, []>("aw_chunk_677_cast")];
            tensor<string, []> var_8733_equation_0 = const()[name = tensor<string, []>("op_8733_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_8733_cast = einsum(equation = var_8733_equation_0, values = (var_8663_cast, var_8642_cast))[name = tensor<string, []>("op_8733_cast")];
            tensor<fp16, []> var_8734_to_fp16 = const()[name = tensor<string, []>("op_8734_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_679_cast = mul(x = var_8733_cast, y = var_8734_to_fp16)[name = tensor<string, []>("aw_chunk_679_cast")];
            tensor<string, []> var_8737_equation_0 = const()[name = tensor<string, []>("op_8737_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_8737_cast = einsum(equation = var_8737_equation_0, values = (var_8667_cast, var_8643_cast))[name = tensor<string, []>("op_8737_cast")];
            tensor<fp16, []> var_8738_to_fp16 = const()[name = tensor<string, []>("op_8738_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_681_cast = mul(x = var_8737_cast, y = var_8738_to_fp16)[name = tensor<string, []>("aw_chunk_681_cast")];
            tensor<string, []> var_8741_equation_0 = const()[name = tensor<string, []>("op_8741_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_8741_cast = einsum(equation = var_8741_equation_0, values = (var_8667_cast, var_8644_cast))[name = tensor<string, []>("op_8741_cast")];
            tensor<fp16, []> var_8742_to_fp16 = const()[name = tensor<string, []>("op_8742_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_683_cast = mul(x = var_8741_cast, y = var_8742_to_fp16)[name = tensor<string, []>("aw_chunk_683_cast")];
            tensor<string, []> var_8745_equation_0 = const()[name = tensor<string, []>("op_8745_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_8745_cast = einsum(equation = var_8745_equation_0, values = (var_8671_cast, var_8645_cast))[name = tensor<string, []>("op_8745_cast")];
            tensor<fp16, []> var_8746_to_fp16 = const()[name = tensor<string, []>("op_8746_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_685_cast = mul(x = var_8745_cast, y = var_8746_to_fp16)[name = tensor<string, []>("aw_chunk_685_cast")];
            tensor<string, []> var_8749_equation_0 = const()[name = tensor<string, []>("op_8749_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_8749_cast = einsum(equation = var_8749_equation_0, values = (var_8671_cast, var_8646_cast))[name = tensor<string, []>("op_8749_cast")];
            tensor<fp16, []> var_8750_to_fp16 = const()[name = tensor<string, []>("op_8750_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_687_cast = mul(x = var_8749_cast, y = var_8750_to_fp16)[name = tensor<string, []>("aw_chunk_687_cast")];
            tensor<string, []> var_8753_equation_0 = const()[name = tensor<string, []>("op_8753_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_8753_cast = einsum(equation = var_8753_equation_0, values = (var_8675_cast, var_8647_cast))[name = tensor<string, []>("op_8753_cast")];
            tensor<fp16, []> var_8754_to_fp16 = const()[name = tensor<string, []>("op_8754_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_689_cast = mul(x = var_8753_cast, y = var_8754_to_fp16)[name = tensor<string, []>("aw_chunk_689_cast")];
            tensor<string, []> var_8757_equation_0 = const()[name = tensor<string, []>("op_8757_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_8757_cast = einsum(equation = var_8757_equation_0, values = (var_8675_cast, var_8648_cast))[name = tensor<string, []>("op_8757_cast")];
            tensor<fp16, []> var_8758_to_fp16 = const()[name = tensor<string, []>("op_8758_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_691_cast = mul(x = var_8757_cast, y = var_8758_to_fp16)[name = tensor<string, []>("aw_chunk_691_cast")];
            tensor<string, []> var_8761_equation_0 = const()[name = tensor<string, []>("op_8761_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_8761_cast = einsum(equation = var_8761_equation_0, values = (var_8679_cast, var_8649_cast))[name = tensor<string, []>("op_8761_cast")];
            tensor<fp16, []> var_8762_to_fp16 = const()[name = tensor<string, []>("op_8762_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_693_cast = mul(x = var_8761_cast, y = var_8762_to_fp16)[name = tensor<string, []>("aw_chunk_693_cast")];
            tensor<string, []> var_8765_equation_0 = const()[name = tensor<string, []>("op_8765_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_8765_cast = einsum(equation = var_8765_equation_0, values = (var_8679_cast, var_8650_cast))[name = tensor<string, []>("op_8765_cast")];
            tensor<fp16, []> var_8766_to_fp16 = const()[name = tensor<string, []>("op_8766_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_695_cast = mul(x = var_8765_cast, y = var_8766_to_fp16)[name = tensor<string, []>("aw_chunk_695_cast")];
            tensor<string, []> var_8769_equation_0 = const()[name = tensor<string, []>("op_8769_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_8769_cast = einsum(equation = var_8769_equation_0, values = (var_8683_cast, var_8651_cast))[name = tensor<string, []>("op_8769_cast")];
            tensor<fp16, []> var_8770_to_fp16 = const()[name = tensor<string, []>("op_8770_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_697_cast = mul(x = var_8769_cast, y = var_8770_to_fp16)[name = tensor<string, []>("aw_chunk_697_cast")];
            tensor<string, []> var_8773_equation_0 = const()[name = tensor<string, []>("op_8773_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_8773_cast = einsum(equation = var_8773_equation_0, values = (var_8683_cast, var_8652_cast))[name = tensor<string, []>("op_8773_cast")];
            tensor<fp16, []> var_8774_to_fp16 = const()[name = tensor<string, []>("op_8774_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_699_cast = mul(x = var_8773_cast, y = var_8774_to_fp16)[name = tensor<string, []>("aw_chunk_699_cast")];
            tensor<string, []> var_8777_equation_0 = const()[name = tensor<string, []>("op_8777_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_8777_cast = einsum(equation = var_8777_equation_0, values = (var_8687_cast, var_8653_cast))[name = tensor<string, []>("op_8777_cast")];
            tensor<fp16, []> var_8778_to_fp16 = const()[name = tensor<string, []>("op_8778_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_701_cast = mul(x = var_8777_cast, y = var_8778_to_fp16)[name = tensor<string, []>("aw_chunk_701_cast")];
            tensor<string, []> var_8781_equation_0 = const()[name = tensor<string, []>("op_8781_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_8781_cast = einsum(equation = var_8781_equation_0, values = (var_8687_cast, var_8654_cast))[name = tensor<string, []>("op_8781_cast")];
            tensor<fp16, []> var_8782_to_fp16 = const()[name = tensor<string, []>("op_8782_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_703_cast = mul(x = var_8781_cast, y = var_8782_to_fp16)[name = tensor<string, []>("aw_chunk_703_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_8784_cast = softmax(axis = var_8182, x = aw_chunk_673_cast)[name = tensor<string, []>("op_8784_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_8785_cast = softmax(axis = var_8182, x = aw_chunk_675_cast)[name = tensor<string, []>("op_8785_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_8786_cast = softmax(axis = var_8182, x = aw_chunk_677_cast)[name = tensor<string, []>("op_8786_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_8787_cast = softmax(axis = var_8182, x = aw_chunk_679_cast)[name = tensor<string, []>("op_8787_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_8788_cast = softmax(axis = var_8182, x = aw_chunk_681_cast)[name = tensor<string, []>("op_8788_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_8789_cast = softmax(axis = var_8182, x = aw_chunk_683_cast)[name = tensor<string, []>("op_8789_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_8790_cast = softmax(axis = var_8182, x = aw_chunk_685_cast)[name = tensor<string, []>("op_8790_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_8791_cast = softmax(axis = var_8182, x = aw_chunk_687_cast)[name = tensor<string, []>("op_8791_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_8792_cast = softmax(axis = var_8182, x = aw_chunk_689_cast)[name = tensor<string, []>("op_8792_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_8793_cast = softmax(axis = var_8182, x = aw_chunk_691_cast)[name = tensor<string, []>("op_8793_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_8794_cast = softmax(axis = var_8182, x = aw_chunk_693_cast)[name = tensor<string, []>("op_8794_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_8795_cast = softmax(axis = var_8182, x = aw_chunk_695_cast)[name = tensor<string, []>("op_8795_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_8796_cast = softmax(axis = var_8182, x = aw_chunk_697_cast)[name = tensor<string, []>("op_8796_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_8797_cast = softmax(axis = var_8182, x = aw_chunk_699_cast)[name = tensor<string, []>("op_8797_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_8798_cast = softmax(axis = var_8182, x = aw_chunk_701_cast)[name = tensor<string, []>("op_8798_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_8799_cast = softmax(axis = var_8182, x = aw_chunk_703_cast)[name = tensor<string, []>("op_8799_cast")];
            tensor<string, []> var_8801_equation_0 = const()[name = tensor<string, []>("op_8801_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8801_cast = einsum(equation = var_8801_equation_0, values = (var_8689_cast, var_8784_cast))[name = tensor<string, []>("op_8801_cast")];
            tensor<string, []> var_8803_equation_0 = const()[name = tensor<string, []>("op_8803_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8803_cast = einsum(equation = var_8803_equation_0, values = (var_8689_cast, var_8785_cast))[name = tensor<string, []>("op_8803_cast")];
            tensor<string, []> var_8805_equation_0 = const()[name = tensor<string, []>("op_8805_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8805_cast = einsum(equation = var_8805_equation_0, values = (var_8693_cast, var_8786_cast))[name = tensor<string, []>("op_8805_cast")];
            tensor<string, []> var_8807_equation_0 = const()[name = tensor<string, []>("op_8807_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8807_cast = einsum(equation = var_8807_equation_0, values = (var_8693_cast, var_8787_cast))[name = tensor<string, []>("op_8807_cast")];
            tensor<string, []> var_8809_equation_0 = const()[name = tensor<string, []>("op_8809_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8809_cast = einsum(equation = var_8809_equation_0, values = (var_8697_cast, var_8788_cast))[name = tensor<string, []>("op_8809_cast")];
            tensor<string, []> var_8811_equation_0 = const()[name = tensor<string, []>("op_8811_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8811_cast = einsum(equation = var_8811_equation_0, values = (var_8697_cast, var_8789_cast))[name = tensor<string, []>("op_8811_cast")];
            tensor<string, []> var_8813_equation_0 = const()[name = tensor<string, []>("op_8813_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8813_cast = einsum(equation = var_8813_equation_0, values = (var_8701_cast, var_8790_cast))[name = tensor<string, []>("op_8813_cast")];
            tensor<string, []> var_8815_equation_0 = const()[name = tensor<string, []>("op_8815_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8815_cast = einsum(equation = var_8815_equation_0, values = (var_8701_cast, var_8791_cast))[name = tensor<string, []>("op_8815_cast")];
            tensor<string, []> var_8817_equation_0 = const()[name = tensor<string, []>("op_8817_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8817_cast = einsum(equation = var_8817_equation_0, values = (var_8705_cast, var_8792_cast))[name = tensor<string, []>("op_8817_cast")];
            tensor<string, []> var_8819_equation_0 = const()[name = tensor<string, []>("op_8819_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8819_cast = einsum(equation = var_8819_equation_0, values = (var_8705_cast, var_8793_cast))[name = tensor<string, []>("op_8819_cast")];
            tensor<string, []> var_8821_equation_0 = const()[name = tensor<string, []>("op_8821_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8821_cast = einsum(equation = var_8821_equation_0, values = (var_8709_cast, var_8794_cast))[name = tensor<string, []>("op_8821_cast")];
            tensor<string, []> var_8823_equation_0 = const()[name = tensor<string, []>("op_8823_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8823_cast = einsum(equation = var_8823_equation_0, values = (var_8709_cast, var_8795_cast))[name = tensor<string, []>("op_8823_cast")];
            tensor<string, []> var_8825_equation_0 = const()[name = tensor<string, []>("op_8825_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8825_cast = einsum(equation = var_8825_equation_0, values = (var_8713_cast, var_8796_cast))[name = tensor<string, []>("op_8825_cast")];
            tensor<string, []> var_8827_equation_0 = const()[name = tensor<string, []>("op_8827_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8827_cast = einsum(equation = var_8827_equation_0, values = (var_8713_cast, var_8797_cast))[name = tensor<string, []>("op_8827_cast")];
            tensor<string, []> var_8829_equation_0 = const()[name = tensor<string, []>("op_8829_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8829_cast = einsum(equation = var_8829_equation_0, values = (var_8717_cast, var_8798_cast))[name = tensor<string, []>("op_8829_cast")];
            tensor<string, []> var_8831_equation_0 = const()[name = tensor<string, []>("op_8831_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_8831_cast = einsum(equation = var_8831_equation_0, values = (var_8717_cast, var_8799_cast))[name = tensor<string, []>("op_8831_cast")];
            tensor<bool, []> var_8833_interleave_0 = const()[name = tensor<string, []>("op_8833_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_8833_cast = concat(axis = var_8160, interleave = var_8833_interleave_0, values = (var_8801_cast, var_8803_cast))[name = tensor<string, []>("op_8833_cast")];
            tensor<bool, []> var_8835_interleave_0 = const()[name = tensor<string, []>("op_8835_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_8835_cast = concat(axis = var_8160, interleave = var_8835_interleave_0, values = (var_8805_cast, var_8807_cast))[name = tensor<string, []>("op_8835_cast")];
            tensor<bool, []> var_8837_interleave_0 = const()[name = tensor<string, []>("op_8837_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_8837_cast = concat(axis = var_8160, interleave = var_8837_interleave_0, values = (var_8809_cast, var_8811_cast))[name = tensor<string, []>("op_8837_cast")];
            tensor<bool, []> var_8839_interleave_0 = const()[name = tensor<string, []>("op_8839_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_8839_cast = concat(axis = var_8160, interleave = var_8839_interleave_0, values = (var_8813_cast, var_8815_cast))[name = tensor<string, []>("op_8839_cast")];
            tensor<bool, []> var_8841_interleave_0 = const()[name = tensor<string, []>("op_8841_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_8841_cast = concat(axis = var_8160, interleave = var_8841_interleave_0, values = (var_8817_cast, var_8819_cast))[name = tensor<string, []>("op_8841_cast")];
            tensor<bool, []> var_8843_interleave_0 = const()[name = tensor<string, []>("op_8843_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_8843_cast = concat(axis = var_8160, interleave = var_8843_interleave_0, values = (var_8821_cast, var_8823_cast))[name = tensor<string, []>("op_8843_cast")];
            tensor<bool, []> var_8845_interleave_0 = const()[name = tensor<string, []>("op_8845_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_8845_cast = concat(axis = var_8160, interleave = var_8845_interleave_0, values = (var_8825_cast, var_8827_cast))[name = tensor<string, []>("op_8845_cast")];
            tensor<bool, []> var_8847_interleave_0 = const()[name = tensor<string, []>("op_8847_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_8847_cast = concat(axis = var_8160, interleave = var_8847_interleave_0, values = (var_8829_cast, var_8831_cast))[name = tensor<string, []>("op_8847_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_8182, interleave = input_385_interleave_0, values = (var_8833_cast, var_8835_cast, var_8837_cast, var_8839_cast, var_8841_cast, var_8843_cast, var_8845_cast, var_8847_cast))[name = tensor<string, []>("input_385_cast")];
            tensor<int32, [2]> var_8853 = const()[name = tensor<string, []>("op_8853"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8855 = const()[name = tensor<string, []>("op_8855"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_8857_pad_type_0 = const()[name = tensor<string, []>("op_8857_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_8857_pad_0 = const()[name = tensor<string, []>("op_8857_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(592571264))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(592878528))), name = tensor<string, []>("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            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, []>(592878720)))];
            tensor<fp16, [2, 640, 1, 1024]> var_8857_cast = conv(bias = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_8855, groups = var_8182, pad = var_8857_pad_0, pad_type = var_8857_pad_type_0, strides = var_8853, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_385_cast)[name = tensor<string, []>("op_8857_cast")];
            tensor<fp16, [2, 640, 1, 1024]> inputs_65_cast = add(x = var_8857_cast, y = inputs_63_cast)[name = tensor<string, []>("inputs_65_cast")];
            tensor<int32, [1]> var_8861 = const()[name = tensor<string, []>("op_8861"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_65_cast = reduce_mean(axes = var_8861, keep_dims = var_8177, x = inputs_65_cast)[name = tensor<string, []>("channels_mean_65_cast")];
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_65_cast = sub(x = inputs_65_cast, y = channels_mean_65_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_65_cast")];
            tensor<int32, [1]> var_8865 = const()[name = tensor<string, []>("op_8865"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_8866_cast = reduce_mean(axes = var_8865, keep_dims = var_8177, x = zero_mean_sq_65_cast)[name = tensor<string, []>("op_8866_cast")];
            tensor<fp16, []> var_8867_to_fp16 = const()[name = tensor<string, []>("op_8867_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_8868_cast = add(x = var_8866_cast, y = var_8867_to_fp16)[name = tensor<string, []>("op_8868_cast")];
            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_8868_cast)[name = tensor<string, []>("denom_65_cast")];
            tensor<fp16, [2, 640, 1, 1024]> out_65_cast = mul(x = zero_mean_65_cast, y = denom_65_cast)[name = tensor<string, []>("out_65_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_8872_to_fp16 = const()[name = tensor<string, []>("op_8872_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(592880064)))];
            tensor<fp16, [2, 640, 1, 1024]> var_8873_cast = add(x = out_65_cast, y = var_8872_to_fp16)[name = tensor<string, []>("op_8873_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_8875_to_fp16 = const()[name = tensor<string, []>("op_8875_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(592881408)))];
            tensor<fp16, [2, 640, 1, 1024]> input_387_cast = mul(x = var_8873_cast, y = var_8875_to_fp16)[name = tensor<string, []>("input_387_cast")];
            tensor<int32, [2]> var_8883 = const()[name = tensor<string, []>("op_8883"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8885 = const()[name = tensor<string, []>("op_8885"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_8887_pad_type_0 = const()[name = tensor<string, []>("op_8887_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_8887_pad_0 = const()[name = tensor<string, []>("op_8887_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2457600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(592882752))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(595340416))), name = tensor<string, []>("up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([5120, 640, 1, 1])];
            tensor<fp16, [5120]> up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3840]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(595340608))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(595344512))), name = tensor<string, []>("up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([5120])];
            tensor<fp16, [2, 5120, 1, 1024]> var_8887_cast = conv(bias = up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_8885, groups = var_8182, pad = var_8887_pad_0, pad_type = var_8887_pad_type_0, strides = var_8883, weight = up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_387_cast)[name = tensor<string, []>("op_8887_cast")];
            tensor<int32, [2]> var_8888_split_sizes_0 = const()[name = tensor<string, []>("op_8888_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_8888_axis_0 = const()[name = tensor<string, []>("op_8888_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 1024]> var_8888_cast_0, tensor<fp16, [2, 2560, 1, 1024]> var_8888_cast_1 = split(axis = var_8888_axis_0, split_sizes = var_8888_split_sizes_0, x = var_8887_cast)[name = tensor<string, []>("op_8888_cast")];
            tensor<string, []> var_8890_mode_0 = const()[name = tensor<string, []>("op_8890_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 1024]> var_8890_cast = gelu(mode = var_8890_mode_0, x = var_8888_cast_1)[name = tensor<string, []>("op_8890_cast")];
            tensor<fp16, [2, 2560, 1, 1024]> input_389_cast = mul(x = var_8888_cast_0, y = var_8890_cast)[name = tensor<string, []>("input_389_cast")];
            tensor<int32, [2]> var_8894 = const()[name = tensor<string, []>("op_8894"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8896 = const()[name = tensor<string, []>("op_8896"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_8898_pad_type_0 = const()[name = tensor<string, []>("op_8898_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_8898_pad_0 = const()[name = tensor<string, []>("op_8898_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(595344704))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(596573568))), name = tensor<string, []>("up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 2560, 1, 1])];
            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, []>(596573760)))];
            tensor<fp16, [2, 640, 1, 1024]> var_8898_cast = conv(bias = up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_8896, groups = var_8182, pad = var_8898_pad_0, pad_type = var_8898_pad_type_0, strides = var_8894, weight = up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_389_cast)[name = tensor<string, []>("op_8898_cast")];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_233_cast = add(x = var_8898_cast, y = inputs_65_cast)[name = tensor<string, []>("hidden_states_233_cast")];
            tensor<int32, [4]> var_8900 = const()[name = tensor<string, []>("op_8900"), val = tensor<int32, [4]>([2, 640, 32, 32])];
            tensor<fp16, [2, 640, 32, 32]> input_391_cast = reshape(shape = var_8900, x = hidden_states_233_cast)[name = tensor<string, []>("input_391_cast")];
            tensor<int32, [2]> var_8904 = const()[name = tensor<string, []>("op_8904"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8906 = const()[name = tensor<string, []>("op_8906"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(596575104))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(596882368))), name = tensor<string, []>("up_blocks_2_attentions_0_proj_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            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, []>(596882560)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_235_cast = conv(bias = up_blocks_2_attentions_0_proj_out_bias_to_fp16, dilations = var_8906, groups = var_8182, pad = hidden_states_235_pad_0, pad_type = hidden_states_235_pad_type_0, strides = var_8904, weight = up_blocks_2_attentions_0_proj_out_weight_to_fp16_palettized, x = input_391_cast)[name = tensor<string, []>("hidden_states_235_cast")];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_237_cast = add(x = hidden_states_235_cast, y = hidden_states_223_cast)[name = tensor<string, []>("hidden_states_237_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_8182, interleave = input_393_interleave_0, values = (hidden_states_237_cast, input_89_cast))[name = tensor<string, []>("input_393_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)[name = tensor<string, []>("reshape_180_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)[name = tensor<string, []>("reduce_mean_135_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> sub_90_cast = sub(x = reshape_180_cast, y = reduce_mean_135_cast)[name = tensor<string, []>("sub_90_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> square_45_cast = square(x = sub_90_cast)[name = tensor<string, []>("square_45_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)[name = tensor<string, []>("reduce_mean_137_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)[name = tensor<string, []>("add_90_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_45_cast = sqrt(x = add_90_cast)[name = tensor<string, []>("sqrt_45_cast")];
            tensor<fp16, [2, 32, 40, 32, 32]> real_div_45_cast = real_div(x = sub_90_cast, y = sqrt_45_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_181_cast")];
            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, []>(596883904)))];
            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, []>(596886528)))];
            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_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_181_cast)[name = tensor<string, []>("add_91_cast")];
            tensor<fp16, [2, 1280, 32, 32]> input_397_cast = silu(x = add_91_cast)[name = tensor<string, []>("input_397_cast")];
            tensor<int32, [2]> var_8924 = const()[name = tensor<string, []>("op_8924"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8926 = const()[name = tensor<string, []>("op_8926"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5529600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(596889152))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(602418816))), name = tensor<string, []>("up_blocks_2_resnets_1_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 1280, 3, 3])];
            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, []>(602419008)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_239_cast = conv(bias = up_blocks_2_resnets_1_conv1_bias_to_fp16, dilations = var_8926, groups = var_8182, pad = hidden_states_239_pad_0, pad_type = hidden_states_239_pad_type_0, strides = var_8924, weight = up_blocks_2_resnets_1_conv1_weight_to_fp16_palettized, x = input_397_cast)[name = tensor<string, []>("hidden_states_239_cast")];
            tensor<int32, [2]> var_8932 = const()[name = tensor<string, []>("op_8932"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8934 = const()[name = tensor<string, []>("op_8934"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [614400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(602420352))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(603034816))), name = tensor<string, []>("up_blocks_2_resnets_1_time_emb_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 1280, 1, 1])];
            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, []>(603035008)))];
            tensor<fp16, [2, 640, 1, 1]> temb_35_cast = conv(bias = up_blocks_2_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_8934, groups = var_8182, pad = temb_35_pad_0, pad_type = temb_35_pad_type_0, strides = var_8932, weight = up_blocks_2_resnets_1_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor<string, []>("temb_35_cast")];
            tensor<fp16, [2, 640, 32, 32]> input_401_cast = add(x = hidden_states_239_cast, y = temb_35_cast)[name = tensor<string, []>("input_401_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)[name = tensor<string, []>("reshape_184_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)[name = tensor<string, []>("reduce_mean_138_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> sub_92_cast = sub(x = reshape_184_cast, y = reduce_mean_138_cast)[name = tensor<string, []>("sub_92_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> square_46_cast = square(x = sub_92_cast)[name = tensor<string, []>("square_46_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)[name = tensor<string, []>("reduce_mean_140_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)[name = tensor<string, []>("add_92_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_46_cast = sqrt(x = add_92_cast)[name = tensor<string, []>("sqrt_46_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> real_div_46_cast = real_div(x = sub_92_cast, y = sqrt_46_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_185_cast")];
            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, []>(603036352)))];
            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, []>(603037696)))];
            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_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_185_cast)[name = tensor<string, []>("add_93_cast")];
            tensor<fp16, [2, 640, 32, 32]> input_405_cast = silu(x = add_93_cast)[name = tensor<string, []>("input_405_cast")];
            tensor<int32, [2]> var_8944 = const()[name = tensor<string, []>("op_8944"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8946 = const()[name = tensor<string, []>("op_8946"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2764800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(603039040))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(605803904))), name = tensor<string, []>("up_blocks_2_resnets_1_conv2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 3, 3])];
            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, []>(605804096)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_241_cast = conv(bias = up_blocks_2_resnets_1_conv2_bias_to_fp16, dilations = var_8946, groups = var_8182, pad = hidden_states_241_pad_0, pad_type = hidden_states_241_pad_type_0, strides = var_8944, weight = up_blocks_2_resnets_1_conv2_weight_to_fp16_palettized, x = input_405_cast)[name = tensor<string, []>("hidden_states_241_cast")];
            tensor<int32, [2]> var_8951 = const()[name = tensor<string, []>("op_8951"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8953 = const()[name = tensor<string, []>("op_8953"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [614400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(605805440))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(606419904))), name = tensor<string, []>("up_blocks_2_resnets_1_conv_shortcut_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 1280, 1, 1])];
            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, []>(606420096)))];
            tensor<fp16, [2, 640, 32, 32]> x_19_cast = conv(bias = up_blocks_2_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_8953, groups = var_8182, pad = x_19_pad_0, pad_type = x_19_pad_type_0, strides = var_8951, weight = up_blocks_2_resnets_1_conv_shortcut_weight_to_fp16_palettized, x = input_393_cast)[name = tensor<string, []>("x_19_cast")];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_243_cast = add(x = x_19_cast, y = hidden_states_241_cast)[name = tensor<string, []>("hidden_states_243_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)[name = tensor<string, []>("reshape_188_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)[name = tensor<string, []>("reduce_mean_141_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> sub_94_cast = sub(x = reshape_188_cast, y = reduce_mean_141_cast)[name = tensor<string, []>("sub_94_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> square_47_cast = square(x = sub_94_cast)[name = tensor<string, []>("square_47_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)[name = tensor<string, []>("reduce_mean_143_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)[name = tensor<string, []>("add_94_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_47_cast = sqrt(x = add_94_cast)[name = tensor<string, []>("sqrt_47_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> real_div_47_cast = real_div(x = sub_94_cast, y = sqrt_47_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_189_cast")];
            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, []>(606421440)))];
            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, []>(606422784)))];
            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_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_189_cast)[name = tensor<string, []>("add_95_cast")];
            tensor<int32, [2]> var_8973 = const()[name = tensor<string, []>("op_8973"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8975 = const()[name = tensor<string, []>("op_8975"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(606424128))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(606731392))), name = tensor<string, []>("up_blocks_2_attentions_1_proj_in_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            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, []>(606731584)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_245_cast = conv(bias = up_blocks_2_attentions_1_proj_in_bias_to_fp16, dilations = var_8975, groups = var_8182, pad = hidden_states_245_pad_0, pad_type = hidden_states_245_pad_type_0, strides = var_8973, weight = up_blocks_2_attentions_1_proj_in_weight_to_fp16_palettized, x = add_95_cast)[name = tensor<string, []>("hidden_states_245_cast")];
            tensor<int32, [4]> var_8980 = const()[name = tensor<string, []>("op_8980"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<fp16, [2, 640, 1, 1024]> inputs_67_cast = reshape(shape = var_8980, x = hidden_states_245_cast)[name = tensor<string, []>("inputs_67_cast")];
            tensor<int32, [1]> var_8990 = const()[name = tensor<string, []>("op_8990"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_67_cast = reduce_mean(axes = var_8990, keep_dims = var_8177, x = inputs_67_cast)[name = tensor<string, []>("channels_mean_67_cast")];
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_67_cast = sub(x = inputs_67_cast, y = channels_mean_67_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_67_cast")];
            tensor<int32, [1]> var_8994 = const()[name = tensor<string, []>("op_8994"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_8995_cast = reduce_mean(axes = var_8994, keep_dims = var_8177, x = zero_mean_sq_67_cast)[name = tensor<string, []>("op_8995_cast")];
            tensor<fp16, []> var_8996_to_fp16 = const()[name = tensor<string, []>("op_8996_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_8997_cast = add(x = var_8995_cast, y = var_8996_to_fp16)[name = tensor<string, []>("op_8997_cast")];
            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_8997_cast)[name = tensor<string, []>("denom_67_cast")];
            tensor<fp16, [2, 640, 1, 1024]> out_67_cast = mul(x = zero_mean_67_cast, y = denom_67_cast)[name = tensor<string, []>("out_67_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_9001_to_fp16 = const()[name = tensor<string, []>("op_9001_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(606732928)))];
            tensor<fp16, [2, 640, 1, 1024]> var_9002_cast = add(x = out_67_cast, y = var_9001_to_fp16)[name = tensor<string, []>("op_9002_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_9004_to_fp16 = const()[name = tensor<string, []>("op_9004_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(606734272)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_247_cast = mul(x = var_9002_cast, y = var_9004_to_fp16)[name = tensor<string, []>("hidden_states_247_cast")];
            tensor<int32, [2]> var_9011 = const()[name = tensor<string, []>("op_9011"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9013 = const()[name = tensor<string, []>("op_9013"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(606735616))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(607042880))), name = tensor<string, []>("up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 1024]> q_45_cast = conv(dilations = var_9013, groups = var_8182, pad = q_45_pad_0, pad_type = q_45_pad_type_0, strides = var_9011, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_247_cast)[name = tensor<string, []>("q_45_cast")];
            tensor<int32, [2]> var_9017 = const()[name = tensor<string, []>("op_9017"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9019 = const()[name = tensor<string, []>("op_9019"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(607043072))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(607350336))), name = tensor<string, []>("up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 1024]> k_89_cast = conv(dilations = var_9019, groups = var_8182, pad = k_89_pad_0, pad_type = k_89_pad_type_0, strides = var_9017, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_247_cast)[name = tensor<string, []>("k_89_cast")];
            tensor<int32, [2]> var_9023 = const()[name = tensor<string, []>("op_9023"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9025 = const()[name = tensor<string, []>("op_9025"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(607350528))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(607657792))), name = tensor<string, []>("up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 1024]> v_45_cast = conv(dilations = var_9025, groups = var_8182, pad = v_45_pad_0, pad_type = v_45_pad_type_0, strides = var_9023, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_247_cast)[name = tensor<string, []>("v_45_cast")];
            tensor<int32, [4]> var_9029_begin_0 = const()[name = tensor<string, []>("op_9029_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9029_end_0 = const()[name = tensor<string, []>("op_9029_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9029_end_mask_0 = const()[name = tensor<string, []>("op_9029_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9029_cast = slice_by_index(begin = var_9029_begin_0, end = var_9029_end_0, end_mask = var_9029_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_9029_cast")];
            tensor<int32, [4]> var_9033_begin_0 = const()[name = tensor<string, []>("op_9033_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_9033_end_0 = const()[name = tensor<string, []>("op_9033_end_0"), val = tensor<int32, [4]>([2, 160, 1, 1024])];
            tensor<bool, [4]> var_9033_end_mask_0 = const()[name = tensor<string, []>("op_9033_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9033_cast = slice_by_index(begin = var_9033_begin_0, end = var_9033_end_0, end_mask = var_9033_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_9033_cast")];
            tensor<int32, [4]> var_9037_begin_0 = const()[name = tensor<string, []>("op_9037_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_9037_end_0 = const()[name = tensor<string, []>("op_9037_end_0"), val = tensor<int32, [4]>([2, 240, 1, 1024])];
            tensor<bool, [4]> var_9037_end_mask_0 = const()[name = tensor<string, []>("op_9037_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9037_cast = slice_by_index(begin = var_9037_begin_0, end = var_9037_end_0, end_mask = var_9037_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_9037_cast")];
            tensor<int32, [4]> var_9041_begin_0 = const()[name = tensor<string, []>("op_9041_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_9041_end_0 = const()[name = tensor<string, []>("op_9041_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_9041_end_mask_0 = const()[name = tensor<string, []>("op_9041_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9041_cast = slice_by_index(begin = var_9041_begin_0, end = var_9041_end_0, end_mask = var_9041_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_9041_cast")];
            tensor<int32, [4]> var_9045_begin_0 = const()[name = tensor<string, []>("op_9045_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_9045_end_0 = const()[name = tensor<string, []>("op_9045_end_0"), val = tensor<int32, [4]>([2, 400, 1, 1024])];
            tensor<bool, [4]> var_9045_end_mask_0 = const()[name = tensor<string, []>("op_9045_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9045_cast = slice_by_index(begin = var_9045_begin_0, end = var_9045_end_0, end_mask = var_9045_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_9045_cast")];
            tensor<int32, [4]> var_9049_begin_0 = const()[name = tensor<string, []>("op_9049_begin_0"), val = tensor<int32, [4]>([0, 400, 0, 0])];
            tensor<int32, [4]> var_9049_end_0 = const()[name = tensor<string, []>("op_9049_end_0"), val = tensor<int32, [4]>([2, 480, 1, 1024])];
            tensor<bool, [4]> var_9049_end_mask_0 = const()[name = tensor<string, []>("op_9049_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9049_cast = slice_by_index(begin = var_9049_begin_0, end = var_9049_end_0, end_mask = var_9049_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_9049_cast")];
            tensor<int32, [4]> var_9053_begin_0 = const()[name = tensor<string, []>("op_9053_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_9053_end_0 = const()[name = tensor<string, []>("op_9053_end_0"), val = tensor<int32, [4]>([2, 560, 1, 1024])];
            tensor<bool, [4]> var_9053_end_mask_0 = const()[name = tensor<string, []>("op_9053_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9053_cast = slice_by_index(begin = var_9053_begin_0, end = var_9053_end_0, end_mask = var_9053_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_9053_cast")];
            tensor<int32, [4]> var_9057_begin_0 = const()[name = tensor<string, []>("op_9057_begin_0"), val = tensor<int32, [4]>([0, 560, 0, 0])];
            tensor<int32, [4]> var_9057_end_0 = const()[name = tensor<string, []>("op_9057_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_9057_end_mask_0 = const()[name = tensor<string, []>("op_9057_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9057_cast = slice_by_index(begin = var_9057_begin_0, end = var_9057_end_0, end_mask = var_9057_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_9057_cast")];
            tensor<int32, [4]> var_9060_begin_0 = const()[name = tensor<string, []>("op_9060_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9060_end_0 = const()[name = tensor<string, []>("op_9060_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_9060_end_mask_0 = const()[name = tensor<string, []>("op_9060_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9060_cast = slice_by_index(begin = var_9060_begin_0, end = var_9060_end_0, end_mask = var_9060_end_mask_0, x = var_9029_cast)[name = tensor<string, []>("op_9060_cast")];
            tensor<int32, [4]> var_9061_begin_0 = const()[name = tensor<string, []>("op_9061_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9061_end_0 = const()[name = tensor<string, []>("op_9061_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9061_end_mask_0 = const()[name = tensor<string, []>("op_9061_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9061_cast = slice_by_index(begin = var_9061_begin_0, end = var_9061_end_0, end_mask = var_9061_end_mask_0, x = var_9029_cast)[name = tensor<string, []>("op_9061_cast")];
            tensor<int32, [4]> var_9062_begin_0 = const()[name = tensor<string, []>("op_9062_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9062_end_0 = const()[name = tensor<string, []>("op_9062_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            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, 80, 1, 512]> var_9062_cast = slice_by_index(begin = var_9062_begin_0, end = var_9062_end_0, end_mask = var_9062_end_mask_0, x = var_9033_cast)[name = tensor<string, []>("op_9062_cast")];
            tensor<int32, [4]> var_9063_begin_0 = const()[name = tensor<string, []>("op_9063_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9063_end_0 = const()[name = tensor<string, []>("op_9063_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9063_end_mask_0 = const()[name = tensor<string, []>("op_9063_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9063_cast = slice_by_index(begin = var_9063_begin_0, end = var_9063_end_0, end_mask = var_9063_end_mask_0, x = var_9033_cast)[name = tensor<string, []>("op_9063_cast")];
            tensor<int32, [4]> var_9064_begin_0 = const()[name = tensor<string, []>("op_9064_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9064_end_0 = const()[name = tensor<string, []>("op_9064_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_9064_end_mask_0 = const()[name = tensor<string, []>("op_9064_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9064_cast = slice_by_index(begin = var_9064_begin_0, end = var_9064_end_0, end_mask = var_9064_end_mask_0, x = var_9037_cast)[name = tensor<string, []>("op_9064_cast")];
            tensor<int32, [4]> var_9065_begin_0 = const()[name = tensor<string, []>("op_9065_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9065_end_0 = const()[name = tensor<string, []>("op_9065_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9065_end_mask_0 = const()[name = tensor<string, []>("op_9065_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9065_cast = slice_by_index(begin = var_9065_begin_0, end = var_9065_end_0, end_mask = var_9065_end_mask_0, x = var_9037_cast)[name = tensor<string, []>("op_9065_cast")];
            tensor<int32, [4]> var_9066_begin_0 = const()[name = tensor<string, []>("op_9066_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9066_end_0 = const()[name = tensor<string, []>("op_9066_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            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, 80, 1, 512]> var_9066_cast = slice_by_index(begin = var_9066_begin_0, end = var_9066_end_0, end_mask = var_9066_end_mask_0, x = var_9041_cast)[name = tensor<string, []>("op_9066_cast")];
            tensor<int32, [4]> var_9067_begin_0 = const()[name = tensor<string, []>("op_9067_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9067_end_0 = const()[name = tensor<string, []>("op_9067_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9067_end_mask_0 = const()[name = tensor<string, []>("op_9067_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9067_cast = slice_by_index(begin = var_9067_begin_0, end = var_9067_end_0, end_mask = var_9067_end_mask_0, x = var_9041_cast)[name = tensor<string, []>("op_9067_cast")];
            tensor<int32, [4]> var_9068_begin_0 = const()[name = tensor<string, []>("op_9068_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9068_end_0 = const()[name = tensor<string, []>("op_9068_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_9068_end_mask_0 = const()[name = tensor<string, []>("op_9068_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9068_cast = slice_by_index(begin = var_9068_begin_0, end = var_9068_end_0, end_mask = var_9068_end_mask_0, x = var_9045_cast)[name = tensor<string, []>("op_9068_cast")];
            tensor<int32, [4]> var_9069_begin_0 = const()[name = tensor<string, []>("op_9069_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9069_end_0 = const()[name = tensor<string, []>("op_9069_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9069_end_mask_0 = const()[name = tensor<string, []>("op_9069_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9069_cast = slice_by_index(begin = var_9069_begin_0, end = var_9069_end_0, end_mask = var_9069_end_mask_0, x = var_9045_cast)[name = tensor<string, []>("op_9069_cast")];
            tensor<int32, [4]> var_9070_begin_0 = const()[name = tensor<string, []>("op_9070_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9070_end_0 = const()[name = tensor<string, []>("op_9070_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            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, 80, 1, 512]> var_9070_cast = slice_by_index(begin = var_9070_begin_0, end = var_9070_end_0, end_mask = var_9070_end_mask_0, x = var_9049_cast)[name = tensor<string, []>("op_9070_cast")];
            tensor<int32, [4]> var_9071_begin_0 = const()[name = tensor<string, []>("op_9071_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9071_end_0 = const()[name = tensor<string, []>("op_9071_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9071_end_mask_0 = const()[name = tensor<string, []>("op_9071_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9071_cast = slice_by_index(begin = var_9071_begin_0, end = var_9071_end_0, end_mask = var_9071_end_mask_0, x = var_9049_cast)[name = tensor<string, []>("op_9071_cast")];
            tensor<int32, [4]> var_9072_begin_0 = const()[name = tensor<string, []>("op_9072_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9072_end_0 = const()[name = tensor<string, []>("op_9072_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_9072_end_mask_0 = const()[name = tensor<string, []>("op_9072_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9072_cast = slice_by_index(begin = var_9072_begin_0, end = var_9072_end_0, end_mask = var_9072_end_mask_0, x = var_9053_cast)[name = tensor<string, []>("op_9072_cast")];
            tensor<int32, [4]> var_9073_begin_0 = const()[name = tensor<string, []>("op_9073_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9073_end_0 = const()[name = tensor<string, []>("op_9073_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9073_end_mask_0 = const()[name = tensor<string, []>("op_9073_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9073_cast = slice_by_index(begin = var_9073_begin_0, end = var_9073_end_0, end_mask = var_9073_end_mask_0, x = var_9053_cast)[name = tensor<string, []>("op_9073_cast")];
            tensor<int32, [4]> var_9074_begin_0 = const()[name = tensor<string, []>("op_9074_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9074_end_0 = const()[name = tensor<string, []>("op_9074_end_0"), val = tensor<int32, [4]>([2, 80, 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, 80, 1, 512]> var_9074_cast = slice_by_index(begin = var_9074_begin_0, end = var_9074_end_0, end_mask = var_9074_end_mask_0, x = var_9057_cast)[name = tensor<string, []>("op_9074_cast")];
            tensor<int32, [4]> var_9075_begin_0 = const()[name = tensor<string, []>("op_9075_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9075_end_0 = const()[name = tensor<string, []>("op_9075_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9075_end_mask_0 = const()[name = tensor<string, []>("op_9075_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9075_cast = slice_by_index(begin = var_9075_begin_0, end = var_9075_end_0, end_mask = var_9075_end_mask_0, x = var_9057_cast)[name = tensor<string, []>("op_9075_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_9080_begin_0 = const()[name = tensor<string, []>("op_9080_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9080_end_0 = const()[name = tensor<string, []>("op_9080_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 80])];
            tensor<bool, [4]> var_9080_end_mask_0 = const()[name = tensor<string, []>("op_9080_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)[name = tensor<string, []>("transpose_9")];
            tensor<fp16, [2, 1024, 1, 80]> var_9080_cast = slice_by_index(begin = var_9080_begin_0, end = var_9080_end_0, end_mask = var_9080_end_mask_0, x = transpose_9)[name = tensor<string, []>("op_9080_cast")];
            tensor<int32, [4]> var_9084_begin_0 = const()[name = tensor<string, []>("op_9084_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 80])];
            tensor<int32, [4]> var_9084_end_0 = const()[name = tensor<string, []>("op_9084_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 160])];
            tensor<bool, [4]> var_9084_end_mask_0 = const()[name = tensor<string, []>("op_9084_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_9084_cast = slice_by_index(begin = var_9084_begin_0, end = var_9084_end_0, end_mask = var_9084_end_mask_0, x = transpose_9)[name = tensor<string, []>("op_9084_cast")];
            tensor<int32, [4]> var_9088_begin_0 = const()[name = tensor<string, []>("op_9088_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_9088_end_0 = const()[name = tensor<string, []>("op_9088_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 240])];
            tensor<bool, [4]> var_9088_end_mask_0 = const()[name = tensor<string, []>("op_9088_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_9088_cast = slice_by_index(begin = var_9088_begin_0, end = var_9088_end_0, end_mask = var_9088_end_mask_0, x = transpose_9)[name = tensor<string, []>("op_9088_cast")];
            tensor<int32, [4]> var_9092_begin_0 = const()[name = tensor<string, []>("op_9092_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 240])];
            tensor<int32, [4]> var_9092_end_0 = const()[name = tensor<string, []>("op_9092_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_9092_end_mask_0 = const()[name = tensor<string, []>("op_9092_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_9092_cast = slice_by_index(begin = var_9092_begin_0, end = var_9092_end_0, end_mask = var_9092_end_mask_0, x = transpose_9)[name = tensor<string, []>("op_9092_cast")];
            tensor<int32, [4]> var_9096_begin_0 = const()[name = tensor<string, []>("op_9096_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_9096_end_0 = const()[name = tensor<string, []>("op_9096_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 400])];
            tensor<bool, [4]> var_9096_end_mask_0 = const()[name = tensor<string, []>("op_9096_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_9096_cast = slice_by_index(begin = var_9096_begin_0, end = var_9096_end_0, end_mask = var_9096_end_mask_0, x = transpose_9)[name = tensor<string, []>("op_9096_cast")];
            tensor<int32, [4]> var_9100_begin_0 = const()[name = tensor<string, []>("op_9100_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 400])];
            tensor<int32, [4]> var_9100_end_0 = const()[name = tensor<string, []>("op_9100_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 480])];
            tensor<bool, [4]> var_9100_end_mask_0 = const()[name = tensor<string, []>("op_9100_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_9100_cast = slice_by_index(begin = var_9100_begin_0, end = var_9100_end_0, end_mask = var_9100_end_mask_0, x = transpose_9)[name = tensor<string, []>("op_9100_cast")];
            tensor<int32, [4]> var_9104_begin_0 = const()[name = tensor<string, []>("op_9104_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 480])];
            tensor<int32, [4]> var_9104_end_0 = const()[name = tensor<string, []>("op_9104_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 560])];
            tensor<bool, [4]> var_9104_end_mask_0 = const()[name = tensor<string, []>("op_9104_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_9104_cast = slice_by_index(begin = var_9104_begin_0, end = var_9104_end_0, end_mask = var_9104_end_mask_0, x = transpose_9)[name = tensor<string, []>("op_9104_cast")];
            tensor<int32, [4]> var_9108_begin_0 = const()[name = tensor<string, []>("op_9108_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 560])];
            tensor<int32, [4]> var_9108_end_0 = const()[name = tensor<string, []>("op_9108_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_9108_end_mask_0 = const()[name = tensor<string, []>("op_9108_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_9108_cast = slice_by_index(begin = var_9108_begin_0, end = var_9108_end_0, end_mask = var_9108_end_mask_0, x = transpose_9)[name = tensor<string, []>("op_9108_cast")];
            tensor<int32, [4]> var_9110_begin_0 = const()[name = tensor<string, []>("op_9110_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9110_end_0 = const()[name = tensor<string, []>("op_9110_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9110_end_mask_0 = const()[name = tensor<string, []>("op_9110_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9110_cast = slice_by_index(begin = var_9110_begin_0, end = var_9110_end_0, end_mask = var_9110_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_9110_cast")];
            tensor<int32, [4]> var_9114_begin_0 = const()[name = tensor<string, []>("op_9114_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_9114_end_0 = const()[name = tensor<string, []>("op_9114_end_0"), val = tensor<int32, [4]>([2, 160, 1, 1024])];
            tensor<bool, [4]> var_9114_end_mask_0 = const()[name = tensor<string, []>("op_9114_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9114_cast = slice_by_index(begin = var_9114_begin_0, end = var_9114_end_0, end_mask = var_9114_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_9114_cast")];
            tensor<int32, [4]> var_9118_begin_0 = const()[name = tensor<string, []>("op_9118_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_9118_end_0 = const()[name = tensor<string, []>("op_9118_end_0"), val = tensor<int32, [4]>([2, 240, 1, 1024])];
            tensor<bool, [4]> var_9118_end_mask_0 = const()[name = tensor<string, []>("op_9118_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9118_cast = slice_by_index(begin = var_9118_begin_0, end = var_9118_end_0, end_mask = var_9118_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_9118_cast")];
            tensor<int32, [4]> var_9122_begin_0 = const()[name = tensor<string, []>("op_9122_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_9122_end_0 = const()[name = tensor<string, []>("op_9122_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_9122_end_mask_0 = const()[name = tensor<string, []>("op_9122_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9122_cast = slice_by_index(begin = var_9122_begin_0, end = var_9122_end_0, end_mask = var_9122_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_9122_cast")];
            tensor<int32, [4]> var_9126_begin_0 = const()[name = tensor<string, []>("op_9126_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_9126_end_0 = const()[name = tensor<string, []>("op_9126_end_0"), val = tensor<int32, [4]>([2, 400, 1, 1024])];
            tensor<bool, [4]> var_9126_end_mask_0 = const()[name = tensor<string, []>("op_9126_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9126_cast = slice_by_index(begin = var_9126_begin_0, end = var_9126_end_0, end_mask = var_9126_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_9126_cast")];
            tensor<int32, [4]> var_9130_begin_0 = const()[name = tensor<string, []>("op_9130_begin_0"), val = tensor<int32, [4]>([0, 400, 0, 0])];
            tensor<int32, [4]> var_9130_end_0 = const()[name = tensor<string, []>("op_9130_end_0"), val = tensor<int32, [4]>([2, 480, 1, 1024])];
            tensor<bool, [4]> var_9130_end_mask_0 = const()[name = tensor<string, []>("op_9130_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9130_cast = slice_by_index(begin = var_9130_begin_0, end = var_9130_end_0, end_mask = var_9130_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_9130_cast")];
            tensor<int32, [4]> var_9134_begin_0 = const()[name = tensor<string, []>("op_9134_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_9134_end_0 = const()[name = tensor<string, []>("op_9134_end_0"), val = tensor<int32, [4]>([2, 560, 1, 1024])];
            tensor<bool, [4]> var_9134_end_mask_0 = const()[name = tensor<string, []>("op_9134_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9134_cast = slice_by_index(begin = var_9134_begin_0, end = var_9134_end_0, end_mask = var_9134_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_9134_cast")];
            tensor<int32, [4]> var_9138_begin_0 = const()[name = tensor<string, []>("op_9138_begin_0"), val = tensor<int32, [4]>([0, 560, 0, 0])];
            tensor<int32, [4]> var_9138_end_0 = const()[name = tensor<string, []>("op_9138_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_9138_end_mask_0 = const()[name = tensor<string, []>("op_9138_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9138_cast = slice_by_index(begin = var_9138_begin_0, end = var_9138_end_0, end_mask = var_9138_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_9138_cast")];
            tensor<string, []> var_9142_equation_0 = const()[name = tensor<string, []>("op_9142_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9142_cast = einsum(equation = var_9142_equation_0, values = (var_9080_cast, var_9060_cast))[name = tensor<string, []>("op_9142_cast")];
            tensor<fp16, []> var_9143_to_fp16 = const()[name = tensor<string, []>("op_9143_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_705_cast = mul(x = var_9142_cast, y = var_9143_to_fp16)[name = tensor<string, []>("aw_chunk_705_cast")];
            tensor<string, []> var_9146_equation_0 = const()[name = tensor<string, []>("op_9146_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9146_cast = einsum(equation = var_9146_equation_0, values = (var_9080_cast, var_9061_cast))[name = tensor<string, []>("op_9146_cast")];
            tensor<fp16, []> var_9147_to_fp16 = const()[name = tensor<string, []>("op_9147_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_707_cast = mul(x = var_9146_cast, y = var_9147_to_fp16)[name = tensor<string, []>("aw_chunk_707_cast")];
            tensor<string, []> var_9150_equation_0 = const()[name = tensor<string, []>("op_9150_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9150_cast = einsum(equation = var_9150_equation_0, values = (var_9084_cast, var_9062_cast))[name = tensor<string, []>("op_9150_cast")];
            tensor<fp16, []> var_9151_to_fp16 = const()[name = tensor<string, []>("op_9151_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_709_cast = mul(x = var_9150_cast, y = var_9151_to_fp16)[name = tensor<string, []>("aw_chunk_709_cast")];
            tensor<string, []> var_9154_equation_0 = const()[name = tensor<string, []>("op_9154_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9154_cast = einsum(equation = var_9154_equation_0, values = (var_9084_cast, var_9063_cast))[name = tensor<string, []>("op_9154_cast")];
            tensor<fp16, []> var_9155_to_fp16 = const()[name = tensor<string, []>("op_9155_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_711_cast = mul(x = var_9154_cast, y = var_9155_to_fp16)[name = tensor<string, []>("aw_chunk_711_cast")];
            tensor<string, []> var_9158_equation_0 = const()[name = tensor<string, []>("op_9158_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9158_cast = einsum(equation = var_9158_equation_0, values = (var_9088_cast, var_9064_cast))[name = tensor<string, []>("op_9158_cast")];
            tensor<fp16, []> var_9159_to_fp16 = const()[name = tensor<string, []>("op_9159_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_713_cast = mul(x = var_9158_cast, y = var_9159_to_fp16)[name = tensor<string, []>("aw_chunk_713_cast")];
            tensor<string, []> var_9162_equation_0 = const()[name = tensor<string, []>("op_9162_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9162_cast = einsum(equation = var_9162_equation_0, values = (var_9088_cast, var_9065_cast))[name = tensor<string, []>("op_9162_cast")];
            tensor<fp16, []> var_9163_to_fp16 = const()[name = tensor<string, []>("op_9163_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_715_cast = mul(x = var_9162_cast, y = var_9163_to_fp16)[name = tensor<string, []>("aw_chunk_715_cast")];
            tensor<string, []> var_9166_equation_0 = const()[name = tensor<string, []>("op_9166_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9166_cast = einsum(equation = var_9166_equation_0, values = (var_9092_cast, var_9066_cast))[name = tensor<string, []>("op_9166_cast")];
            tensor<fp16, []> var_9167_to_fp16 = const()[name = tensor<string, []>("op_9167_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_717_cast = mul(x = var_9166_cast, y = var_9167_to_fp16)[name = tensor<string, []>("aw_chunk_717_cast")];
            tensor<string, []> var_9170_equation_0 = const()[name = tensor<string, []>("op_9170_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9170_cast = einsum(equation = var_9170_equation_0, values = (var_9092_cast, var_9067_cast))[name = tensor<string, []>("op_9170_cast")];
            tensor<fp16, []> var_9171_to_fp16 = const()[name = tensor<string, []>("op_9171_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_719_cast = mul(x = var_9170_cast, y = var_9171_to_fp16)[name = tensor<string, []>("aw_chunk_719_cast")];
            tensor<string, []> var_9174_equation_0 = const()[name = tensor<string, []>("op_9174_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9174_cast = einsum(equation = var_9174_equation_0, values = (var_9096_cast, var_9068_cast))[name = tensor<string, []>("op_9174_cast")];
            tensor<fp16, []> var_9175_to_fp16 = const()[name = tensor<string, []>("op_9175_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_721_cast = mul(x = var_9174_cast, y = var_9175_to_fp16)[name = tensor<string, []>("aw_chunk_721_cast")];
            tensor<string, []> var_9178_equation_0 = const()[name = tensor<string, []>("op_9178_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9178_cast = einsum(equation = var_9178_equation_0, values = (var_9096_cast, var_9069_cast))[name = tensor<string, []>("op_9178_cast")];
            tensor<fp16, []> var_9179_to_fp16 = const()[name = tensor<string, []>("op_9179_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_723_cast = mul(x = var_9178_cast, y = var_9179_to_fp16)[name = tensor<string, []>("aw_chunk_723_cast")];
            tensor<string, []> var_9182_equation_0 = const()[name = tensor<string, []>("op_9182_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9182_cast = einsum(equation = var_9182_equation_0, values = (var_9100_cast, var_9070_cast))[name = tensor<string, []>("op_9182_cast")];
            tensor<fp16, []> var_9183_to_fp16 = const()[name = tensor<string, []>("op_9183_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_725_cast = mul(x = var_9182_cast, y = var_9183_to_fp16)[name = tensor<string, []>("aw_chunk_725_cast")];
            tensor<string, []> var_9186_equation_0 = const()[name = tensor<string, []>("op_9186_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9186_cast = einsum(equation = var_9186_equation_0, values = (var_9100_cast, var_9071_cast))[name = tensor<string, []>("op_9186_cast")];
            tensor<fp16, []> var_9187_to_fp16 = const()[name = tensor<string, []>("op_9187_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_727_cast = mul(x = var_9186_cast, y = var_9187_to_fp16)[name = tensor<string, []>("aw_chunk_727_cast")];
            tensor<string, []> var_9190_equation_0 = const()[name = tensor<string, []>("op_9190_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9190_cast = einsum(equation = var_9190_equation_0, values = (var_9104_cast, var_9072_cast))[name = tensor<string, []>("op_9190_cast")];
            tensor<fp16, []> var_9191_to_fp16 = const()[name = tensor<string, []>("op_9191_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_729_cast = mul(x = var_9190_cast, y = var_9191_to_fp16)[name = tensor<string, []>("aw_chunk_729_cast")];
            tensor<string, []> var_9194_equation_0 = const()[name = tensor<string, []>("op_9194_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9194_cast = einsum(equation = var_9194_equation_0, values = (var_9104_cast, var_9073_cast))[name = tensor<string, []>("op_9194_cast")];
            tensor<fp16, []> var_9195_to_fp16 = const()[name = tensor<string, []>("op_9195_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_731_cast = mul(x = var_9194_cast, y = var_9195_to_fp16)[name = tensor<string, []>("aw_chunk_731_cast")];
            tensor<string, []> var_9198_equation_0 = const()[name = tensor<string, []>("op_9198_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9198_cast = einsum(equation = var_9198_equation_0, values = (var_9108_cast, var_9074_cast))[name = tensor<string, []>("op_9198_cast")];
            tensor<fp16, []> var_9199_to_fp16 = const()[name = tensor<string, []>("op_9199_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_733_cast = mul(x = var_9198_cast, y = var_9199_to_fp16)[name = tensor<string, []>("aw_chunk_733_cast")];
            tensor<string, []> var_9202_equation_0 = const()[name = tensor<string, []>("op_9202_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9202_cast = einsum(equation = var_9202_equation_0, values = (var_9108_cast, var_9075_cast))[name = tensor<string, []>("op_9202_cast")];
            tensor<fp16, []> var_9203_to_fp16 = const()[name = tensor<string, []>("op_9203_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_735_cast = mul(x = var_9202_cast, y = var_9203_to_fp16)[name = tensor<string, []>("aw_chunk_735_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9205_cast = softmax(axis = var_8182, x = aw_chunk_705_cast)[name = tensor<string, []>("op_9205_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9206_cast = softmax(axis = var_8182, x = aw_chunk_707_cast)[name = tensor<string, []>("op_9206_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9207_cast = softmax(axis = var_8182, x = aw_chunk_709_cast)[name = tensor<string, []>("op_9207_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9208_cast = softmax(axis = var_8182, x = aw_chunk_711_cast)[name = tensor<string, []>("op_9208_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9209_cast = softmax(axis = var_8182, x = aw_chunk_713_cast)[name = tensor<string, []>("op_9209_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9210_cast = softmax(axis = var_8182, x = aw_chunk_715_cast)[name = tensor<string, []>("op_9210_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9211_cast = softmax(axis = var_8182, x = aw_chunk_717_cast)[name = tensor<string, []>("op_9211_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9212_cast = softmax(axis = var_8182, x = aw_chunk_719_cast)[name = tensor<string, []>("op_9212_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9213_cast = softmax(axis = var_8182, x = aw_chunk_721_cast)[name = tensor<string, []>("op_9213_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9214_cast = softmax(axis = var_8182, x = aw_chunk_723_cast)[name = tensor<string, []>("op_9214_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9215_cast = softmax(axis = var_8182, x = aw_chunk_725_cast)[name = tensor<string, []>("op_9215_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9216_cast = softmax(axis = var_8182, x = aw_chunk_727_cast)[name = tensor<string, []>("op_9216_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9217_cast = softmax(axis = var_8182, x = aw_chunk_729_cast)[name = tensor<string, []>("op_9217_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9218_cast = softmax(axis = var_8182, x = aw_chunk_731_cast)[name = tensor<string, []>("op_9218_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9219_cast = softmax(axis = var_8182, x = aw_chunk_733_cast)[name = tensor<string, []>("op_9219_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9220_cast = softmax(axis = var_8182, x = aw_chunk_735_cast)[name = tensor<string, []>("op_9220_cast")];
            tensor<string, []> var_9222_equation_0 = const()[name = tensor<string, []>("op_9222_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9222_cast = einsum(equation = var_9222_equation_0, values = (var_9110_cast, var_9205_cast))[name = tensor<string, []>("op_9222_cast")];
            tensor<string, []> var_9224_equation_0 = const()[name = tensor<string, []>("op_9224_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9224_cast = einsum(equation = var_9224_equation_0, values = (var_9110_cast, var_9206_cast))[name = tensor<string, []>("op_9224_cast")];
            tensor<string, []> var_9226_equation_0 = const()[name = tensor<string, []>("op_9226_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9226_cast = einsum(equation = var_9226_equation_0, values = (var_9114_cast, var_9207_cast))[name = tensor<string, []>("op_9226_cast")];
            tensor<string, []> var_9228_equation_0 = const()[name = tensor<string, []>("op_9228_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9228_cast = einsum(equation = var_9228_equation_0, values = (var_9114_cast, var_9208_cast))[name = tensor<string, []>("op_9228_cast")];
            tensor<string, []> var_9230_equation_0 = const()[name = tensor<string, []>("op_9230_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9230_cast = einsum(equation = var_9230_equation_0, values = (var_9118_cast, var_9209_cast))[name = tensor<string, []>("op_9230_cast")];
            tensor<string, []> var_9232_equation_0 = const()[name = tensor<string, []>("op_9232_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9232_cast = einsum(equation = var_9232_equation_0, values = (var_9118_cast, var_9210_cast))[name = tensor<string, []>("op_9232_cast")];
            tensor<string, []> var_9234_equation_0 = const()[name = tensor<string, []>("op_9234_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9234_cast = einsum(equation = var_9234_equation_0, values = (var_9122_cast, var_9211_cast))[name = tensor<string, []>("op_9234_cast")];
            tensor<string, []> var_9236_equation_0 = const()[name = tensor<string, []>("op_9236_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9236_cast = einsum(equation = var_9236_equation_0, values = (var_9122_cast, var_9212_cast))[name = tensor<string, []>("op_9236_cast")];
            tensor<string, []> var_9238_equation_0 = const()[name = tensor<string, []>("op_9238_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9238_cast = einsum(equation = var_9238_equation_0, values = (var_9126_cast, var_9213_cast))[name = tensor<string, []>("op_9238_cast")];
            tensor<string, []> var_9240_equation_0 = const()[name = tensor<string, []>("op_9240_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9240_cast = einsum(equation = var_9240_equation_0, values = (var_9126_cast, var_9214_cast))[name = tensor<string, []>("op_9240_cast")];
            tensor<string, []> var_9242_equation_0 = const()[name = tensor<string, []>("op_9242_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9242_cast = einsum(equation = var_9242_equation_0, values = (var_9130_cast, var_9215_cast))[name = tensor<string, []>("op_9242_cast")];
            tensor<string, []> var_9244_equation_0 = const()[name = tensor<string, []>("op_9244_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9244_cast = einsum(equation = var_9244_equation_0, values = (var_9130_cast, var_9216_cast))[name = tensor<string, []>("op_9244_cast")];
            tensor<string, []> var_9246_equation_0 = const()[name = tensor<string, []>("op_9246_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9246_cast = einsum(equation = var_9246_equation_0, values = (var_9134_cast, var_9217_cast))[name = tensor<string, []>("op_9246_cast")];
            tensor<string, []> var_9248_equation_0 = const()[name = tensor<string, []>("op_9248_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9248_cast = einsum(equation = var_9248_equation_0, values = (var_9134_cast, var_9218_cast))[name = tensor<string, []>("op_9248_cast")];
            tensor<string, []> var_9250_equation_0 = const()[name = tensor<string, []>("op_9250_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9250_cast = einsum(equation = var_9250_equation_0, values = (var_9138_cast, var_9219_cast))[name = tensor<string, []>("op_9250_cast")];
            tensor<string, []> var_9252_equation_0 = const()[name = tensor<string, []>("op_9252_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9252_cast = einsum(equation = var_9252_equation_0, values = (var_9138_cast, var_9220_cast))[name = tensor<string, []>("op_9252_cast")];
            tensor<bool, []> var_9254_interleave_0 = const()[name = tensor<string, []>("op_9254_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_9254_cast = concat(axis = var_8160, interleave = var_9254_interleave_0, values = (var_9222_cast, var_9224_cast))[name = tensor<string, []>("op_9254_cast")];
            tensor<bool, []> var_9256_interleave_0 = const()[name = tensor<string, []>("op_9256_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_9256_cast = concat(axis = var_8160, interleave = var_9256_interleave_0, values = (var_9226_cast, var_9228_cast))[name = tensor<string, []>("op_9256_cast")];
            tensor<bool, []> var_9258_interleave_0 = const()[name = tensor<string, []>("op_9258_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_9258_cast = concat(axis = var_8160, interleave = var_9258_interleave_0, values = (var_9230_cast, var_9232_cast))[name = tensor<string, []>("op_9258_cast")];
            tensor<bool, []> var_9260_interleave_0 = const()[name = tensor<string, []>("op_9260_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_9260_cast = concat(axis = var_8160, interleave = var_9260_interleave_0, values = (var_9234_cast, var_9236_cast))[name = tensor<string, []>("op_9260_cast")];
            tensor<bool, []> var_9262_interleave_0 = const()[name = tensor<string, []>("op_9262_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_9262_cast = concat(axis = var_8160, interleave = var_9262_interleave_0, values = (var_9238_cast, var_9240_cast))[name = tensor<string, []>("op_9262_cast")];
            tensor<bool, []> var_9264_interleave_0 = const()[name = tensor<string, []>("op_9264_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_9264_cast = concat(axis = var_8160, interleave = var_9264_interleave_0, values = (var_9242_cast, var_9244_cast))[name = tensor<string, []>("op_9264_cast")];
            tensor<bool, []> var_9266_interleave_0 = const()[name = tensor<string, []>("op_9266_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_9266_cast = concat(axis = var_8160, interleave = var_9266_interleave_0, values = (var_9246_cast, var_9248_cast))[name = tensor<string, []>("op_9266_cast")];
            tensor<bool, []> var_9268_interleave_0 = const()[name = tensor<string, []>("op_9268_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_9268_cast = concat(axis = var_8160, interleave = var_9268_interleave_0, values = (var_9250_cast, var_9252_cast))[name = tensor<string, []>("op_9268_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_8182, interleave = input_409_interleave_0, values = (var_9254_cast, var_9256_cast, var_9258_cast, var_9260_cast, var_9262_cast, var_9264_cast, var_9266_cast, var_9268_cast))[name = tensor<string, []>("input_409_cast")];
            tensor<int32, [2]> var_9274 = const()[name = tensor<string, []>("op_9274"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9276 = const()[name = tensor<string, []>("op_9276"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_9278_pad_type_0 = const()[name = tensor<string, []>("op_9278_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_9278_pad_0 = const()[name = tensor<string, []>("op_9278_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(607657984))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(607965248))), name = tensor<string, []>("up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            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, []>(607965440)))];
            tensor<fp16, [2, 640, 1, 1024]> var_9278_cast = conv(bias = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_9276, groups = var_8182, pad = var_9278_pad_0, pad_type = var_9278_pad_type_0, strides = var_9274, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_409_cast)[name = tensor<string, []>("op_9278_cast")];
            tensor<fp16, [2, 640, 1, 1024]> inputs_69_cast = add(x = var_9278_cast, y = inputs_67_cast)[name = tensor<string, []>("inputs_69_cast")];
            tensor<int32, [1]> var_9282 = const()[name = tensor<string, []>("op_9282"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_69_cast = reduce_mean(axes = var_9282, keep_dims = var_8177, x = inputs_69_cast)[name = tensor<string, []>("channels_mean_69_cast")];
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_69_cast = sub(x = inputs_69_cast, y = channels_mean_69_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_69_cast")];
            tensor<int32, [1]> var_9286 = const()[name = tensor<string, []>("op_9286"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_9287_cast = reduce_mean(axes = var_9286, keep_dims = var_8177, x = zero_mean_sq_69_cast)[name = tensor<string, []>("op_9287_cast")];
            tensor<fp16, []> var_9288_to_fp16 = const()[name = tensor<string, []>("op_9288_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_9289_cast = add(x = var_9287_cast, y = var_9288_to_fp16)[name = tensor<string, []>("op_9289_cast")];
            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_9289_cast)[name = tensor<string, []>("denom_69_cast")];
            tensor<fp16, [2, 640, 1, 1024]> out_69_cast = mul(x = zero_mean_69_cast, y = denom_69_cast)[name = tensor<string, []>("out_69_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_9293_to_fp16 = const()[name = tensor<string, []>("op_9293_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(607966784)))];
            tensor<fp16, [2, 640, 1, 1024]> var_9294_cast = add(x = out_69_cast, y = var_9293_to_fp16)[name = tensor<string, []>("op_9294_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_9296_to_fp16 = const()[name = tensor<string, []>("op_9296_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(607968128)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_249_cast = mul(x = var_9294_cast, y = var_9296_to_fp16)[name = tensor<string, []>("hidden_states_249_cast")];
            tensor<int32, [2]> var_9303 = const()[name = tensor<string, []>("op_9303"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9305 = const()[name = tensor<string, []>("op_9305"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(607969472))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(608276736))), name = tensor<string, []>("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 1024]> q_47_cast = conv(dilations = var_9305, groups = var_8182, pad = q_47_pad_0, pad_type = q_47_pad_type_0, strides = var_9303, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_249_cast)[name = tensor<string, []>("q_47_cast")];
            tensor<int32, [2]> var_9309 = const()[name = tensor<string, []>("op_9309"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9311 = const()[name = tensor<string, []>("op_9311"), 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, 768, 1, 1]> up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [368640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(608276928))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(608645632))), name = tensor<string, []>("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 768, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> k_93_cast = conv(dilations = var_9311, groups = var_8182, pad = k_93_pad_0, pad_type = k_93_pad_type_0, strides = var_9309, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_93_cast")];
            tensor<int32, [2]> var_9315 = const()[name = tensor<string, []>("op_9315"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9317 = const()[name = tensor<string, []>("op_9317"), 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, 768, 1, 1]> up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [368640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(608645824))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(609014528))), name = tensor<string, []>("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 768, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> v_47_cast = conv(dilations = var_9317, groups = var_8182, pad = v_47_pad_0, pad_type = v_47_pad_type_0, strides = var_9315, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_47_cast")];
            tensor<int32, [4]> var_9321_begin_0 = const()[name = tensor<string, []>("op_9321_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9321_end_0 = const()[name = tensor<string, []>("op_9321_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9321_end_mask_0 = const()[name = tensor<string, []>("op_9321_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9321_cast = slice_by_index(begin = var_9321_begin_0, end = var_9321_end_0, end_mask = var_9321_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_9321_cast")];
            tensor<int32, [4]> var_9325_begin_0 = const()[name = tensor<string, []>("op_9325_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_9325_end_0 = const()[name = tensor<string, []>("op_9325_end_0"), val = tensor<int32, [4]>([2, 160, 1, 1024])];
            tensor<bool, [4]> var_9325_end_mask_0 = const()[name = tensor<string, []>("op_9325_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9325_cast = slice_by_index(begin = var_9325_begin_0, end = var_9325_end_0, end_mask = var_9325_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_9325_cast")];
            tensor<int32, [4]> var_9329_begin_0 = const()[name = tensor<string, []>("op_9329_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_9329_end_0 = const()[name = tensor<string, []>("op_9329_end_0"), val = tensor<int32, [4]>([2, 240, 1, 1024])];
            tensor<bool, [4]> var_9329_end_mask_0 = const()[name = tensor<string, []>("op_9329_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9329_cast = slice_by_index(begin = var_9329_begin_0, end = var_9329_end_0, end_mask = var_9329_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_9329_cast")];
            tensor<int32, [4]> var_9333_begin_0 = const()[name = tensor<string, []>("op_9333_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_9333_end_0 = const()[name = tensor<string, []>("op_9333_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_9333_end_mask_0 = const()[name = tensor<string, []>("op_9333_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9333_cast = slice_by_index(begin = var_9333_begin_0, end = var_9333_end_0, end_mask = var_9333_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_9333_cast")];
            tensor<int32, [4]> var_9337_begin_0 = const()[name = tensor<string, []>("op_9337_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_9337_end_0 = const()[name = tensor<string, []>("op_9337_end_0"), val = tensor<int32, [4]>([2, 400, 1, 1024])];
            tensor<bool, [4]> var_9337_end_mask_0 = const()[name = tensor<string, []>("op_9337_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9337_cast = slice_by_index(begin = var_9337_begin_0, end = var_9337_end_0, end_mask = var_9337_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_9337_cast")];
            tensor<int32, [4]> var_9341_begin_0 = const()[name = tensor<string, []>("op_9341_begin_0"), val = tensor<int32, [4]>([0, 400, 0, 0])];
            tensor<int32, [4]> var_9341_end_0 = const()[name = tensor<string, []>("op_9341_end_0"), val = tensor<int32, [4]>([2, 480, 1, 1024])];
            tensor<bool, [4]> var_9341_end_mask_0 = const()[name = tensor<string, []>("op_9341_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9341_cast = slice_by_index(begin = var_9341_begin_0, end = var_9341_end_0, end_mask = var_9341_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_9341_cast")];
            tensor<int32, [4]> var_9345_begin_0 = const()[name = tensor<string, []>("op_9345_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_9345_end_0 = const()[name = tensor<string, []>("op_9345_end_0"), val = tensor<int32, [4]>([2, 560, 1, 1024])];
            tensor<bool, [4]> var_9345_end_mask_0 = const()[name = tensor<string, []>("op_9345_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9345_cast = slice_by_index(begin = var_9345_begin_0, end = var_9345_end_0, end_mask = var_9345_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_9345_cast")];
            tensor<int32, [4]> var_9349_begin_0 = const()[name = tensor<string, []>("op_9349_begin_0"), val = tensor<int32, [4]>([0, 560, 0, 0])];
            tensor<int32, [4]> var_9349_end_0 = const()[name = tensor<string, []>("op_9349_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_9349_end_mask_0 = const()[name = tensor<string, []>("op_9349_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9349_cast = slice_by_index(begin = var_9349_begin_0, end = var_9349_end_0, end_mask = var_9349_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_9349_cast")];
            tensor<int32, [4]> var_9352_begin_0 = const()[name = tensor<string, []>("op_9352_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9352_end_0 = const()[name = tensor<string, []>("op_9352_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_9352_end_mask_0 = const()[name = tensor<string, []>("op_9352_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9352_cast = slice_by_index(begin = var_9352_begin_0, end = var_9352_end_0, end_mask = var_9352_end_mask_0, x = var_9321_cast)[name = tensor<string, []>("op_9352_cast")];
            tensor<int32, [4]> var_9353_begin_0 = const()[name = tensor<string, []>("op_9353_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9353_end_0 = const()[name = tensor<string, []>("op_9353_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9353_end_mask_0 = const()[name = tensor<string, []>("op_9353_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9353_cast = slice_by_index(begin = var_9353_begin_0, end = var_9353_end_0, end_mask = var_9353_end_mask_0, x = var_9321_cast)[name = tensor<string, []>("op_9353_cast")];
            tensor<int32, [4]> var_9354_begin_0 = const()[name = tensor<string, []>("op_9354_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9354_end_0 = const()[name = tensor<string, []>("op_9354_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_9354_end_mask_0 = const()[name = tensor<string, []>("op_9354_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9354_cast = slice_by_index(begin = var_9354_begin_0, end = var_9354_end_0, end_mask = var_9354_end_mask_0, x = var_9325_cast)[name = tensor<string, []>("op_9354_cast")];
            tensor<int32, [4]> var_9355_begin_0 = const()[name = tensor<string, []>("op_9355_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9355_end_0 = const()[name = tensor<string, []>("op_9355_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9355_end_mask_0 = const()[name = tensor<string, []>("op_9355_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9355_cast = slice_by_index(begin = var_9355_begin_0, end = var_9355_end_0, end_mask = var_9355_end_mask_0, x = var_9325_cast)[name = tensor<string, []>("op_9355_cast")];
            tensor<int32, [4]> var_9356_begin_0 = const()[name = tensor<string, []>("op_9356_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9356_end_0 = const()[name = tensor<string, []>("op_9356_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_9356_end_mask_0 = const()[name = tensor<string, []>("op_9356_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9356_cast = slice_by_index(begin = var_9356_begin_0, end = var_9356_end_0, end_mask = var_9356_end_mask_0, x = var_9329_cast)[name = tensor<string, []>("op_9356_cast")];
            tensor<int32, [4]> var_9357_begin_0 = const()[name = tensor<string, []>("op_9357_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9357_end_0 = const()[name = tensor<string, []>("op_9357_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9357_end_mask_0 = const()[name = tensor<string, []>("op_9357_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9357_cast = slice_by_index(begin = var_9357_begin_0, end = var_9357_end_0, end_mask = var_9357_end_mask_0, x = var_9329_cast)[name = tensor<string, []>("op_9357_cast")];
            tensor<int32, [4]> var_9358_begin_0 = const()[name = tensor<string, []>("op_9358_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9358_end_0 = const()[name = tensor<string, []>("op_9358_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_9358_end_mask_0 = const()[name = tensor<string, []>("op_9358_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9358_cast = slice_by_index(begin = var_9358_begin_0, end = var_9358_end_0, end_mask = var_9358_end_mask_0, x = var_9333_cast)[name = tensor<string, []>("op_9358_cast")];
            tensor<int32, [4]> var_9359_begin_0 = const()[name = tensor<string, []>("op_9359_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9359_end_0 = const()[name = tensor<string, []>("op_9359_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9359_end_mask_0 = const()[name = tensor<string, []>("op_9359_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9359_cast = slice_by_index(begin = var_9359_begin_0, end = var_9359_end_0, end_mask = var_9359_end_mask_0, x = var_9333_cast)[name = tensor<string, []>("op_9359_cast")];
            tensor<int32, [4]> var_9360_begin_0 = const()[name = tensor<string, []>("op_9360_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9360_end_0 = const()[name = tensor<string, []>("op_9360_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_9360_end_mask_0 = const()[name = tensor<string, []>("op_9360_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9360_cast = slice_by_index(begin = var_9360_begin_0, end = var_9360_end_0, end_mask = var_9360_end_mask_0, x = var_9337_cast)[name = tensor<string, []>("op_9360_cast")];
            tensor<int32, [4]> var_9361_begin_0 = const()[name = tensor<string, []>("op_9361_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9361_end_0 = const()[name = tensor<string, []>("op_9361_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9361_end_mask_0 = const()[name = tensor<string, []>("op_9361_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9361_cast = slice_by_index(begin = var_9361_begin_0, end = var_9361_end_0, end_mask = var_9361_end_mask_0, x = var_9337_cast)[name = tensor<string, []>("op_9361_cast")];
            tensor<int32, [4]> var_9362_begin_0 = const()[name = tensor<string, []>("op_9362_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9362_end_0 = const()[name = tensor<string, []>("op_9362_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_9362_end_mask_0 = const()[name = tensor<string, []>("op_9362_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9362_cast = slice_by_index(begin = var_9362_begin_0, end = var_9362_end_0, end_mask = var_9362_end_mask_0, x = var_9341_cast)[name = tensor<string, []>("op_9362_cast")];
            tensor<int32, [4]> var_9363_begin_0 = const()[name = tensor<string, []>("op_9363_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9363_end_0 = const()[name = tensor<string, []>("op_9363_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9363_end_mask_0 = const()[name = tensor<string, []>("op_9363_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9363_cast = slice_by_index(begin = var_9363_begin_0, end = var_9363_end_0, end_mask = var_9363_end_mask_0, x = var_9341_cast)[name = tensor<string, []>("op_9363_cast")];
            tensor<int32, [4]> var_9364_begin_0 = const()[name = tensor<string, []>("op_9364_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9364_end_0 = const()[name = tensor<string, []>("op_9364_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_9364_end_mask_0 = const()[name = tensor<string, []>("op_9364_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9364_cast = slice_by_index(begin = var_9364_begin_0, end = var_9364_end_0, end_mask = var_9364_end_mask_0, x = var_9345_cast)[name = tensor<string, []>("op_9364_cast")];
            tensor<int32, [4]> var_9365_begin_0 = const()[name = tensor<string, []>("op_9365_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9365_end_0 = const()[name = tensor<string, []>("op_9365_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9365_end_mask_0 = const()[name = tensor<string, []>("op_9365_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9365_cast = slice_by_index(begin = var_9365_begin_0, end = var_9365_end_0, end_mask = var_9365_end_mask_0, x = var_9345_cast)[name = tensor<string, []>("op_9365_cast")];
            tensor<int32, [4]> var_9366_begin_0 = const()[name = tensor<string, []>("op_9366_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9366_end_0 = const()[name = tensor<string, []>("op_9366_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_9366_end_mask_0 = const()[name = tensor<string, []>("op_9366_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9366_cast = slice_by_index(begin = var_9366_begin_0, end = var_9366_end_0, end_mask = var_9366_end_mask_0, x = var_9349_cast)[name = tensor<string, []>("op_9366_cast")];
            tensor<int32, [4]> var_9367_begin_0 = const()[name = tensor<string, []>("op_9367_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9367_end_0 = const()[name = tensor<string, []>("op_9367_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9367_end_mask_0 = const()[name = tensor<string, []>("op_9367_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9367_cast = slice_by_index(begin = var_9367_begin_0, end = var_9367_end_0, end_mask = var_9367_end_mask_0, x = var_9349_cast)[name = tensor<string, []>("op_9367_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_9372_begin_0 = const()[name = tensor<string, []>("op_9372_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9372_end_0 = const()[name = tensor<string, []>("op_9372_end_0"), val = tensor<int32, [4]>([2, 77, 1, 80])];
            tensor<bool, [4]> var_9372_end_mask_0 = const()[name = tensor<string, []>("op_9372_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)[name = tensor<string, []>("transpose_8")];
            tensor<fp16, [2, 77, 1, 80]> var_9372_cast = slice_by_index(begin = var_9372_begin_0, end = var_9372_end_0, end_mask = var_9372_end_mask_0, x = transpose_8)[name = tensor<string, []>("op_9372_cast")];
            tensor<int32, [4]> var_9376_begin_0 = const()[name = tensor<string, []>("op_9376_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 80])];
            tensor<int32, [4]> var_9376_end_0 = const()[name = tensor<string, []>("op_9376_end_0"), val = tensor<int32, [4]>([2, 77, 1, 160])];
            tensor<bool, [4]> var_9376_end_mask_0 = const()[name = tensor<string, []>("op_9376_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_9376_cast = slice_by_index(begin = var_9376_begin_0, end = var_9376_end_0, end_mask = var_9376_end_mask_0, x = transpose_8)[name = tensor<string, []>("op_9376_cast")];
            tensor<int32, [4]> var_9380_begin_0 = const()[name = tensor<string, []>("op_9380_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_9380_end_0 = const()[name = tensor<string, []>("op_9380_end_0"), val = tensor<int32, [4]>([2, 77, 1, 240])];
            tensor<bool, [4]> var_9380_end_mask_0 = const()[name = tensor<string, []>("op_9380_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_9380_cast = slice_by_index(begin = var_9380_begin_0, end = var_9380_end_0, end_mask = var_9380_end_mask_0, x = transpose_8)[name = tensor<string, []>("op_9380_cast")];
            tensor<int32, [4]> var_9384_begin_0 = const()[name = tensor<string, []>("op_9384_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 240])];
            tensor<int32, [4]> var_9384_end_0 = const()[name = tensor<string, []>("op_9384_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_9384_end_mask_0 = const()[name = tensor<string, []>("op_9384_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_9384_cast = slice_by_index(begin = var_9384_begin_0, end = var_9384_end_0, end_mask = var_9384_end_mask_0, x = transpose_8)[name = tensor<string, []>("op_9384_cast")];
            tensor<int32, [4]> var_9388_begin_0 = const()[name = tensor<string, []>("op_9388_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_9388_end_0 = const()[name = tensor<string, []>("op_9388_end_0"), val = tensor<int32, [4]>([2, 77, 1, 400])];
            tensor<bool, [4]> var_9388_end_mask_0 = const()[name = tensor<string, []>("op_9388_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_9388_cast = slice_by_index(begin = var_9388_begin_0, end = var_9388_end_0, end_mask = var_9388_end_mask_0, x = transpose_8)[name = tensor<string, []>("op_9388_cast")];
            tensor<int32, [4]> var_9392_begin_0 = const()[name = tensor<string, []>("op_9392_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 400])];
            tensor<int32, [4]> var_9392_end_0 = const()[name = tensor<string, []>("op_9392_end_0"), val = tensor<int32, [4]>([2, 77, 1, 480])];
            tensor<bool, [4]> var_9392_end_mask_0 = const()[name = tensor<string, []>("op_9392_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_9392_cast = slice_by_index(begin = var_9392_begin_0, end = var_9392_end_0, end_mask = var_9392_end_mask_0, x = transpose_8)[name = tensor<string, []>("op_9392_cast")];
            tensor<int32, [4]> var_9396_begin_0 = const()[name = tensor<string, []>("op_9396_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 480])];
            tensor<int32, [4]> var_9396_end_0 = const()[name = tensor<string, []>("op_9396_end_0"), val = tensor<int32, [4]>([2, 77, 1, 560])];
            tensor<bool, [4]> var_9396_end_mask_0 = const()[name = tensor<string, []>("op_9396_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_9396_cast = slice_by_index(begin = var_9396_begin_0, end = var_9396_end_0, end_mask = var_9396_end_mask_0, x = transpose_8)[name = tensor<string, []>("op_9396_cast")];
            tensor<int32, [4]> var_9400_begin_0 = const()[name = tensor<string, []>("op_9400_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 560])];
            tensor<int32, [4]> var_9400_end_0 = const()[name = tensor<string, []>("op_9400_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_9400_end_mask_0 = const()[name = tensor<string, []>("op_9400_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_9400_cast = slice_by_index(begin = var_9400_begin_0, end = var_9400_end_0, end_mask = var_9400_end_mask_0, x = transpose_8)[name = tensor<string, []>("op_9400_cast")];
            tensor<int32, [4]> var_9402_begin_0 = const()[name = tensor<string, []>("op_9402_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9402_end_0 = const()[name = tensor<string, []>("op_9402_end_0"), val = tensor<int32, [4]>([2, 80, 1, 77])];
            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, 80, 1, 77]> var_9402_cast = slice_by_index(begin = var_9402_begin_0, end = var_9402_end_0, end_mask = var_9402_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_9402_cast")];
            tensor<int32, [4]> var_9406_begin_0 = const()[name = tensor<string, []>("op_9406_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_9406_end_0 = const()[name = tensor<string, []>("op_9406_end_0"), val = tensor<int32, [4]>([2, 160, 1, 77])];
            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, 80, 1, 77]> var_9406_cast = slice_by_index(begin = var_9406_begin_0, end = var_9406_end_0, end_mask = var_9406_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_9406_cast")];
            tensor<int32, [4]> var_9410_begin_0 = const()[name = tensor<string, []>("op_9410_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_9410_end_0 = const()[name = tensor<string, []>("op_9410_end_0"), val = tensor<int32, [4]>([2, 240, 1, 77])];
            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, 80, 1, 77]> var_9410_cast = slice_by_index(begin = var_9410_begin_0, end = var_9410_end_0, end_mask = var_9410_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_9410_cast")];
            tensor<int32, [4]> var_9414_begin_0 = const()[name = tensor<string, []>("op_9414_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_9414_end_0 = const()[name = tensor<string, []>("op_9414_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_9414_end_mask_0 = const()[name = tensor<string, []>("op_9414_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_9414_cast = slice_by_index(begin = var_9414_begin_0, end = var_9414_end_0, end_mask = var_9414_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_9414_cast")];
            tensor<int32, [4]> var_9418_begin_0 = const()[name = tensor<string, []>("op_9418_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_9418_end_0 = const()[name = tensor<string, []>("op_9418_end_0"), val = tensor<int32, [4]>([2, 400, 1, 77])];
            tensor<bool, [4]> var_9418_end_mask_0 = const()[name = tensor<string, []>("op_9418_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_9418_cast = slice_by_index(begin = var_9418_begin_0, end = var_9418_end_0, end_mask = var_9418_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_9418_cast")];
            tensor<int32, [4]> var_9422_begin_0 = const()[name = tensor<string, []>("op_9422_begin_0"), val = tensor<int32, [4]>([0, 400, 0, 0])];
            tensor<int32, [4]> var_9422_end_0 = const()[name = tensor<string, []>("op_9422_end_0"), val = tensor<int32, [4]>([2, 480, 1, 77])];
            tensor<bool, [4]> var_9422_end_mask_0 = const()[name = tensor<string, []>("op_9422_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_9422_cast = slice_by_index(begin = var_9422_begin_0, end = var_9422_end_0, end_mask = var_9422_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_9422_cast")];
            tensor<int32, [4]> var_9426_begin_0 = const()[name = tensor<string, []>("op_9426_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_9426_end_0 = const()[name = tensor<string, []>("op_9426_end_0"), val = tensor<int32, [4]>([2, 560, 1, 77])];
            tensor<bool, [4]> var_9426_end_mask_0 = const()[name = tensor<string, []>("op_9426_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_9426_cast = slice_by_index(begin = var_9426_begin_0, end = var_9426_end_0, end_mask = var_9426_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_9426_cast")];
            tensor<int32, [4]> var_9430_begin_0 = const()[name = tensor<string, []>("op_9430_begin_0"), val = tensor<int32, [4]>([0, 560, 0, 0])];
            tensor<int32, [4]> var_9430_end_0 = const()[name = tensor<string, []>("op_9430_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_9430_end_mask_0 = const()[name = tensor<string, []>("op_9430_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_9430_cast = slice_by_index(begin = var_9430_begin_0, end = var_9430_end_0, end_mask = var_9430_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_9430_cast")];
            tensor<string, []> var_9434_equation_0 = const()[name = tensor<string, []>("op_9434_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_9434_cast = einsum(equation = var_9434_equation_0, values = (var_9372_cast, var_9352_cast))[name = tensor<string, []>("op_9434_cast")];
            tensor<fp16, []> var_9435_to_fp16 = const()[name = tensor<string, []>("op_9435_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_737_cast = mul(x = var_9434_cast, y = var_9435_to_fp16)[name = tensor<string, []>("aw_chunk_737_cast")];
            tensor<string, []> var_9438_equation_0 = const()[name = tensor<string, []>("op_9438_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_9438_cast = einsum(equation = var_9438_equation_0, values = (var_9372_cast, var_9353_cast))[name = tensor<string, []>("op_9438_cast")];
            tensor<fp16, []> var_9439_to_fp16 = const()[name = tensor<string, []>("op_9439_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_739_cast = mul(x = var_9438_cast, y = var_9439_to_fp16)[name = tensor<string, []>("aw_chunk_739_cast")];
            tensor<string, []> var_9442_equation_0 = const()[name = tensor<string, []>("op_9442_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_9442_cast = einsum(equation = var_9442_equation_0, values = (var_9376_cast, var_9354_cast))[name = tensor<string, []>("op_9442_cast")];
            tensor<fp16, []> var_9443_to_fp16 = const()[name = tensor<string, []>("op_9443_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_741_cast = mul(x = var_9442_cast, y = var_9443_to_fp16)[name = tensor<string, []>("aw_chunk_741_cast")];
            tensor<string, []> var_9446_equation_0 = const()[name = tensor<string, []>("op_9446_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_9446_cast = einsum(equation = var_9446_equation_0, values = (var_9376_cast, var_9355_cast))[name = tensor<string, []>("op_9446_cast")];
            tensor<fp16, []> var_9447_to_fp16 = const()[name = tensor<string, []>("op_9447_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_743_cast = mul(x = var_9446_cast, y = var_9447_to_fp16)[name = tensor<string, []>("aw_chunk_743_cast")];
            tensor<string, []> var_9450_equation_0 = const()[name = tensor<string, []>("op_9450_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_9450_cast = einsum(equation = var_9450_equation_0, values = (var_9380_cast, var_9356_cast))[name = tensor<string, []>("op_9450_cast")];
            tensor<fp16, []> var_9451_to_fp16 = const()[name = tensor<string, []>("op_9451_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_745_cast = mul(x = var_9450_cast, y = var_9451_to_fp16)[name = tensor<string, []>("aw_chunk_745_cast")];
            tensor<string, []> var_9454_equation_0 = const()[name = tensor<string, []>("op_9454_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_9454_cast = einsum(equation = var_9454_equation_0, values = (var_9380_cast, var_9357_cast))[name = tensor<string, []>("op_9454_cast")];
            tensor<fp16, []> var_9455_to_fp16 = const()[name = tensor<string, []>("op_9455_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_747_cast = mul(x = var_9454_cast, y = var_9455_to_fp16)[name = tensor<string, []>("aw_chunk_747_cast")];
            tensor<string, []> var_9458_equation_0 = const()[name = tensor<string, []>("op_9458_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_9458_cast = einsum(equation = var_9458_equation_0, values = (var_9384_cast, var_9358_cast))[name = tensor<string, []>("op_9458_cast")];
            tensor<fp16, []> var_9459_to_fp16 = const()[name = tensor<string, []>("op_9459_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_749_cast = mul(x = var_9458_cast, y = var_9459_to_fp16)[name = tensor<string, []>("aw_chunk_749_cast")];
            tensor<string, []> var_9462_equation_0 = const()[name = tensor<string, []>("op_9462_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_9462_cast = einsum(equation = var_9462_equation_0, values = (var_9384_cast, var_9359_cast))[name = tensor<string, []>("op_9462_cast")];
            tensor<fp16, []> var_9463_to_fp16 = const()[name = tensor<string, []>("op_9463_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_751_cast = mul(x = var_9462_cast, y = var_9463_to_fp16)[name = tensor<string, []>("aw_chunk_751_cast")];
            tensor<string, []> var_9466_equation_0 = const()[name = tensor<string, []>("op_9466_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_9466_cast = einsum(equation = var_9466_equation_0, values = (var_9388_cast, var_9360_cast))[name = tensor<string, []>("op_9466_cast")];
            tensor<fp16, []> var_9467_to_fp16 = const()[name = tensor<string, []>("op_9467_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_753_cast = mul(x = var_9466_cast, y = var_9467_to_fp16)[name = tensor<string, []>("aw_chunk_753_cast")];
            tensor<string, []> var_9470_equation_0 = const()[name = tensor<string, []>("op_9470_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_9470_cast = einsum(equation = var_9470_equation_0, values = (var_9388_cast, var_9361_cast))[name = tensor<string, []>("op_9470_cast")];
            tensor<fp16, []> var_9471_to_fp16 = const()[name = tensor<string, []>("op_9471_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_755_cast = mul(x = var_9470_cast, y = var_9471_to_fp16)[name = tensor<string, []>("aw_chunk_755_cast")];
            tensor<string, []> var_9474_equation_0 = const()[name = tensor<string, []>("op_9474_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_9474_cast = einsum(equation = var_9474_equation_0, values = (var_9392_cast, var_9362_cast))[name = tensor<string, []>("op_9474_cast")];
            tensor<fp16, []> var_9475_to_fp16 = const()[name = tensor<string, []>("op_9475_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_757_cast = mul(x = var_9474_cast, y = var_9475_to_fp16)[name = tensor<string, []>("aw_chunk_757_cast")];
            tensor<string, []> var_9478_equation_0 = const()[name = tensor<string, []>("op_9478_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_9478_cast = einsum(equation = var_9478_equation_0, values = (var_9392_cast, var_9363_cast))[name = tensor<string, []>("op_9478_cast")];
            tensor<fp16, []> var_9479_to_fp16 = const()[name = tensor<string, []>("op_9479_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_759_cast = mul(x = var_9478_cast, y = var_9479_to_fp16)[name = tensor<string, []>("aw_chunk_759_cast")];
            tensor<string, []> var_9482_equation_0 = const()[name = tensor<string, []>("op_9482_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_9482_cast = einsum(equation = var_9482_equation_0, values = (var_9396_cast, var_9364_cast))[name = tensor<string, []>("op_9482_cast")];
            tensor<fp16, []> var_9483_to_fp16 = const()[name = tensor<string, []>("op_9483_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_761_cast = mul(x = var_9482_cast, y = var_9483_to_fp16)[name = tensor<string, []>("aw_chunk_761_cast")];
            tensor<string, []> var_9486_equation_0 = const()[name = tensor<string, []>("op_9486_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_9486_cast = einsum(equation = var_9486_equation_0, values = (var_9396_cast, var_9365_cast))[name = tensor<string, []>("op_9486_cast")];
            tensor<fp16, []> var_9487_to_fp16 = const()[name = tensor<string, []>("op_9487_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_763_cast = mul(x = var_9486_cast, y = var_9487_to_fp16)[name = tensor<string, []>("aw_chunk_763_cast")];
            tensor<string, []> var_9490_equation_0 = const()[name = tensor<string, []>("op_9490_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_9490_cast = einsum(equation = var_9490_equation_0, values = (var_9400_cast, var_9366_cast))[name = tensor<string, []>("op_9490_cast")];
            tensor<fp16, []> var_9491_to_fp16 = const()[name = tensor<string, []>("op_9491_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_765_cast = mul(x = var_9490_cast, y = var_9491_to_fp16)[name = tensor<string, []>("aw_chunk_765_cast")];
            tensor<string, []> var_9494_equation_0 = const()[name = tensor<string, []>("op_9494_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_9494_cast = einsum(equation = var_9494_equation_0, values = (var_9400_cast, var_9367_cast))[name = tensor<string, []>("op_9494_cast")];
            tensor<fp16, []> var_9495_to_fp16 = const()[name = tensor<string, []>("op_9495_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_767_cast = mul(x = var_9494_cast, y = var_9495_to_fp16)[name = tensor<string, []>("aw_chunk_767_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_9497_cast = softmax(axis = var_8182, x = aw_chunk_737_cast)[name = tensor<string, []>("op_9497_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_9498_cast = softmax(axis = var_8182, x = aw_chunk_739_cast)[name = tensor<string, []>("op_9498_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_9499_cast = softmax(axis = var_8182, x = aw_chunk_741_cast)[name = tensor<string, []>("op_9499_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_9500_cast = softmax(axis = var_8182, x = aw_chunk_743_cast)[name = tensor<string, []>("op_9500_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_9501_cast = softmax(axis = var_8182, x = aw_chunk_745_cast)[name = tensor<string, []>("op_9501_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_9502_cast = softmax(axis = var_8182, x = aw_chunk_747_cast)[name = tensor<string, []>("op_9502_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_9503_cast = softmax(axis = var_8182, x = aw_chunk_749_cast)[name = tensor<string, []>("op_9503_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_9504_cast = softmax(axis = var_8182, x = aw_chunk_751_cast)[name = tensor<string, []>("op_9504_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_9505_cast = softmax(axis = var_8182, x = aw_chunk_753_cast)[name = tensor<string, []>("op_9505_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_9506_cast = softmax(axis = var_8182, x = aw_chunk_755_cast)[name = tensor<string, []>("op_9506_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_9507_cast = softmax(axis = var_8182, x = aw_chunk_757_cast)[name = tensor<string, []>("op_9507_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_9508_cast = softmax(axis = var_8182, x = aw_chunk_759_cast)[name = tensor<string, []>("op_9508_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_9509_cast = softmax(axis = var_8182, x = aw_chunk_761_cast)[name = tensor<string, []>("op_9509_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_9510_cast = softmax(axis = var_8182, x = aw_chunk_763_cast)[name = tensor<string, []>("op_9510_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_9511_cast = softmax(axis = var_8182, x = aw_chunk_765_cast)[name = tensor<string, []>("op_9511_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_9512_cast = softmax(axis = var_8182, x = aw_chunk_767_cast)[name = tensor<string, []>("op_9512_cast")];
            tensor<string, []> var_9514_equation_0 = const()[name = tensor<string, []>("op_9514_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9514_cast = einsum(equation = var_9514_equation_0, values = (var_9402_cast, var_9497_cast))[name = tensor<string, []>("op_9514_cast")];
            tensor<string, []> var_9516_equation_0 = const()[name = tensor<string, []>("op_9516_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9516_cast = einsum(equation = var_9516_equation_0, values = (var_9402_cast, var_9498_cast))[name = tensor<string, []>("op_9516_cast")];
            tensor<string, []> var_9518_equation_0 = const()[name = tensor<string, []>("op_9518_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9518_cast = einsum(equation = var_9518_equation_0, values = (var_9406_cast, var_9499_cast))[name = tensor<string, []>("op_9518_cast")];
            tensor<string, []> var_9520_equation_0 = const()[name = tensor<string, []>("op_9520_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9520_cast = einsum(equation = var_9520_equation_0, values = (var_9406_cast, var_9500_cast))[name = tensor<string, []>("op_9520_cast")];
            tensor<string, []> var_9522_equation_0 = const()[name = tensor<string, []>("op_9522_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9522_cast = einsum(equation = var_9522_equation_0, values = (var_9410_cast, var_9501_cast))[name = tensor<string, []>("op_9522_cast")];
            tensor<string, []> var_9524_equation_0 = const()[name = tensor<string, []>("op_9524_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9524_cast = einsum(equation = var_9524_equation_0, values = (var_9410_cast, var_9502_cast))[name = tensor<string, []>("op_9524_cast")];
            tensor<string, []> var_9526_equation_0 = const()[name = tensor<string, []>("op_9526_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9526_cast = einsum(equation = var_9526_equation_0, values = (var_9414_cast, var_9503_cast))[name = tensor<string, []>("op_9526_cast")];
            tensor<string, []> var_9528_equation_0 = const()[name = tensor<string, []>("op_9528_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9528_cast = einsum(equation = var_9528_equation_0, values = (var_9414_cast, var_9504_cast))[name = tensor<string, []>("op_9528_cast")];
            tensor<string, []> var_9530_equation_0 = const()[name = tensor<string, []>("op_9530_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9530_cast = einsum(equation = var_9530_equation_0, values = (var_9418_cast, var_9505_cast))[name = tensor<string, []>("op_9530_cast")];
            tensor<string, []> var_9532_equation_0 = const()[name = tensor<string, []>("op_9532_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9532_cast = einsum(equation = var_9532_equation_0, values = (var_9418_cast, var_9506_cast))[name = tensor<string, []>("op_9532_cast")];
            tensor<string, []> var_9534_equation_0 = const()[name = tensor<string, []>("op_9534_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9534_cast = einsum(equation = var_9534_equation_0, values = (var_9422_cast, var_9507_cast))[name = tensor<string, []>("op_9534_cast")];
            tensor<string, []> var_9536_equation_0 = const()[name = tensor<string, []>("op_9536_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9536_cast = einsum(equation = var_9536_equation_0, values = (var_9422_cast, var_9508_cast))[name = tensor<string, []>("op_9536_cast")];
            tensor<string, []> var_9538_equation_0 = const()[name = tensor<string, []>("op_9538_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9538_cast = einsum(equation = var_9538_equation_0, values = (var_9426_cast, var_9509_cast))[name = tensor<string, []>("op_9538_cast")];
            tensor<string, []> var_9540_equation_0 = const()[name = tensor<string, []>("op_9540_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9540_cast = einsum(equation = var_9540_equation_0, values = (var_9426_cast, var_9510_cast))[name = tensor<string, []>("op_9540_cast")];
            tensor<string, []> var_9542_equation_0 = const()[name = tensor<string, []>("op_9542_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9542_cast = einsum(equation = var_9542_equation_0, values = (var_9430_cast, var_9511_cast))[name = tensor<string, []>("op_9542_cast")];
            tensor<string, []> var_9544_equation_0 = const()[name = tensor<string, []>("op_9544_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9544_cast = einsum(equation = var_9544_equation_0, values = (var_9430_cast, var_9512_cast))[name = tensor<string, []>("op_9544_cast")];
            tensor<bool, []> var_9546_interleave_0 = const()[name = tensor<string, []>("op_9546_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_9546_cast = concat(axis = var_8160, interleave = var_9546_interleave_0, values = (var_9514_cast, var_9516_cast))[name = tensor<string, []>("op_9546_cast")];
            tensor<bool, []> var_9548_interleave_0 = const()[name = tensor<string, []>("op_9548_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_9548_cast = concat(axis = var_8160, interleave = var_9548_interleave_0, values = (var_9518_cast, var_9520_cast))[name = tensor<string, []>("op_9548_cast")];
            tensor<bool, []> var_9550_interleave_0 = const()[name = tensor<string, []>("op_9550_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_9550_cast = concat(axis = var_8160, interleave = var_9550_interleave_0, values = (var_9522_cast, var_9524_cast))[name = tensor<string, []>("op_9550_cast")];
            tensor<bool, []> var_9552_interleave_0 = const()[name = tensor<string, []>("op_9552_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_9552_cast = concat(axis = var_8160, interleave = var_9552_interleave_0, values = (var_9526_cast, var_9528_cast))[name = tensor<string, []>("op_9552_cast")];
            tensor<bool, []> var_9554_interleave_0 = const()[name = tensor<string, []>("op_9554_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_9554_cast = concat(axis = var_8160, interleave = var_9554_interleave_0, values = (var_9530_cast, var_9532_cast))[name = tensor<string, []>("op_9554_cast")];
            tensor<bool, []> var_9556_interleave_0 = const()[name = tensor<string, []>("op_9556_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_9556_cast = concat(axis = var_8160, interleave = var_9556_interleave_0, values = (var_9534_cast, var_9536_cast))[name = tensor<string, []>("op_9556_cast")];
            tensor<bool, []> var_9558_interleave_0 = const()[name = tensor<string, []>("op_9558_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_9558_cast = concat(axis = var_8160, interleave = var_9558_interleave_0, values = (var_9538_cast, var_9540_cast))[name = tensor<string, []>("op_9558_cast")];
            tensor<bool, []> var_9560_interleave_0 = const()[name = tensor<string, []>("op_9560_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_9560_cast = concat(axis = var_8160, interleave = var_9560_interleave_0, values = (var_9542_cast, var_9544_cast))[name = tensor<string, []>("op_9560_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_8182, interleave = input_411_interleave_0, values = (var_9546_cast, var_9548_cast, var_9550_cast, var_9552_cast, var_9554_cast, var_9556_cast, var_9558_cast, var_9560_cast))[name = tensor<string, []>("input_411_cast")];
            tensor<int32, [2]> var_9566 = const()[name = tensor<string, []>("op_9566"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9568 = const()[name = tensor<string, []>("op_9568"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_9570_pad_type_0 = const()[name = tensor<string, []>("op_9570_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_9570_pad_0 = const()[name = tensor<string, []>("op_9570_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(609014720))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(609321984))), name = tensor<string, []>("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            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, []>(609322176)))];
            tensor<fp16, [2, 640, 1, 1024]> var_9570_cast = conv(bias = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_9568, groups = var_8182, pad = var_9570_pad_0, pad_type = var_9570_pad_type_0, strides = var_9566, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_411_cast)[name = tensor<string, []>("op_9570_cast")];
            tensor<fp16, [2, 640, 1, 1024]> inputs_71_cast = add(x = var_9570_cast, y = inputs_69_cast)[name = tensor<string, []>("inputs_71_cast")];
            tensor<int32, [1]> var_9574 = const()[name = tensor<string, []>("op_9574"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_71_cast = reduce_mean(axes = var_9574, keep_dims = var_8177, x = inputs_71_cast)[name = tensor<string, []>("channels_mean_71_cast")];
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_71_cast = sub(x = inputs_71_cast, y = channels_mean_71_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_71_cast")];
            tensor<int32, [1]> var_9578 = const()[name = tensor<string, []>("op_9578"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_9579_cast = reduce_mean(axes = var_9578, keep_dims = var_8177, x = zero_mean_sq_71_cast)[name = tensor<string, []>("op_9579_cast")];
            tensor<fp16, []> var_9580_to_fp16 = const()[name = tensor<string, []>("op_9580_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_9581_cast = add(x = var_9579_cast, y = var_9580_to_fp16)[name = tensor<string, []>("op_9581_cast")];
            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_9581_cast)[name = tensor<string, []>("denom_71_cast")];
            tensor<fp16, [2, 640, 1, 1024]> out_71_cast = mul(x = zero_mean_71_cast, y = denom_71_cast)[name = tensor<string, []>("out_71_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_9585_to_fp16 = const()[name = tensor<string, []>("op_9585_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(609323520)))];
            tensor<fp16, [2, 640, 1, 1024]> var_9586_cast = add(x = out_71_cast, y = var_9585_to_fp16)[name = tensor<string, []>("op_9586_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, []>(609324864)))];
            tensor<fp16, [2, 640, 1, 1024]> input_413_cast = mul(x = var_9586_cast, y = var_9588_to_fp16)[name = tensor<string, []>("input_413_cast")];
            tensor<int32, [2]> var_9596 = const()[name = tensor<string, []>("op_9596"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9598 = const()[name = tensor<string, []>("op_9598"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_9600_pad_type_0 = const()[name = tensor<string, []>("op_9600_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_9600_pad_0 = const()[name = tensor<string, []>("op_9600_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2457600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(609326208))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(611783872))), name = tensor<string, []>("up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([5120, 640, 1, 1])];
            tensor<fp16, [5120]> up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3840]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(611784064))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(611787968))), name = tensor<string, []>("up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([5120])];
            tensor<fp16, [2, 5120, 1, 1024]> var_9600_cast = conv(bias = up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_9598, groups = var_8182, pad = var_9600_pad_0, pad_type = var_9600_pad_type_0, strides = var_9596, weight = up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_413_cast)[name = tensor<string, []>("op_9600_cast")];
            tensor<int32, [2]> var_9601_split_sizes_0 = const()[name = tensor<string, []>("op_9601_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_9601_axis_0 = const()[name = tensor<string, []>("op_9601_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 1024]> var_9601_cast_0, tensor<fp16, [2, 2560, 1, 1024]> var_9601_cast_1 = split(axis = var_9601_axis_0, split_sizes = var_9601_split_sizes_0, x = var_9600_cast)[name = tensor<string, []>("op_9601_cast")];
            tensor<string, []> var_9603_mode_0 = const()[name = tensor<string, []>("op_9603_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 1024]> var_9603_cast = gelu(mode = var_9603_mode_0, x = var_9601_cast_1)[name = tensor<string, []>("op_9603_cast")];
            tensor<fp16, [2, 2560, 1, 1024]> input_415_cast = mul(x = var_9601_cast_0, y = var_9603_cast)[name = tensor<string, []>("input_415_cast")];
            tensor<int32, [2]> var_9607 = const()[name = tensor<string, []>("op_9607"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9609 = const()[name = tensor<string, []>("op_9609"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_9611_pad_type_0 = const()[name = tensor<string, []>("op_9611_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_9611_pad_0 = const()[name = tensor<string, []>("op_9611_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(611788160))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(613017024))), name = tensor<string, []>("up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 2560, 1, 1])];
            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, []>(613017216)))];
            tensor<fp16, [2, 640, 1, 1024]> var_9611_cast = conv(bias = up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_9609, groups = var_8182, pad = var_9611_pad_0, pad_type = var_9611_pad_type_0, strides = var_9607, weight = up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_415_cast)[name = tensor<string, []>("op_9611_cast")];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_253_cast = add(x = var_9611_cast, y = inputs_71_cast)[name = tensor<string, []>("hidden_states_253_cast")];
            tensor<int32, [4]> var_9613 = const()[name = tensor<string, []>("op_9613"), val = tensor<int32, [4]>([2, 640, 32, 32])];
            tensor<fp16, [2, 640, 32, 32]> input_417_cast = reshape(shape = var_9613, x = hidden_states_253_cast)[name = tensor<string, []>("input_417_cast")];
            tensor<int32, [2]> var_9617 = const()[name = tensor<string, []>("op_9617"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9619 = const()[name = tensor<string, []>("op_9619"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(613018560))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(613325824))), name = tensor<string, []>("up_blocks_2_attentions_1_proj_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            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, []>(613326016)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_255_cast = conv(bias = up_blocks_2_attentions_1_proj_out_bias_to_fp16, dilations = var_9619, groups = var_8182, pad = hidden_states_255_pad_0, pad_type = hidden_states_255_pad_type_0, strides = var_9617, weight = up_blocks_2_attentions_1_proj_out_weight_to_fp16_palettized, x = input_417_cast)[name = tensor<string, []>("hidden_states_255_cast")];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_257_cast = add(x = hidden_states_255_cast, y = hidden_states_243_cast)[name = tensor<string, []>("hidden_states_257_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_8182, interleave = input_419_interleave_0, values = (hidden_states_257_cast, input_63_cast))[name = tensor<string, []>("input_419_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)[name = tensor<string, []>("reshape_192_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)[name = tensor<string, []>("reduce_mean_144_cast")];
            tensor<fp16, [2, 32, 30, 32, 32]> sub_96_cast = sub(x = reshape_192_cast, y = reduce_mean_144_cast)[name = tensor<string, []>("sub_96_cast")];
            tensor<fp16, [2, 32, 30, 32, 32]> square_48_cast = square(x = sub_96_cast)[name = tensor<string, []>("square_48_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)[name = tensor<string, []>("reduce_mean_146_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)[name = tensor<string, []>("add_96_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_48_cast = sqrt(x = add_96_cast)[name = tensor<string, []>("sqrt_48_cast")];
            tensor<fp16, [2, 32, 30, 32, 32]> real_div_48_cast = real_div(x = sub_96_cast, y = sqrt_48_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_193_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, []>(613327360)))];
            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, []>(613329344)))];
            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, []>(613331328)))];
            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, []>(613333312)))];
            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)[name = tensor<string, []>("add_97_cast")];
            tensor<fp16, [2, 960, 32, 32]> input_423_cast = silu(x = add_97_cast)[name = tensor<string, []>("input_423_cast")];
            tensor<int32, [2]> var_9637 = const()[name = tensor<string, []>("op_9637"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9639 = const()[name = tensor<string, []>("op_9639"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4147200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(613335296))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(617482560))), name = tensor<string, []>("up_blocks_2_resnets_2_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 960, 3, 3])];
            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, []>(617482752)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_259_cast = conv(bias = up_blocks_2_resnets_2_conv1_bias_to_fp16, dilations = var_9639, groups = var_8182, pad = hidden_states_259_pad_0, pad_type = hidden_states_259_pad_type_0, strides = var_9637, weight = up_blocks_2_resnets_2_conv1_weight_to_fp16_palettized, x = input_423_cast)[name = tensor<string, []>("hidden_states_259_cast")];
            tensor<int32, [2]> var_9645 = const()[name = tensor<string, []>("op_9645"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9647 = const()[name = tensor<string, []>("op_9647"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [614400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(617484096))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(618098560))), name = tensor<string, []>("up_blocks_2_resnets_2_time_emb_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 1280, 1, 1])];
            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, []>(618098752)))];
            tensor<fp16, [2, 640, 1, 1]> temb_37_cast = conv(bias = up_blocks_2_resnets_2_time_emb_proj_bias_to_fp16, dilations = var_9647, groups = var_8182, pad = temb_37_pad_0, pad_type = temb_37_pad_type_0, strides = var_9645, weight = up_blocks_2_resnets_2_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor<string, []>("temb_37_cast")];
            tensor<fp16, [2, 640, 32, 32]> input_427_cast = add(x = hidden_states_259_cast, y = temb_37_cast)[name = tensor<string, []>("input_427_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)[name = tensor<string, []>("reshape_196_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)[name = tensor<string, []>("reduce_mean_147_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> sub_98_cast = sub(x = reshape_196_cast, y = reduce_mean_147_cast)[name = tensor<string, []>("sub_98_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> square_49_cast = square(x = sub_98_cast)[name = tensor<string, []>("square_49_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)[name = tensor<string, []>("reduce_mean_149_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)[name = tensor<string, []>("add_98_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_49_cast = sqrt(x = add_98_cast)[name = tensor<string, []>("sqrt_49_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> real_div_49_cast = real_div(x = sub_98_cast, y = sqrt_49_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_197_cast")];
            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, []>(618100096)))];
            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, []>(618101440)))];
            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_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_197_cast)[name = tensor<string, []>("add_99_cast")];
            tensor<fp16, [2, 640, 32, 32]> input_431_cast = silu(x = add_99_cast)[name = tensor<string, []>("input_431_cast")];
            tensor<int32, [2]> var_9657 = const()[name = tensor<string, []>("op_9657"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9659 = const()[name = tensor<string, []>("op_9659"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2764800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(618102784))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(620867648))), name = tensor<string, []>("up_blocks_2_resnets_2_conv2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 3, 3])];
            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, []>(620867840)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_261_cast = conv(bias = up_blocks_2_resnets_2_conv2_bias_to_fp16, dilations = var_9659, groups = var_8182, pad = hidden_states_261_pad_0, pad_type = hidden_states_261_pad_type_0, strides = var_9657, weight = up_blocks_2_resnets_2_conv2_weight_to_fp16_palettized, x = input_431_cast)[name = tensor<string, []>("hidden_states_261_cast")];
            tensor<int32, [2]> var_9664 = const()[name = tensor<string, []>("op_9664"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9666 = const()[name = tensor<string, []>("op_9666"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [460800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(620869184))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(621330048))), name = tensor<string, []>("up_blocks_2_resnets_2_conv_shortcut_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 960, 1, 1])];
            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, []>(621330240)))];
            tensor<fp16, [2, 640, 32, 32]> x_21_cast = conv(bias = up_blocks_2_resnets_2_conv_shortcut_bias_to_fp16, dilations = var_9666, groups = var_8182, pad = x_21_pad_0, pad_type = x_21_pad_type_0, strides = var_9664, weight = up_blocks_2_resnets_2_conv_shortcut_weight_to_fp16_palettized, x = input_419_cast)[name = tensor<string, []>("x_21_cast")];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_263_cast = add(x = x_21_cast, y = hidden_states_261_cast)[name = tensor<string, []>("hidden_states_263_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)[name = tensor<string, []>("reshape_200_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)[name = tensor<string, []>("reduce_mean_150_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> sub_100_cast = sub(x = reshape_200_cast, y = reduce_mean_150_cast)[name = tensor<string, []>("sub_100_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> square_50_cast = square(x = sub_100_cast)[name = tensor<string, []>("square_50_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)[name = tensor<string, []>("reduce_mean_152_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)[name = tensor<string, []>("add_100_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_50_cast = sqrt(x = add_100_cast)[name = tensor<string, []>("sqrt_50_cast")];
            tensor<fp16, [2, 32, 20, 32, 32]> real_div_50_cast = real_div(x = sub_100_cast, y = sqrt_50_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_201_cast")];
            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, []>(621331584)))];
            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, []>(621332928)))];
            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_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_201_cast)[name = tensor<string, []>("add_101_cast")];
            tensor<int32, [2]> var_9686 = const()[name = tensor<string, []>("op_9686"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9688 = const()[name = tensor<string, []>("op_9688"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(621334272))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(621641536))), name = tensor<string, []>("up_blocks_2_attentions_2_proj_in_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            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, []>(621641728)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_265_cast = conv(bias = up_blocks_2_attentions_2_proj_in_bias_to_fp16, dilations = var_9688, groups = var_8182, pad = hidden_states_265_pad_0, pad_type = hidden_states_265_pad_type_0, strides = var_9686, weight = up_blocks_2_attentions_2_proj_in_weight_to_fp16_palettized, x = add_101_cast)[name = tensor<string, []>("hidden_states_265_cast")];
            tensor<int32, [4]> var_9693 = const()[name = tensor<string, []>("op_9693"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<fp16, [2, 640, 1, 1024]> inputs_73_cast = reshape(shape = var_9693, x = hidden_states_265_cast)[name = tensor<string, []>("inputs_73_cast")];
            tensor<int32, [1]> var_9703 = const()[name = tensor<string, []>("op_9703"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_73_cast = reduce_mean(axes = var_9703, keep_dims = var_8177, x = inputs_73_cast)[name = tensor<string, []>("channels_mean_73_cast")];
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_73_cast = sub(x = inputs_73_cast, y = channels_mean_73_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_73_cast")];
            tensor<int32, [1]> var_9707 = const()[name = tensor<string, []>("op_9707"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_9708_cast = reduce_mean(axes = var_9707, keep_dims = var_8177, x = zero_mean_sq_73_cast)[name = tensor<string, []>("op_9708_cast")];
            tensor<fp16, []> var_9709_to_fp16 = const()[name = tensor<string, []>("op_9709_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_9710_cast = add(x = var_9708_cast, y = var_9709_to_fp16)[name = tensor<string, []>("op_9710_cast")];
            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_9710_cast)[name = tensor<string, []>("denom_73_cast")];
            tensor<fp16, [2, 640, 1, 1024]> out_73_cast = mul(x = zero_mean_73_cast, y = denom_73_cast)[name = tensor<string, []>("out_73_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_9714_to_fp16 = const()[name = tensor<string, []>("op_9714_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(621643072)))];
            tensor<fp16, [2, 640, 1, 1024]> var_9715_cast = add(x = out_73_cast, y = var_9714_to_fp16)[name = tensor<string, []>("op_9715_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_9717_to_fp16 = const()[name = tensor<string, []>("op_9717_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(621644416)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_267_cast = mul(x = var_9715_cast, y = var_9717_to_fp16)[name = tensor<string, []>("hidden_states_267_cast")];
            tensor<int32, [2]> var_9724 = const()[name = tensor<string, []>("op_9724"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9726 = const()[name = tensor<string, []>("op_9726"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(621645760))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(621953024))), name = tensor<string, []>("up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 1024]> q_49_cast = conv(dilations = var_9726, groups = var_8182, pad = q_49_pad_0, pad_type = q_49_pad_type_0, strides = var_9724, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_267_cast)[name = tensor<string, []>("q_49_cast")];
            tensor<int32, [2]> var_9730 = const()[name = tensor<string, []>("op_9730"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9732 = const()[name = tensor<string, []>("op_9732"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(621953216))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(622260480))), name = tensor<string, []>("up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 1024]> k_97_cast = conv(dilations = var_9732, groups = var_8182, pad = k_97_pad_0, pad_type = k_97_pad_type_0, strides = var_9730, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_267_cast)[name = tensor<string, []>("k_97_cast")];
            tensor<int32, [2]> var_9736 = const()[name = tensor<string, []>("op_9736"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9738 = const()[name = tensor<string, []>("op_9738"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(622260672))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(622567936))), name = tensor<string, []>("up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 1024]> v_49_cast = conv(dilations = var_9738, groups = var_8182, pad = v_49_pad_0, pad_type = v_49_pad_type_0, strides = var_9736, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_267_cast)[name = tensor<string, []>("v_49_cast")];
            tensor<int32, [4]> var_9742_begin_0 = const()[name = tensor<string, []>("op_9742_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9742_end_0 = const()[name = tensor<string, []>("op_9742_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9742_end_mask_0 = const()[name = tensor<string, []>("op_9742_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9742_cast = slice_by_index(begin = var_9742_begin_0, end = var_9742_end_0, end_mask = var_9742_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_9742_cast")];
            tensor<int32, [4]> var_9746_begin_0 = const()[name = tensor<string, []>("op_9746_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_9746_end_0 = const()[name = tensor<string, []>("op_9746_end_0"), val = tensor<int32, [4]>([2, 160, 1, 1024])];
            tensor<bool, [4]> var_9746_end_mask_0 = const()[name = tensor<string, []>("op_9746_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9746_cast = slice_by_index(begin = var_9746_begin_0, end = var_9746_end_0, end_mask = var_9746_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_9746_cast")];
            tensor<int32, [4]> var_9750_begin_0 = const()[name = tensor<string, []>("op_9750_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_9750_end_0 = const()[name = tensor<string, []>("op_9750_end_0"), val = tensor<int32, [4]>([2, 240, 1, 1024])];
            tensor<bool, [4]> var_9750_end_mask_0 = const()[name = tensor<string, []>("op_9750_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9750_cast = slice_by_index(begin = var_9750_begin_0, end = var_9750_end_0, end_mask = var_9750_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_9750_cast")];
            tensor<int32, [4]> var_9754_begin_0 = const()[name = tensor<string, []>("op_9754_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_9754_end_0 = const()[name = tensor<string, []>("op_9754_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_9754_end_mask_0 = const()[name = tensor<string, []>("op_9754_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9754_cast = slice_by_index(begin = var_9754_begin_0, end = var_9754_end_0, end_mask = var_9754_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_9754_cast")];
            tensor<int32, [4]> var_9758_begin_0 = const()[name = tensor<string, []>("op_9758_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_9758_end_0 = const()[name = tensor<string, []>("op_9758_end_0"), val = tensor<int32, [4]>([2, 400, 1, 1024])];
            tensor<bool, [4]> var_9758_end_mask_0 = const()[name = tensor<string, []>("op_9758_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9758_cast = slice_by_index(begin = var_9758_begin_0, end = var_9758_end_0, end_mask = var_9758_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_9758_cast")];
            tensor<int32, [4]> var_9762_begin_0 = const()[name = tensor<string, []>("op_9762_begin_0"), val = tensor<int32, [4]>([0, 400, 0, 0])];
            tensor<int32, [4]> var_9762_end_0 = const()[name = tensor<string, []>("op_9762_end_0"), val = tensor<int32, [4]>([2, 480, 1, 1024])];
            tensor<bool, [4]> var_9762_end_mask_0 = const()[name = tensor<string, []>("op_9762_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9762_cast = slice_by_index(begin = var_9762_begin_0, end = var_9762_end_0, end_mask = var_9762_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_9762_cast")];
            tensor<int32, [4]> var_9766_begin_0 = const()[name = tensor<string, []>("op_9766_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_9766_end_0 = const()[name = tensor<string, []>("op_9766_end_0"), val = tensor<int32, [4]>([2, 560, 1, 1024])];
            tensor<bool, [4]> var_9766_end_mask_0 = const()[name = tensor<string, []>("op_9766_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9766_cast = slice_by_index(begin = var_9766_begin_0, end = var_9766_end_0, end_mask = var_9766_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_9766_cast")];
            tensor<int32, [4]> var_9770_begin_0 = const()[name = tensor<string, []>("op_9770_begin_0"), val = tensor<int32, [4]>([0, 560, 0, 0])];
            tensor<int32, [4]> var_9770_end_0 = const()[name = tensor<string, []>("op_9770_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_9770_end_mask_0 = const()[name = tensor<string, []>("op_9770_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9770_cast = slice_by_index(begin = var_9770_begin_0, end = var_9770_end_0, end_mask = var_9770_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_9770_cast")];
            tensor<int32, [4]> var_9773_begin_0 = const()[name = tensor<string, []>("op_9773_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9773_end_0 = const()[name = tensor<string, []>("op_9773_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_9773_end_mask_0 = const()[name = tensor<string, []>("op_9773_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9773_cast = slice_by_index(begin = var_9773_begin_0, end = var_9773_end_0, end_mask = var_9773_end_mask_0, x = var_9742_cast)[name = tensor<string, []>("op_9773_cast")];
            tensor<int32, [4]> var_9774_begin_0 = const()[name = tensor<string, []>("op_9774_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9774_end_0 = const()[name = tensor<string, []>("op_9774_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9774_end_mask_0 = const()[name = tensor<string, []>("op_9774_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9774_cast = slice_by_index(begin = var_9774_begin_0, end = var_9774_end_0, end_mask = var_9774_end_mask_0, x = var_9742_cast)[name = tensor<string, []>("op_9774_cast")];
            tensor<int32, [4]> var_9775_begin_0 = const()[name = tensor<string, []>("op_9775_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9775_end_0 = const()[name = tensor<string, []>("op_9775_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_9775_end_mask_0 = const()[name = tensor<string, []>("op_9775_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9775_cast = slice_by_index(begin = var_9775_begin_0, end = var_9775_end_0, end_mask = var_9775_end_mask_0, x = var_9746_cast)[name = tensor<string, []>("op_9775_cast")];
            tensor<int32, [4]> var_9776_begin_0 = const()[name = tensor<string, []>("op_9776_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9776_end_0 = const()[name = tensor<string, []>("op_9776_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9776_end_mask_0 = const()[name = tensor<string, []>("op_9776_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9776_cast = slice_by_index(begin = var_9776_begin_0, end = var_9776_end_0, end_mask = var_9776_end_mask_0, x = var_9746_cast)[name = tensor<string, []>("op_9776_cast")];
            tensor<int32, [4]> var_9777_begin_0 = const()[name = tensor<string, []>("op_9777_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9777_end_0 = const()[name = tensor<string, []>("op_9777_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_9777_end_mask_0 = const()[name = tensor<string, []>("op_9777_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9777_cast = slice_by_index(begin = var_9777_begin_0, end = var_9777_end_0, end_mask = var_9777_end_mask_0, x = var_9750_cast)[name = tensor<string, []>("op_9777_cast")];
            tensor<int32, [4]> var_9778_begin_0 = const()[name = tensor<string, []>("op_9778_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9778_end_0 = const()[name = tensor<string, []>("op_9778_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9778_end_mask_0 = const()[name = tensor<string, []>("op_9778_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9778_cast = slice_by_index(begin = var_9778_begin_0, end = var_9778_end_0, end_mask = var_9778_end_mask_0, x = var_9750_cast)[name = tensor<string, []>("op_9778_cast")];
            tensor<int32, [4]> var_9779_begin_0 = const()[name = tensor<string, []>("op_9779_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9779_end_0 = const()[name = tensor<string, []>("op_9779_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_9779_end_mask_0 = const()[name = tensor<string, []>("op_9779_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9779_cast = slice_by_index(begin = var_9779_begin_0, end = var_9779_end_0, end_mask = var_9779_end_mask_0, x = var_9754_cast)[name = tensor<string, []>("op_9779_cast")];
            tensor<int32, [4]> var_9780_begin_0 = const()[name = tensor<string, []>("op_9780_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9780_end_0 = const()[name = tensor<string, []>("op_9780_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9780_end_mask_0 = const()[name = tensor<string, []>("op_9780_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9780_cast = slice_by_index(begin = var_9780_begin_0, end = var_9780_end_0, end_mask = var_9780_end_mask_0, x = var_9754_cast)[name = tensor<string, []>("op_9780_cast")];
            tensor<int32, [4]> var_9781_begin_0 = const()[name = tensor<string, []>("op_9781_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9781_end_0 = const()[name = tensor<string, []>("op_9781_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_9781_end_mask_0 = const()[name = tensor<string, []>("op_9781_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9781_cast = slice_by_index(begin = var_9781_begin_0, end = var_9781_end_0, end_mask = var_9781_end_mask_0, x = var_9758_cast)[name = tensor<string, []>("op_9781_cast")];
            tensor<int32, [4]> var_9782_begin_0 = const()[name = tensor<string, []>("op_9782_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9782_end_0 = const()[name = tensor<string, []>("op_9782_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9782_end_mask_0 = const()[name = tensor<string, []>("op_9782_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9782_cast = slice_by_index(begin = var_9782_begin_0, end = var_9782_end_0, end_mask = var_9782_end_mask_0, x = var_9758_cast)[name = tensor<string, []>("op_9782_cast")];
            tensor<int32, [4]> var_9783_begin_0 = const()[name = tensor<string, []>("op_9783_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9783_end_0 = const()[name = tensor<string, []>("op_9783_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_9783_end_mask_0 = const()[name = tensor<string, []>("op_9783_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9783_cast = slice_by_index(begin = var_9783_begin_0, end = var_9783_end_0, end_mask = var_9783_end_mask_0, x = var_9762_cast)[name = tensor<string, []>("op_9783_cast")];
            tensor<int32, [4]> var_9784_begin_0 = const()[name = tensor<string, []>("op_9784_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9784_end_0 = const()[name = tensor<string, []>("op_9784_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9784_end_mask_0 = const()[name = tensor<string, []>("op_9784_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9784_cast = slice_by_index(begin = var_9784_begin_0, end = var_9784_end_0, end_mask = var_9784_end_mask_0, x = var_9762_cast)[name = tensor<string, []>("op_9784_cast")];
            tensor<int32, [4]> var_9785_begin_0 = const()[name = tensor<string, []>("op_9785_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9785_end_0 = const()[name = tensor<string, []>("op_9785_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_9785_end_mask_0 = const()[name = tensor<string, []>("op_9785_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9785_cast = slice_by_index(begin = var_9785_begin_0, end = var_9785_end_0, end_mask = var_9785_end_mask_0, x = var_9766_cast)[name = tensor<string, []>("op_9785_cast")];
            tensor<int32, [4]> var_9786_begin_0 = const()[name = tensor<string, []>("op_9786_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9786_end_0 = const()[name = tensor<string, []>("op_9786_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9786_end_mask_0 = const()[name = tensor<string, []>("op_9786_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9786_cast = slice_by_index(begin = var_9786_begin_0, end = var_9786_end_0, end_mask = var_9786_end_mask_0, x = var_9766_cast)[name = tensor<string, []>("op_9786_cast")];
            tensor<int32, [4]> var_9787_begin_0 = const()[name = tensor<string, []>("op_9787_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9787_end_0 = const()[name = tensor<string, []>("op_9787_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_9787_end_mask_0 = const()[name = tensor<string, []>("op_9787_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9787_cast = slice_by_index(begin = var_9787_begin_0, end = var_9787_end_0, end_mask = var_9787_end_mask_0, x = var_9770_cast)[name = tensor<string, []>("op_9787_cast")];
            tensor<int32, [4]> var_9788_begin_0 = const()[name = tensor<string, []>("op_9788_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9788_end_0 = const()[name = tensor<string, []>("op_9788_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9788_end_mask_0 = const()[name = tensor<string, []>("op_9788_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_9788_cast = slice_by_index(begin = var_9788_begin_0, end = var_9788_end_0, end_mask = var_9788_end_mask_0, x = var_9770_cast)[name = tensor<string, []>("op_9788_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_9793_begin_0 = const()[name = tensor<string, []>("op_9793_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9793_end_0 = const()[name = tensor<string, []>("op_9793_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 80])];
            tensor<bool, [4]> var_9793_end_mask_0 = const()[name = tensor<string, []>("op_9793_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)[name = tensor<string, []>("transpose_7")];
            tensor<fp16, [2, 1024, 1, 80]> var_9793_cast = slice_by_index(begin = var_9793_begin_0, end = var_9793_end_0, end_mask = var_9793_end_mask_0, x = transpose_7)[name = tensor<string, []>("op_9793_cast")];
            tensor<int32, [4]> var_9797_begin_0 = const()[name = tensor<string, []>("op_9797_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 80])];
            tensor<int32, [4]> var_9797_end_0 = const()[name = tensor<string, []>("op_9797_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 160])];
            tensor<bool, [4]> var_9797_end_mask_0 = const()[name = tensor<string, []>("op_9797_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_9797_cast = slice_by_index(begin = var_9797_begin_0, end = var_9797_end_0, end_mask = var_9797_end_mask_0, x = transpose_7)[name = tensor<string, []>("op_9797_cast")];
            tensor<int32, [4]> var_9801_begin_0 = const()[name = tensor<string, []>("op_9801_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_9801_end_0 = const()[name = tensor<string, []>("op_9801_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 240])];
            tensor<bool, [4]> var_9801_end_mask_0 = const()[name = tensor<string, []>("op_9801_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_9801_cast = slice_by_index(begin = var_9801_begin_0, end = var_9801_end_0, end_mask = var_9801_end_mask_0, x = transpose_7)[name = tensor<string, []>("op_9801_cast")];
            tensor<int32, [4]> var_9805_begin_0 = const()[name = tensor<string, []>("op_9805_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 240])];
            tensor<int32, [4]> var_9805_end_0 = const()[name = tensor<string, []>("op_9805_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_9805_end_mask_0 = const()[name = tensor<string, []>("op_9805_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_9805_cast = slice_by_index(begin = var_9805_begin_0, end = var_9805_end_0, end_mask = var_9805_end_mask_0, x = transpose_7)[name = tensor<string, []>("op_9805_cast")];
            tensor<int32, [4]> var_9809_begin_0 = const()[name = tensor<string, []>("op_9809_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_9809_end_0 = const()[name = tensor<string, []>("op_9809_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 400])];
            tensor<bool, [4]> var_9809_end_mask_0 = const()[name = tensor<string, []>("op_9809_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_9809_cast = slice_by_index(begin = var_9809_begin_0, end = var_9809_end_0, end_mask = var_9809_end_mask_0, x = transpose_7)[name = tensor<string, []>("op_9809_cast")];
            tensor<int32, [4]> var_9813_begin_0 = const()[name = tensor<string, []>("op_9813_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 400])];
            tensor<int32, [4]> var_9813_end_0 = const()[name = tensor<string, []>("op_9813_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 480])];
            tensor<bool, [4]> var_9813_end_mask_0 = const()[name = tensor<string, []>("op_9813_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_9813_cast = slice_by_index(begin = var_9813_begin_0, end = var_9813_end_0, end_mask = var_9813_end_mask_0, x = transpose_7)[name = tensor<string, []>("op_9813_cast")];
            tensor<int32, [4]> var_9817_begin_0 = const()[name = tensor<string, []>("op_9817_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 480])];
            tensor<int32, [4]> var_9817_end_0 = const()[name = tensor<string, []>("op_9817_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 560])];
            tensor<bool, [4]> var_9817_end_mask_0 = const()[name = tensor<string, []>("op_9817_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_9817_cast = slice_by_index(begin = var_9817_begin_0, end = var_9817_end_0, end_mask = var_9817_end_mask_0, x = transpose_7)[name = tensor<string, []>("op_9817_cast")];
            tensor<int32, [4]> var_9821_begin_0 = const()[name = tensor<string, []>("op_9821_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 560])];
            tensor<int32, [4]> var_9821_end_0 = const()[name = tensor<string, []>("op_9821_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_9821_end_mask_0 = const()[name = tensor<string, []>("op_9821_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 80]> var_9821_cast = slice_by_index(begin = var_9821_begin_0, end = var_9821_end_0, end_mask = var_9821_end_mask_0, x = transpose_7)[name = tensor<string, []>("op_9821_cast")];
            tensor<int32, [4]> var_9823_begin_0 = const()[name = tensor<string, []>("op_9823_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9823_end_0 = const()[name = tensor<string, []>("op_9823_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_9823_end_mask_0 = const()[name = tensor<string, []>("op_9823_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9823_cast = slice_by_index(begin = var_9823_begin_0, end = var_9823_end_0, end_mask = var_9823_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_9823_cast")];
            tensor<int32, [4]> var_9827_begin_0 = const()[name = tensor<string, []>("op_9827_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_9827_end_0 = const()[name = tensor<string, []>("op_9827_end_0"), val = tensor<int32, [4]>([2, 160, 1, 1024])];
            tensor<bool, [4]> var_9827_end_mask_0 = const()[name = tensor<string, []>("op_9827_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9827_cast = slice_by_index(begin = var_9827_begin_0, end = var_9827_end_0, end_mask = var_9827_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_9827_cast")];
            tensor<int32, [4]> var_9831_begin_0 = const()[name = tensor<string, []>("op_9831_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_9831_end_0 = const()[name = tensor<string, []>("op_9831_end_0"), val = tensor<int32, [4]>([2, 240, 1, 1024])];
            tensor<bool, [4]> var_9831_end_mask_0 = const()[name = tensor<string, []>("op_9831_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9831_cast = slice_by_index(begin = var_9831_begin_0, end = var_9831_end_0, end_mask = var_9831_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_9831_cast")];
            tensor<int32, [4]> var_9835_begin_0 = const()[name = tensor<string, []>("op_9835_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_9835_end_0 = const()[name = tensor<string, []>("op_9835_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_9835_end_mask_0 = const()[name = tensor<string, []>("op_9835_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9835_cast = slice_by_index(begin = var_9835_begin_0, end = var_9835_end_0, end_mask = var_9835_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_9835_cast")];
            tensor<int32, [4]> var_9839_begin_0 = const()[name = tensor<string, []>("op_9839_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_9839_end_0 = const()[name = tensor<string, []>("op_9839_end_0"), val = tensor<int32, [4]>([2, 400, 1, 1024])];
            tensor<bool, [4]> var_9839_end_mask_0 = const()[name = tensor<string, []>("op_9839_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9839_cast = slice_by_index(begin = var_9839_begin_0, end = var_9839_end_0, end_mask = var_9839_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_9839_cast")];
            tensor<int32, [4]> var_9843_begin_0 = const()[name = tensor<string, []>("op_9843_begin_0"), val = tensor<int32, [4]>([0, 400, 0, 0])];
            tensor<int32, [4]> var_9843_end_0 = const()[name = tensor<string, []>("op_9843_end_0"), val = tensor<int32, [4]>([2, 480, 1, 1024])];
            tensor<bool, [4]> var_9843_end_mask_0 = const()[name = tensor<string, []>("op_9843_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9843_cast = slice_by_index(begin = var_9843_begin_0, end = var_9843_end_0, end_mask = var_9843_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_9843_cast")];
            tensor<int32, [4]> var_9847_begin_0 = const()[name = tensor<string, []>("op_9847_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_9847_end_0 = const()[name = tensor<string, []>("op_9847_end_0"), val = tensor<int32, [4]>([2, 560, 1, 1024])];
            tensor<bool, [4]> var_9847_end_mask_0 = const()[name = tensor<string, []>("op_9847_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9847_cast = slice_by_index(begin = var_9847_begin_0, end = var_9847_end_0, end_mask = var_9847_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_9847_cast")];
            tensor<int32, [4]> var_9851_begin_0 = const()[name = tensor<string, []>("op_9851_begin_0"), val = tensor<int32, [4]>([0, 560, 0, 0])];
            tensor<int32, [4]> var_9851_end_0 = const()[name = tensor<string, []>("op_9851_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_9851_end_mask_0 = const()[name = tensor<string, []>("op_9851_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_9851_cast = slice_by_index(begin = var_9851_begin_0, end = var_9851_end_0, end_mask = var_9851_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_9851_cast")];
            tensor<string, []> var_9855_equation_0 = const()[name = tensor<string, []>("op_9855_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9855_cast = einsum(equation = var_9855_equation_0, values = (var_9793_cast, var_9773_cast))[name = tensor<string, []>("op_9855_cast")];
            tensor<fp16, []> var_9856_to_fp16 = const()[name = tensor<string, []>("op_9856_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_769_cast = mul(x = var_9855_cast, y = var_9856_to_fp16)[name = tensor<string, []>("aw_chunk_769_cast")];
            tensor<string, []> var_9859_equation_0 = const()[name = tensor<string, []>("op_9859_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9859_cast = einsum(equation = var_9859_equation_0, values = (var_9793_cast, var_9774_cast))[name = tensor<string, []>("op_9859_cast")];
            tensor<fp16, []> var_9860_to_fp16 = const()[name = tensor<string, []>("op_9860_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_771_cast = mul(x = var_9859_cast, y = var_9860_to_fp16)[name = tensor<string, []>("aw_chunk_771_cast")];
            tensor<string, []> var_9863_equation_0 = const()[name = tensor<string, []>("op_9863_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9863_cast = einsum(equation = var_9863_equation_0, values = (var_9797_cast, var_9775_cast))[name = tensor<string, []>("op_9863_cast")];
            tensor<fp16, []> var_9864_to_fp16 = const()[name = tensor<string, []>("op_9864_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_773_cast = mul(x = var_9863_cast, y = var_9864_to_fp16)[name = tensor<string, []>("aw_chunk_773_cast")];
            tensor<string, []> var_9867_equation_0 = const()[name = tensor<string, []>("op_9867_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9867_cast = einsum(equation = var_9867_equation_0, values = (var_9797_cast, var_9776_cast))[name = tensor<string, []>("op_9867_cast")];
            tensor<fp16, []> var_9868_to_fp16 = const()[name = tensor<string, []>("op_9868_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_775_cast = mul(x = var_9867_cast, y = var_9868_to_fp16)[name = tensor<string, []>("aw_chunk_775_cast")];
            tensor<string, []> var_9871_equation_0 = const()[name = tensor<string, []>("op_9871_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9871_cast = einsum(equation = var_9871_equation_0, values = (var_9801_cast, var_9777_cast))[name = tensor<string, []>("op_9871_cast")];
            tensor<fp16, []> var_9872_to_fp16 = const()[name = tensor<string, []>("op_9872_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_777_cast = mul(x = var_9871_cast, y = var_9872_to_fp16)[name = tensor<string, []>("aw_chunk_777_cast")];
            tensor<string, []> var_9875_equation_0 = const()[name = tensor<string, []>("op_9875_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9875_cast = einsum(equation = var_9875_equation_0, values = (var_9801_cast, var_9778_cast))[name = tensor<string, []>("op_9875_cast")];
            tensor<fp16, []> var_9876_to_fp16 = const()[name = tensor<string, []>("op_9876_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_779_cast = mul(x = var_9875_cast, y = var_9876_to_fp16)[name = tensor<string, []>("aw_chunk_779_cast")];
            tensor<string, []> var_9879_equation_0 = const()[name = tensor<string, []>("op_9879_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9879_cast = einsum(equation = var_9879_equation_0, values = (var_9805_cast, var_9779_cast))[name = tensor<string, []>("op_9879_cast")];
            tensor<fp16, []> var_9880_to_fp16 = const()[name = tensor<string, []>("op_9880_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_781_cast = mul(x = var_9879_cast, y = var_9880_to_fp16)[name = tensor<string, []>("aw_chunk_781_cast")];
            tensor<string, []> var_9883_equation_0 = const()[name = tensor<string, []>("op_9883_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9883_cast = einsum(equation = var_9883_equation_0, values = (var_9805_cast, var_9780_cast))[name = tensor<string, []>("op_9883_cast")];
            tensor<fp16, []> var_9884_to_fp16 = const()[name = tensor<string, []>("op_9884_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_783_cast = mul(x = var_9883_cast, y = var_9884_to_fp16)[name = tensor<string, []>("aw_chunk_783_cast")];
            tensor<string, []> var_9887_equation_0 = const()[name = tensor<string, []>("op_9887_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9887_cast = einsum(equation = var_9887_equation_0, values = (var_9809_cast, var_9781_cast))[name = tensor<string, []>("op_9887_cast")];
            tensor<fp16, []> var_9888_to_fp16 = const()[name = tensor<string, []>("op_9888_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_785_cast = mul(x = var_9887_cast, y = var_9888_to_fp16)[name = tensor<string, []>("aw_chunk_785_cast")];
            tensor<string, []> var_9891_equation_0 = const()[name = tensor<string, []>("op_9891_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9891_cast = einsum(equation = var_9891_equation_0, values = (var_9809_cast, var_9782_cast))[name = tensor<string, []>("op_9891_cast")];
            tensor<fp16, []> var_9892_to_fp16 = const()[name = tensor<string, []>("op_9892_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_787_cast = mul(x = var_9891_cast, y = var_9892_to_fp16)[name = tensor<string, []>("aw_chunk_787_cast")];
            tensor<string, []> var_9895_equation_0 = const()[name = tensor<string, []>("op_9895_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9895_cast = einsum(equation = var_9895_equation_0, values = (var_9813_cast, var_9783_cast))[name = tensor<string, []>("op_9895_cast")];
            tensor<fp16, []> var_9896_to_fp16 = const()[name = tensor<string, []>("op_9896_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_789_cast = mul(x = var_9895_cast, y = var_9896_to_fp16)[name = tensor<string, []>("aw_chunk_789_cast")];
            tensor<string, []> var_9899_equation_0 = const()[name = tensor<string, []>("op_9899_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9899_cast = einsum(equation = var_9899_equation_0, values = (var_9813_cast, var_9784_cast))[name = tensor<string, []>("op_9899_cast")];
            tensor<fp16, []> var_9900_to_fp16 = const()[name = tensor<string, []>("op_9900_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_791_cast = mul(x = var_9899_cast, y = var_9900_to_fp16)[name = tensor<string, []>("aw_chunk_791_cast")];
            tensor<string, []> var_9903_equation_0 = const()[name = tensor<string, []>("op_9903_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9903_cast = einsum(equation = var_9903_equation_0, values = (var_9817_cast, var_9785_cast))[name = tensor<string, []>("op_9903_cast")];
            tensor<fp16, []> var_9904_to_fp16 = const()[name = tensor<string, []>("op_9904_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_793_cast = mul(x = var_9903_cast, y = var_9904_to_fp16)[name = tensor<string, []>("aw_chunk_793_cast")];
            tensor<string, []> var_9907_equation_0 = const()[name = tensor<string, []>("op_9907_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9907_cast = einsum(equation = var_9907_equation_0, values = (var_9817_cast, var_9786_cast))[name = tensor<string, []>("op_9907_cast")];
            tensor<fp16, []> var_9908_to_fp16 = const()[name = tensor<string, []>("op_9908_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_795_cast = mul(x = var_9907_cast, y = var_9908_to_fp16)[name = tensor<string, []>("aw_chunk_795_cast")];
            tensor<string, []> var_9911_equation_0 = const()[name = tensor<string, []>("op_9911_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9911_cast = einsum(equation = var_9911_equation_0, values = (var_9821_cast, var_9787_cast))[name = tensor<string, []>("op_9911_cast")];
            tensor<fp16, []> var_9912_to_fp16 = const()[name = tensor<string, []>("op_9912_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_797_cast = mul(x = var_9911_cast, y = var_9912_to_fp16)[name = tensor<string, []>("aw_chunk_797_cast")];
            tensor<string, []> var_9915_equation_0 = const()[name = tensor<string, []>("op_9915_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_9915_cast = einsum(equation = var_9915_equation_0, values = (var_9821_cast, var_9788_cast))[name = tensor<string, []>("op_9915_cast")];
            tensor<fp16, []> var_9916_to_fp16 = const()[name = tensor<string, []>("op_9916_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_799_cast = mul(x = var_9915_cast, y = var_9916_to_fp16)[name = tensor<string, []>("aw_chunk_799_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9918_cast = softmax(axis = var_8182, x = aw_chunk_769_cast)[name = tensor<string, []>("op_9918_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9919_cast = softmax(axis = var_8182, x = aw_chunk_771_cast)[name = tensor<string, []>("op_9919_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9920_cast = softmax(axis = var_8182, x = aw_chunk_773_cast)[name = tensor<string, []>("op_9920_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9921_cast = softmax(axis = var_8182, x = aw_chunk_775_cast)[name = tensor<string, []>("op_9921_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9922_cast = softmax(axis = var_8182, x = aw_chunk_777_cast)[name = tensor<string, []>("op_9922_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9923_cast = softmax(axis = var_8182, x = aw_chunk_779_cast)[name = tensor<string, []>("op_9923_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9924_cast = softmax(axis = var_8182, x = aw_chunk_781_cast)[name = tensor<string, []>("op_9924_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9925_cast = softmax(axis = var_8182, x = aw_chunk_783_cast)[name = tensor<string, []>("op_9925_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9926_cast = softmax(axis = var_8182, x = aw_chunk_785_cast)[name = tensor<string, []>("op_9926_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9927_cast = softmax(axis = var_8182, x = aw_chunk_787_cast)[name = tensor<string, []>("op_9927_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9928_cast = softmax(axis = var_8182, x = aw_chunk_789_cast)[name = tensor<string, []>("op_9928_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9929_cast = softmax(axis = var_8182, x = aw_chunk_791_cast)[name = tensor<string, []>("op_9929_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9930_cast = softmax(axis = var_8182, x = aw_chunk_793_cast)[name = tensor<string, []>("op_9930_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9931_cast = softmax(axis = var_8182, x = aw_chunk_795_cast)[name = tensor<string, []>("op_9931_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9932_cast = softmax(axis = var_8182, x = aw_chunk_797_cast)[name = tensor<string, []>("op_9932_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_9933_cast = softmax(axis = var_8182, x = aw_chunk_799_cast)[name = tensor<string, []>("op_9933_cast")];
            tensor<string, []> var_9935_equation_0 = const()[name = tensor<string, []>("op_9935_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9935_cast = einsum(equation = var_9935_equation_0, values = (var_9823_cast, var_9918_cast))[name = tensor<string, []>("op_9935_cast")];
            tensor<string, []> var_9937_equation_0 = const()[name = tensor<string, []>("op_9937_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9937_cast = einsum(equation = var_9937_equation_0, values = (var_9823_cast, var_9919_cast))[name = tensor<string, []>("op_9937_cast")];
            tensor<string, []> var_9939_equation_0 = const()[name = tensor<string, []>("op_9939_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9939_cast = einsum(equation = var_9939_equation_0, values = (var_9827_cast, var_9920_cast))[name = tensor<string, []>("op_9939_cast")];
            tensor<string, []> var_9941_equation_0 = const()[name = tensor<string, []>("op_9941_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9941_cast = einsum(equation = var_9941_equation_0, values = (var_9827_cast, var_9921_cast))[name = tensor<string, []>("op_9941_cast")];
            tensor<string, []> var_9943_equation_0 = const()[name = tensor<string, []>("op_9943_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9943_cast = einsum(equation = var_9943_equation_0, values = (var_9831_cast, var_9922_cast))[name = tensor<string, []>("op_9943_cast")];
            tensor<string, []> var_9945_equation_0 = const()[name = tensor<string, []>("op_9945_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9945_cast = einsum(equation = var_9945_equation_0, values = (var_9831_cast, var_9923_cast))[name = tensor<string, []>("op_9945_cast")];
            tensor<string, []> var_9947_equation_0 = const()[name = tensor<string, []>("op_9947_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9947_cast = einsum(equation = var_9947_equation_0, values = (var_9835_cast, var_9924_cast))[name = tensor<string, []>("op_9947_cast")];
            tensor<string, []> var_9949_equation_0 = const()[name = tensor<string, []>("op_9949_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9949_cast = einsum(equation = var_9949_equation_0, values = (var_9835_cast, var_9925_cast))[name = tensor<string, []>("op_9949_cast")];
            tensor<string, []> var_9951_equation_0 = const()[name = tensor<string, []>("op_9951_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9951_cast = einsum(equation = var_9951_equation_0, values = (var_9839_cast, var_9926_cast))[name = tensor<string, []>("op_9951_cast")];
            tensor<string, []> var_9953_equation_0 = const()[name = tensor<string, []>("op_9953_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9953_cast = einsum(equation = var_9953_equation_0, values = (var_9839_cast, var_9927_cast))[name = tensor<string, []>("op_9953_cast")];
            tensor<string, []> var_9955_equation_0 = const()[name = tensor<string, []>("op_9955_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9955_cast = einsum(equation = var_9955_equation_0, values = (var_9843_cast, var_9928_cast))[name = tensor<string, []>("op_9955_cast")];
            tensor<string, []> var_9957_equation_0 = const()[name = tensor<string, []>("op_9957_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9957_cast = einsum(equation = var_9957_equation_0, values = (var_9843_cast, var_9929_cast))[name = tensor<string, []>("op_9957_cast")];
            tensor<string, []> var_9959_equation_0 = const()[name = tensor<string, []>("op_9959_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9959_cast = einsum(equation = var_9959_equation_0, values = (var_9847_cast, var_9930_cast))[name = tensor<string, []>("op_9959_cast")];
            tensor<string, []> var_9961_equation_0 = const()[name = tensor<string, []>("op_9961_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9961_cast = einsum(equation = var_9961_equation_0, values = (var_9847_cast, var_9931_cast))[name = tensor<string, []>("op_9961_cast")];
            tensor<string, []> var_9963_equation_0 = const()[name = tensor<string, []>("op_9963_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9963_cast = einsum(equation = var_9963_equation_0, values = (var_9851_cast, var_9932_cast))[name = tensor<string, []>("op_9963_cast")];
            tensor<string, []> var_9965_equation_0 = const()[name = tensor<string, []>("op_9965_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_9965_cast = einsum(equation = var_9965_equation_0, values = (var_9851_cast, var_9933_cast))[name = tensor<string, []>("op_9965_cast")];
            tensor<bool, []> var_9967_interleave_0 = const()[name = tensor<string, []>("op_9967_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_9967_cast = concat(axis = var_8160, interleave = var_9967_interleave_0, values = (var_9935_cast, var_9937_cast))[name = tensor<string, []>("op_9967_cast")];
            tensor<bool, []> var_9969_interleave_0 = const()[name = tensor<string, []>("op_9969_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_9969_cast = concat(axis = var_8160, interleave = var_9969_interleave_0, values = (var_9939_cast, var_9941_cast))[name = tensor<string, []>("op_9969_cast")];
            tensor<bool, []> var_9971_interleave_0 = const()[name = tensor<string, []>("op_9971_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_9971_cast = concat(axis = var_8160, interleave = var_9971_interleave_0, values = (var_9943_cast, var_9945_cast))[name = tensor<string, []>("op_9971_cast")];
            tensor<bool, []> var_9973_interleave_0 = const()[name = tensor<string, []>("op_9973_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_9973_cast = concat(axis = var_8160, interleave = var_9973_interleave_0, values = (var_9947_cast, var_9949_cast))[name = tensor<string, []>("op_9973_cast")];
            tensor<bool, []> var_9975_interleave_0 = const()[name = tensor<string, []>("op_9975_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_9975_cast = concat(axis = var_8160, interleave = var_9975_interleave_0, values = (var_9951_cast, var_9953_cast))[name = tensor<string, []>("op_9975_cast")];
            tensor<bool, []> var_9977_interleave_0 = const()[name = tensor<string, []>("op_9977_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_9977_cast = concat(axis = var_8160, interleave = var_9977_interleave_0, values = (var_9955_cast, var_9957_cast))[name = tensor<string, []>("op_9977_cast")];
            tensor<bool, []> var_9979_interleave_0 = const()[name = tensor<string, []>("op_9979_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_9979_cast = concat(axis = var_8160, interleave = var_9979_interleave_0, values = (var_9959_cast, var_9961_cast))[name = tensor<string, []>("op_9979_cast")];
            tensor<bool, []> var_9981_interleave_0 = const()[name = tensor<string, []>("op_9981_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_9981_cast = concat(axis = var_8160, interleave = var_9981_interleave_0, values = (var_9963_cast, var_9965_cast))[name = tensor<string, []>("op_9981_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_8182, interleave = input_435_interleave_0, values = (var_9967_cast, var_9969_cast, var_9971_cast, var_9973_cast, var_9975_cast, var_9977_cast, var_9979_cast, var_9981_cast))[name = tensor<string, []>("input_435_cast")];
            tensor<int32, [2]> var_9987 = const()[name = tensor<string, []>("op_9987"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9989 = const()[name = tensor<string, []>("op_9989"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_9991_pad_type_0 = const()[name = tensor<string, []>("op_9991_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_9991_pad_0 = const()[name = tensor<string, []>("op_9991_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(622568128))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(622875392))), name = tensor<string, []>("up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            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, []>(622875584)))];
            tensor<fp16, [2, 640, 1, 1024]> var_9991_cast = conv(bias = up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_9989, groups = var_8182, pad = var_9991_pad_0, pad_type = var_9991_pad_type_0, strides = var_9987, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_435_cast)[name = tensor<string, []>("op_9991_cast")];
            tensor<fp16, [2, 640, 1, 1024]> inputs_75_cast = add(x = var_9991_cast, y = inputs_73_cast)[name = tensor<string, []>("inputs_75_cast")];
            tensor<int32, [1]> var_9995 = const()[name = tensor<string, []>("op_9995"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_75_cast = reduce_mean(axes = var_9995, keep_dims = var_8177, x = inputs_75_cast)[name = tensor<string, []>("channels_mean_75_cast")];
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_75_cast = sub(x = inputs_75_cast, y = channels_mean_75_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_75_cast")];
            tensor<int32, [1]> var_9999 = const()[name = tensor<string, []>("op_9999"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_10000_cast = reduce_mean(axes = var_9999, keep_dims = var_8177, x = zero_mean_sq_75_cast)[name = tensor<string, []>("op_10000_cast")];
            tensor<fp16, []> var_10001_to_fp16 = const()[name = tensor<string, []>("op_10001_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_10002_cast = add(x = var_10000_cast, y = var_10001_to_fp16)[name = tensor<string, []>("op_10002_cast")];
            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_10002_cast)[name = tensor<string, []>("denom_75_cast")];
            tensor<fp16, [2, 640, 1, 1024]> out_75_cast = mul(x = zero_mean_75_cast, y = denom_75_cast)[name = tensor<string, []>("out_75_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_10006_to_fp16 = const()[name = tensor<string, []>("op_10006_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(622876928)))];
            tensor<fp16, [2, 640, 1, 1024]> var_10007_cast = add(x = out_75_cast, y = var_10006_to_fp16)[name = tensor<string, []>("op_10007_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_10009_to_fp16 = const()[name = tensor<string, []>("op_10009_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(622878272)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_269_cast = mul(x = var_10007_cast, y = var_10009_to_fp16)[name = tensor<string, []>("hidden_states_269_cast")];
            tensor<int32, [2]> var_10016 = const()[name = tensor<string, []>("op_10016"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10018 = const()[name = tensor<string, []>("op_10018"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(622879616))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(623186880))), name = tensor<string, []>("up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            tensor<fp16, [2, 640, 1, 1024]> q_51_cast = conv(dilations = var_10018, groups = var_8182, pad = q_51_pad_0, pad_type = q_51_pad_type_0, strides = var_10016, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_269_cast)[name = tensor<string, []>("q_51_cast")];
            tensor<int32, [2]> var_10022 = const()[name = tensor<string, []>("op_10022"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10024 = const()[name = tensor<string, []>("op_10024"), 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, 768, 1, 1]> up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [368640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(623187072))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(623555776))), name = tensor<string, []>("up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 768, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> k_101_cast = conv(dilations = var_10024, groups = var_8182, pad = k_101_pad_0, pad_type = k_101_pad_type_0, strides = var_10022, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_101_cast")];
            tensor<int32, [2]> var_10028 = const()[name = tensor<string, []>("op_10028"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10030 = const()[name = tensor<string, []>("op_10030"), 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, 768, 1, 1]> up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [368640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(623555968))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(623924672))), name = tensor<string, []>("up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 768, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> v_51_cast = conv(dilations = var_10030, groups = var_8182, pad = v_51_pad_0, pad_type = v_51_pad_type_0, strides = var_10028, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_51_cast")];
            tensor<int32, [4]> var_10034_begin_0 = const()[name = tensor<string, []>("op_10034_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10034_end_0 = const()[name = tensor<string, []>("op_10034_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_10034_end_mask_0 = const()[name = tensor<string, []>("op_10034_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_10034_cast = slice_by_index(begin = var_10034_begin_0, end = var_10034_end_0, end_mask = var_10034_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_10034_cast")];
            tensor<int32, [4]> var_10038_begin_0 = const()[name = tensor<string, []>("op_10038_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_10038_end_0 = const()[name = tensor<string, []>("op_10038_end_0"), val = tensor<int32, [4]>([2, 160, 1, 1024])];
            tensor<bool, [4]> var_10038_end_mask_0 = const()[name = tensor<string, []>("op_10038_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_10038_cast = slice_by_index(begin = var_10038_begin_0, end = var_10038_end_0, end_mask = var_10038_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_10038_cast")];
            tensor<int32, [4]> var_10042_begin_0 = const()[name = tensor<string, []>("op_10042_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_10042_end_0 = const()[name = tensor<string, []>("op_10042_end_0"), val = tensor<int32, [4]>([2, 240, 1, 1024])];
            tensor<bool, [4]> var_10042_end_mask_0 = const()[name = tensor<string, []>("op_10042_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_10042_cast = slice_by_index(begin = var_10042_begin_0, end = var_10042_end_0, end_mask = var_10042_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_10042_cast")];
            tensor<int32, [4]> var_10046_begin_0 = const()[name = tensor<string, []>("op_10046_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_10046_end_0 = const()[name = tensor<string, []>("op_10046_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_10046_end_mask_0 = const()[name = tensor<string, []>("op_10046_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_10046_cast = slice_by_index(begin = var_10046_begin_0, end = var_10046_end_0, end_mask = var_10046_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_10046_cast")];
            tensor<int32, [4]> var_10050_begin_0 = const()[name = tensor<string, []>("op_10050_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_10050_end_0 = const()[name = tensor<string, []>("op_10050_end_0"), val = tensor<int32, [4]>([2, 400, 1, 1024])];
            tensor<bool, [4]> var_10050_end_mask_0 = const()[name = tensor<string, []>("op_10050_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_10050_cast = slice_by_index(begin = var_10050_begin_0, end = var_10050_end_0, end_mask = var_10050_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_10050_cast")];
            tensor<int32, [4]> var_10054_begin_0 = const()[name = tensor<string, []>("op_10054_begin_0"), val = tensor<int32, [4]>([0, 400, 0, 0])];
            tensor<int32, [4]> var_10054_end_0 = const()[name = tensor<string, []>("op_10054_end_0"), val = tensor<int32, [4]>([2, 480, 1, 1024])];
            tensor<bool, [4]> var_10054_end_mask_0 = const()[name = tensor<string, []>("op_10054_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_10054_cast = slice_by_index(begin = var_10054_begin_0, end = var_10054_end_0, end_mask = var_10054_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_10054_cast")];
            tensor<int32, [4]> var_10058_begin_0 = const()[name = tensor<string, []>("op_10058_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_10058_end_0 = const()[name = tensor<string, []>("op_10058_end_0"), val = tensor<int32, [4]>([2, 560, 1, 1024])];
            tensor<bool, [4]> var_10058_end_mask_0 = const()[name = tensor<string, []>("op_10058_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_10058_cast = slice_by_index(begin = var_10058_begin_0, end = var_10058_end_0, end_mask = var_10058_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_10058_cast")];
            tensor<int32, [4]> var_10062_begin_0 = const()[name = tensor<string, []>("op_10062_begin_0"), val = tensor<int32, [4]>([0, 560, 0, 0])];
            tensor<int32, [4]> var_10062_end_0 = const()[name = tensor<string, []>("op_10062_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_10062_end_mask_0 = const()[name = tensor<string, []>("op_10062_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 1024]> var_10062_cast = slice_by_index(begin = var_10062_begin_0, end = var_10062_end_0, end_mask = var_10062_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_10062_cast")];
            tensor<int32, [4]> var_10065_begin_0 = const()[name = tensor<string, []>("op_10065_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10065_end_0 = const()[name = tensor<string, []>("op_10065_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_10065_end_mask_0 = const()[name = tensor<string, []>("op_10065_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_10065_cast = slice_by_index(begin = var_10065_begin_0, end = var_10065_end_0, end_mask = var_10065_end_mask_0, x = var_10034_cast)[name = tensor<string, []>("op_10065_cast")];
            tensor<int32, [4]> var_10066_begin_0 = const()[name = tensor<string, []>("op_10066_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10066_end_0 = const()[name = tensor<string, []>("op_10066_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            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, 80, 1, 512]> var_10066_cast = slice_by_index(begin = var_10066_begin_0, end = var_10066_end_0, end_mask = var_10066_end_mask_0, x = var_10034_cast)[name = tensor<string, []>("op_10066_cast")];
            tensor<int32, [4]> var_10067_begin_0 = const()[name = tensor<string, []>("op_10067_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10067_end_0 = const()[name = tensor<string, []>("op_10067_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_10067_end_mask_0 = const()[name = tensor<string, []>("op_10067_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_10067_cast = slice_by_index(begin = var_10067_begin_0, end = var_10067_end_0, end_mask = var_10067_end_mask_0, x = var_10038_cast)[name = tensor<string, []>("op_10067_cast")];
            tensor<int32, [4]> var_10068_begin_0 = const()[name = tensor<string, []>("op_10068_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10068_end_0 = const()[name = tensor<string, []>("op_10068_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_10068_end_mask_0 = const()[name = tensor<string, []>("op_10068_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_10068_cast = slice_by_index(begin = var_10068_begin_0, end = var_10068_end_0, end_mask = var_10068_end_mask_0, x = var_10038_cast)[name = tensor<string, []>("op_10068_cast")];
            tensor<int32, [4]> var_10069_begin_0 = const()[name = tensor<string, []>("op_10069_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10069_end_0 = const()[name = tensor<string, []>("op_10069_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_10069_end_mask_0 = const()[name = tensor<string, []>("op_10069_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_10069_cast = slice_by_index(begin = var_10069_begin_0, end = var_10069_end_0, end_mask = var_10069_end_mask_0, x = var_10042_cast)[name = tensor<string, []>("op_10069_cast")];
            tensor<int32, [4]> var_10070_begin_0 = const()[name = tensor<string, []>("op_10070_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10070_end_0 = const()[name = tensor<string, []>("op_10070_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_10070_end_mask_0 = const()[name = tensor<string, []>("op_10070_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_10070_cast = slice_by_index(begin = var_10070_begin_0, end = var_10070_end_0, end_mask = var_10070_end_mask_0, x = var_10042_cast)[name = tensor<string, []>("op_10070_cast")];
            tensor<int32, [4]> var_10071_begin_0 = const()[name = tensor<string, []>("op_10071_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10071_end_0 = const()[name = tensor<string, []>("op_10071_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_10071_end_mask_0 = const()[name = tensor<string, []>("op_10071_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_10071_cast = slice_by_index(begin = var_10071_begin_0, end = var_10071_end_0, end_mask = var_10071_end_mask_0, x = var_10046_cast)[name = tensor<string, []>("op_10071_cast")];
            tensor<int32, [4]> var_10072_begin_0 = const()[name = tensor<string, []>("op_10072_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10072_end_0 = const()[name = tensor<string, []>("op_10072_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_10072_end_mask_0 = const()[name = tensor<string, []>("op_10072_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_10072_cast = slice_by_index(begin = var_10072_begin_0, end = var_10072_end_0, end_mask = var_10072_end_mask_0, x = var_10046_cast)[name = tensor<string, []>("op_10072_cast")];
            tensor<int32, [4]> var_10073_begin_0 = const()[name = tensor<string, []>("op_10073_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10073_end_0 = const()[name = tensor<string, []>("op_10073_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_10073_end_mask_0 = const()[name = tensor<string, []>("op_10073_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_10073_cast = slice_by_index(begin = var_10073_begin_0, end = var_10073_end_0, end_mask = var_10073_end_mask_0, x = var_10050_cast)[name = tensor<string, []>("op_10073_cast")];
            tensor<int32, [4]> var_10074_begin_0 = const()[name = tensor<string, []>("op_10074_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10074_end_0 = const()[name = tensor<string, []>("op_10074_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_10074_end_mask_0 = const()[name = tensor<string, []>("op_10074_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_10074_cast = slice_by_index(begin = var_10074_begin_0, end = var_10074_end_0, end_mask = var_10074_end_mask_0, x = var_10050_cast)[name = tensor<string, []>("op_10074_cast")];
            tensor<int32, [4]> var_10075_begin_0 = const()[name = tensor<string, []>("op_10075_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10075_end_0 = const()[name = tensor<string, []>("op_10075_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_10075_end_mask_0 = const()[name = tensor<string, []>("op_10075_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_10075_cast = slice_by_index(begin = var_10075_begin_0, end = var_10075_end_0, end_mask = var_10075_end_mask_0, x = var_10054_cast)[name = tensor<string, []>("op_10075_cast")];
            tensor<int32, [4]> var_10076_begin_0 = const()[name = tensor<string, []>("op_10076_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10076_end_0 = const()[name = tensor<string, []>("op_10076_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_10076_end_mask_0 = const()[name = tensor<string, []>("op_10076_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_10076_cast = slice_by_index(begin = var_10076_begin_0, end = var_10076_end_0, end_mask = var_10076_end_mask_0, x = var_10054_cast)[name = tensor<string, []>("op_10076_cast")];
            tensor<int32, [4]> var_10077_begin_0 = const()[name = tensor<string, []>("op_10077_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10077_end_0 = const()[name = tensor<string, []>("op_10077_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_10077_end_mask_0 = const()[name = tensor<string, []>("op_10077_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_10077_cast = slice_by_index(begin = var_10077_begin_0, end = var_10077_end_0, end_mask = var_10077_end_mask_0, x = var_10058_cast)[name = tensor<string, []>("op_10077_cast")];
            tensor<int32, [4]> var_10078_begin_0 = const()[name = tensor<string, []>("op_10078_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10078_end_0 = const()[name = tensor<string, []>("op_10078_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_10078_end_mask_0 = const()[name = tensor<string, []>("op_10078_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_10078_cast = slice_by_index(begin = var_10078_begin_0, end = var_10078_end_0, end_mask = var_10078_end_mask_0, x = var_10058_cast)[name = tensor<string, []>("op_10078_cast")];
            tensor<int32, [4]> var_10079_begin_0 = const()[name = tensor<string, []>("op_10079_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10079_end_0 = const()[name = tensor<string, []>("op_10079_end_0"), val = tensor<int32, [4]>([2, 80, 1, 512])];
            tensor<bool, [4]> var_10079_end_mask_0 = const()[name = tensor<string, []>("op_10079_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_10079_cast = slice_by_index(begin = var_10079_begin_0, end = var_10079_end_0, end_mask = var_10079_end_mask_0, x = var_10062_cast)[name = tensor<string, []>("op_10079_cast")];
            tensor<int32, [4]> var_10080_begin_0 = const()[name = tensor<string, []>("op_10080_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10080_end_0 = const()[name = tensor<string, []>("op_10080_end_0"), val = tensor<int32, [4]>([2, 80, 1, 1024])];
            tensor<bool, [4]> var_10080_end_mask_0 = const()[name = tensor<string, []>("op_10080_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 80, 1, 512]> var_10080_cast = slice_by_index(begin = var_10080_begin_0, end = var_10080_end_0, end_mask = var_10080_end_mask_0, x = var_10062_cast)[name = tensor<string, []>("op_10080_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_10085_begin_0 = const()[name = tensor<string, []>("op_10085_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10085_end_0 = const()[name = tensor<string, []>("op_10085_end_0"), val = tensor<int32, [4]>([2, 77, 1, 80])];
            tensor<bool, [4]> var_10085_end_mask_0 = const()[name = tensor<string, []>("op_10085_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)[name = tensor<string, []>("transpose_6")];
            tensor<fp16, [2, 77, 1, 80]> var_10085_cast = slice_by_index(begin = var_10085_begin_0, end = var_10085_end_0, end_mask = var_10085_end_mask_0, x = transpose_6)[name = tensor<string, []>("op_10085_cast")];
            tensor<int32, [4]> var_10089_begin_0 = const()[name = tensor<string, []>("op_10089_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 80])];
            tensor<int32, [4]> var_10089_end_0 = const()[name = tensor<string, []>("op_10089_end_0"), val = tensor<int32, [4]>([2, 77, 1, 160])];
            tensor<bool, [4]> var_10089_end_mask_0 = const()[name = tensor<string, []>("op_10089_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_10089_cast = slice_by_index(begin = var_10089_begin_0, end = var_10089_end_0, end_mask = var_10089_end_mask_0, x = transpose_6)[name = tensor<string, []>("op_10089_cast")];
            tensor<int32, [4]> var_10093_begin_0 = const()[name = tensor<string, []>("op_10093_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_10093_end_0 = const()[name = tensor<string, []>("op_10093_end_0"), val = tensor<int32, [4]>([2, 77, 1, 240])];
            tensor<bool, [4]> var_10093_end_mask_0 = const()[name = tensor<string, []>("op_10093_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_10093_cast = slice_by_index(begin = var_10093_begin_0, end = var_10093_end_0, end_mask = var_10093_end_mask_0, x = transpose_6)[name = tensor<string, []>("op_10093_cast")];
            tensor<int32, [4]> var_10097_begin_0 = const()[name = tensor<string, []>("op_10097_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 240])];
            tensor<int32, [4]> var_10097_end_0 = const()[name = tensor<string, []>("op_10097_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_10097_end_mask_0 = const()[name = tensor<string, []>("op_10097_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_10097_cast = slice_by_index(begin = var_10097_begin_0, end = var_10097_end_0, end_mask = var_10097_end_mask_0, x = transpose_6)[name = tensor<string, []>("op_10097_cast")];
            tensor<int32, [4]> var_10101_begin_0 = const()[name = tensor<string, []>("op_10101_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_10101_end_0 = const()[name = tensor<string, []>("op_10101_end_0"), val = tensor<int32, [4]>([2, 77, 1, 400])];
            tensor<bool, [4]> var_10101_end_mask_0 = const()[name = tensor<string, []>("op_10101_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_10101_cast = slice_by_index(begin = var_10101_begin_0, end = var_10101_end_0, end_mask = var_10101_end_mask_0, x = transpose_6)[name = tensor<string, []>("op_10101_cast")];
            tensor<int32, [4]> var_10105_begin_0 = const()[name = tensor<string, []>("op_10105_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 400])];
            tensor<int32, [4]> var_10105_end_0 = const()[name = tensor<string, []>("op_10105_end_0"), val = tensor<int32, [4]>([2, 77, 1, 480])];
            tensor<bool, [4]> var_10105_end_mask_0 = const()[name = tensor<string, []>("op_10105_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_10105_cast = slice_by_index(begin = var_10105_begin_0, end = var_10105_end_0, end_mask = var_10105_end_mask_0, x = transpose_6)[name = tensor<string, []>("op_10105_cast")];
            tensor<int32, [4]> var_10109_begin_0 = const()[name = tensor<string, []>("op_10109_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 480])];
            tensor<int32, [4]> var_10109_end_0 = const()[name = tensor<string, []>("op_10109_end_0"), val = tensor<int32, [4]>([2, 77, 1, 560])];
            tensor<bool, [4]> var_10109_end_mask_0 = const()[name = tensor<string, []>("op_10109_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_10109_cast = slice_by_index(begin = var_10109_begin_0, end = var_10109_end_0, end_mask = var_10109_end_mask_0, x = transpose_6)[name = tensor<string, []>("op_10109_cast")];
            tensor<int32, [4]> var_10113_begin_0 = const()[name = tensor<string, []>("op_10113_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 560])];
            tensor<int32, [4]> var_10113_end_0 = const()[name = tensor<string, []>("op_10113_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_10113_end_mask_0 = const()[name = tensor<string, []>("op_10113_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 80]> var_10113_cast = slice_by_index(begin = var_10113_begin_0, end = var_10113_end_0, end_mask = var_10113_end_mask_0, x = transpose_6)[name = tensor<string, []>("op_10113_cast")];
            tensor<int32, [4]> var_10115_begin_0 = const()[name = tensor<string, []>("op_10115_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10115_end_0 = const()[name = tensor<string, []>("op_10115_end_0"), val = tensor<int32, [4]>([2, 80, 1, 77])];
            tensor<bool, [4]> var_10115_end_mask_0 = const()[name = tensor<string, []>("op_10115_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_10115_cast = slice_by_index(begin = var_10115_begin_0, end = var_10115_end_0, end_mask = var_10115_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_10115_cast")];
            tensor<int32, [4]> var_10119_begin_0 = const()[name = tensor<string, []>("op_10119_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_10119_end_0 = const()[name = tensor<string, []>("op_10119_end_0"), val = tensor<int32, [4]>([2, 160, 1, 77])];
            tensor<bool, [4]> var_10119_end_mask_0 = const()[name = tensor<string, []>("op_10119_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_10119_cast = slice_by_index(begin = var_10119_begin_0, end = var_10119_end_0, end_mask = var_10119_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_10119_cast")];
            tensor<int32, [4]> var_10123_begin_0 = const()[name = tensor<string, []>("op_10123_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_10123_end_0 = const()[name = tensor<string, []>("op_10123_end_0"), val = tensor<int32, [4]>([2, 240, 1, 77])];
            tensor<bool, [4]> var_10123_end_mask_0 = const()[name = tensor<string, []>("op_10123_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_10123_cast = slice_by_index(begin = var_10123_begin_0, end = var_10123_end_0, end_mask = var_10123_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_10123_cast")];
            tensor<int32, [4]> var_10127_begin_0 = const()[name = tensor<string, []>("op_10127_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_10127_end_0 = const()[name = tensor<string, []>("op_10127_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_10127_end_mask_0 = const()[name = tensor<string, []>("op_10127_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_10127_cast = slice_by_index(begin = var_10127_begin_0, end = var_10127_end_0, end_mask = var_10127_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_10127_cast")];
            tensor<int32, [4]> var_10131_begin_0 = const()[name = tensor<string, []>("op_10131_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_10131_end_0 = const()[name = tensor<string, []>("op_10131_end_0"), val = tensor<int32, [4]>([2, 400, 1, 77])];
            tensor<bool, [4]> var_10131_end_mask_0 = const()[name = tensor<string, []>("op_10131_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_10131_cast = slice_by_index(begin = var_10131_begin_0, end = var_10131_end_0, end_mask = var_10131_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_10131_cast")];
            tensor<int32, [4]> var_10135_begin_0 = const()[name = tensor<string, []>("op_10135_begin_0"), val = tensor<int32, [4]>([0, 400, 0, 0])];
            tensor<int32, [4]> var_10135_end_0 = const()[name = tensor<string, []>("op_10135_end_0"), val = tensor<int32, [4]>([2, 480, 1, 77])];
            tensor<bool, [4]> var_10135_end_mask_0 = const()[name = tensor<string, []>("op_10135_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_10135_cast = slice_by_index(begin = var_10135_begin_0, end = var_10135_end_0, end_mask = var_10135_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_10135_cast")];
            tensor<int32, [4]> var_10139_begin_0 = const()[name = tensor<string, []>("op_10139_begin_0"), val = tensor<int32, [4]>([0, 480, 0, 0])];
            tensor<int32, [4]> var_10139_end_0 = const()[name = tensor<string, []>("op_10139_end_0"), val = tensor<int32, [4]>([2, 560, 1, 77])];
            tensor<bool, [4]> var_10139_end_mask_0 = const()[name = tensor<string, []>("op_10139_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_10139_cast = slice_by_index(begin = var_10139_begin_0, end = var_10139_end_0, end_mask = var_10139_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_10139_cast")];
            tensor<int32, [4]> var_10143_begin_0 = const()[name = tensor<string, []>("op_10143_begin_0"), val = tensor<int32, [4]>([0, 560, 0, 0])];
            tensor<int32, [4]> var_10143_end_0 = const()[name = tensor<string, []>("op_10143_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_10143_end_mask_0 = const()[name = tensor<string, []>("op_10143_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 80, 1, 77]> var_10143_cast = slice_by_index(begin = var_10143_begin_0, end = var_10143_end_0, end_mask = var_10143_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_10143_cast")];
            tensor<string, []> var_10147_equation_0 = const()[name = tensor<string, []>("op_10147_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10147_cast = einsum(equation = var_10147_equation_0, values = (var_10085_cast, var_10065_cast))[name = tensor<string, []>("op_10147_cast")];
            tensor<fp16, []> var_10148_to_fp16 = const()[name = tensor<string, []>("op_10148_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_801_cast = mul(x = var_10147_cast, y = var_10148_to_fp16)[name = tensor<string, []>("aw_chunk_801_cast")];
            tensor<string, []> var_10151_equation_0 = const()[name = tensor<string, []>("op_10151_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10151_cast = einsum(equation = var_10151_equation_0, values = (var_10085_cast, var_10066_cast))[name = tensor<string, []>("op_10151_cast")];
            tensor<fp16, []> var_10152_to_fp16 = const()[name = tensor<string, []>("op_10152_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_803_cast = mul(x = var_10151_cast, y = var_10152_to_fp16)[name = tensor<string, []>("aw_chunk_803_cast")];
            tensor<string, []> var_10155_equation_0 = const()[name = tensor<string, []>("op_10155_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10155_cast = einsum(equation = var_10155_equation_0, values = (var_10089_cast, var_10067_cast))[name = tensor<string, []>("op_10155_cast")];
            tensor<fp16, []> var_10156_to_fp16 = const()[name = tensor<string, []>("op_10156_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_805_cast = mul(x = var_10155_cast, y = var_10156_to_fp16)[name = tensor<string, []>("aw_chunk_805_cast")];
            tensor<string, []> var_10159_equation_0 = const()[name = tensor<string, []>("op_10159_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10159_cast = einsum(equation = var_10159_equation_0, values = (var_10089_cast, var_10068_cast))[name = tensor<string, []>("op_10159_cast")];
            tensor<fp16, []> var_10160_to_fp16 = const()[name = tensor<string, []>("op_10160_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_807_cast = mul(x = var_10159_cast, y = var_10160_to_fp16)[name = tensor<string, []>("aw_chunk_807_cast")];
            tensor<string, []> var_10163_equation_0 = const()[name = tensor<string, []>("op_10163_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10163_cast = einsum(equation = var_10163_equation_0, values = (var_10093_cast, var_10069_cast))[name = tensor<string, []>("op_10163_cast")];
            tensor<fp16, []> var_10164_to_fp16 = const()[name = tensor<string, []>("op_10164_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_809_cast = mul(x = var_10163_cast, y = var_10164_to_fp16)[name = tensor<string, []>("aw_chunk_809_cast")];
            tensor<string, []> var_10167_equation_0 = const()[name = tensor<string, []>("op_10167_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10167_cast = einsum(equation = var_10167_equation_0, values = (var_10093_cast, var_10070_cast))[name = tensor<string, []>("op_10167_cast")];
            tensor<fp16, []> var_10168_to_fp16 = const()[name = tensor<string, []>("op_10168_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_811_cast = mul(x = var_10167_cast, y = var_10168_to_fp16)[name = tensor<string, []>("aw_chunk_811_cast")];
            tensor<string, []> var_10171_equation_0 = const()[name = tensor<string, []>("op_10171_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10171_cast = einsum(equation = var_10171_equation_0, values = (var_10097_cast, var_10071_cast))[name = tensor<string, []>("op_10171_cast")];
            tensor<fp16, []> var_10172_to_fp16 = const()[name = tensor<string, []>("op_10172_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_813_cast = mul(x = var_10171_cast, y = var_10172_to_fp16)[name = tensor<string, []>("aw_chunk_813_cast")];
            tensor<string, []> var_10175_equation_0 = const()[name = tensor<string, []>("op_10175_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10175_cast = einsum(equation = var_10175_equation_0, values = (var_10097_cast, var_10072_cast))[name = tensor<string, []>("op_10175_cast")];
            tensor<fp16, []> var_10176_to_fp16 = const()[name = tensor<string, []>("op_10176_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_815_cast = mul(x = var_10175_cast, y = var_10176_to_fp16)[name = tensor<string, []>("aw_chunk_815_cast")];
            tensor<string, []> var_10179_equation_0 = const()[name = tensor<string, []>("op_10179_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10179_cast = einsum(equation = var_10179_equation_0, values = (var_10101_cast, var_10073_cast))[name = tensor<string, []>("op_10179_cast")];
            tensor<fp16, []> var_10180_to_fp16 = const()[name = tensor<string, []>("op_10180_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_817_cast = mul(x = var_10179_cast, y = var_10180_to_fp16)[name = tensor<string, []>("aw_chunk_817_cast")];
            tensor<string, []> var_10183_equation_0 = const()[name = tensor<string, []>("op_10183_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10183_cast = einsum(equation = var_10183_equation_0, values = (var_10101_cast, var_10074_cast))[name = tensor<string, []>("op_10183_cast")];
            tensor<fp16, []> var_10184_to_fp16 = const()[name = tensor<string, []>("op_10184_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_819_cast = mul(x = var_10183_cast, y = var_10184_to_fp16)[name = tensor<string, []>("aw_chunk_819_cast")];
            tensor<string, []> var_10187_equation_0 = const()[name = tensor<string, []>("op_10187_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10187_cast = einsum(equation = var_10187_equation_0, values = (var_10105_cast, var_10075_cast))[name = tensor<string, []>("op_10187_cast")];
            tensor<fp16, []> var_10188_to_fp16 = const()[name = tensor<string, []>("op_10188_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_821_cast = mul(x = var_10187_cast, y = var_10188_to_fp16)[name = tensor<string, []>("aw_chunk_821_cast")];
            tensor<string, []> var_10191_equation_0 = const()[name = tensor<string, []>("op_10191_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10191_cast = einsum(equation = var_10191_equation_0, values = (var_10105_cast, var_10076_cast))[name = tensor<string, []>("op_10191_cast")];
            tensor<fp16, []> var_10192_to_fp16 = const()[name = tensor<string, []>("op_10192_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_823_cast = mul(x = var_10191_cast, y = var_10192_to_fp16)[name = tensor<string, []>("aw_chunk_823_cast")];
            tensor<string, []> var_10195_equation_0 = const()[name = tensor<string, []>("op_10195_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10195_cast = einsum(equation = var_10195_equation_0, values = (var_10109_cast, var_10077_cast))[name = tensor<string, []>("op_10195_cast")];
            tensor<fp16, []> var_10196_to_fp16 = const()[name = tensor<string, []>("op_10196_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_825_cast = mul(x = var_10195_cast, y = var_10196_to_fp16)[name = tensor<string, []>("aw_chunk_825_cast")];
            tensor<string, []> var_10199_equation_0 = const()[name = tensor<string, []>("op_10199_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10199_cast = einsum(equation = var_10199_equation_0, values = (var_10109_cast, var_10078_cast))[name = tensor<string, []>("op_10199_cast")];
            tensor<fp16, []> var_10200_to_fp16 = const()[name = tensor<string, []>("op_10200_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_827_cast = mul(x = var_10199_cast, y = var_10200_to_fp16)[name = tensor<string, []>("aw_chunk_827_cast")];
            tensor<string, []> var_10203_equation_0 = const()[name = tensor<string, []>("op_10203_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10203_cast = einsum(equation = var_10203_equation_0, values = (var_10113_cast, var_10079_cast))[name = tensor<string, []>("op_10203_cast")];
            tensor<fp16, []> var_10204_to_fp16 = const()[name = tensor<string, []>("op_10204_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_829_cast = mul(x = var_10203_cast, y = var_10204_to_fp16)[name = tensor<string, []>("aw_chunk_829_cast")];
            tensor<string, []> var_10207_equation_0 = const()[name = tensor<string, []>("op_10207_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10207_cast = einsum(equation = var_10207_equation_0, values = (var_10113_cast, var_10080_cast))[name = tensor<string, []>("op_10207_cast")];
            tensor<fp16, []> var_10208_to_fp16 = const()[name = tensor<string, []>("op_10208_to_fp16"), val = tensor<fp16, []>(0x1.cap-4)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_831_cast = mul(x = var_10207_cast, y = var_10208_to_fp16)[name = tensor<string, []>("aw_chunk_831_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10210_cast = softmax(axis = var_8182, x = aw_chunk_801_cast)[name = tensor<string, []>("op_10210_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10211_cast = softmax(axis = var_8182, x = aw_chunk_803_cast)[name = tensor<string, []>("op_10211_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10212_cast = softmax(axis = var_8182, x = aw_chunk_805_cast)[name = tensor<string, []>("op_10212_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10213_cast = softmax(axis = var_8182, x = aw_chunk_807_cast)[name = tensor<string, []>("op_10213_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10214_cast = softmax(axis = var_8182, x = aw_chunk_809_cast)[name = tensor<string, []>("op_10214_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10215_cast = softmax(axis = var_8182, x = aw_chunk_811_cast)[name = tensor<string, []>("op_10215_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10216_cast = softmax(axis = var_8182, x = aw_chunk_813_cast)[name = tensor<string, []>("op_10216_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10217_cast = softmax(axis = var_8182, x = aw_chunk_815_cast)[name = tensor<string, []>("op_10217_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10218_cast = softmax(axis = var_8182, x = aw_chunk_817_cast)[name = tensor<string, []>("op_10218_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10219_cast = softmax(axis = var_8182, x = aw_chunk_819_cast)[name = tensor<string, []>("op_10219_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10220_cast = softmax(axis = var_8182, x = aw_chunk_821_cast)[name = tensor<string, []>("op_10220_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10221_cast = softmax(axis = var_8182, x = aw_chunk_823_cast)[name = tensor<string, []>("op_10221_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10222_cast = softmax(axis = var_8182, x = aw_chunk_825_cast)[name = tensor<string, []>("op_10222_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10223_cast = softmax(axis = var_8182, x = aw_chunk_827_cast)[name = tensor<string, []>("op_10223_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10224_cast = softmax(axis = var_8182, x = aw_chunk_829_cast)[name = tensor<string, []>("op_10224_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10225_cast = softmax(axis = var_8182, x = aw_chunk_831_cast)[name = tensor<string, []>("op_10225_cast")];
            tensor<string, []> var_10227_equation_0 = const()[name = tensor<string, []>("op_10227_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_10227_cast = einsum(equation = var_10227_equation_0, values = (var_10115_cast, var_10210_cast))[name = tensor<string, []>("op_10227_cast")];
            tensor<string, []> var_10229_equation_0 = const()[name = tensor<string, []>("op_10229_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_10229_cast = einsum(equation = var_10229_equation_0, values = (var_10115_cast, var_10211_cast))[name = tensor<string, []>("op_10229_cast")];
            tensor<string, []> var_10231_equation_0 = const()[name = tensor<string, []>("op_10231_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_10231_cast = einsum(equation = var_10231_equation_0, values = (var_10119_cast, var_10212_cast))[name = tensor<string, []>("op_10231_cast")];
            tensor<string, []> var_10233_equation_0 = const()[name = tensor<string, []>("op_10233_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_10233_cast = einsum(equation = var_10233_equation_0, values = (var_10119_cast, var_10213_cast))[name = tensor<string, []>("op_10233_cast")];
            tensor<string, []> var_10235_equation_0 = const()[name = tensor<string, []>("op_10235_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_10235_cast = einsum(equation = var_10235_equation_0, values = (var_10123_cast, var_10214_cast))[name = tensor<string, []>("op_10235_cast")];
            tensor<string, []> var_10237_equation_0 = const()[name = tensor<string, []>("op_10237_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_10237_cast = einsum(equation = var_10237_equation_0, values = (var_10123_cast, var_10215_cast))[name = tensor<string, []>("op_10237_cast")];
            tensor<string, []> var_10239_equation_0 = const()[name = tensor<string, []>("op_10239_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_10239_cast = einsum(equation = var_10239_equation_0, values = (var_10127_cast, var_10216_cast))[name = tensor<string, []>("op_10239_cast")];
            tensor<string, []> var_10241_equation_0 = const()[name = tensor<string, []>("op_10241_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_10241_cast = einsum(equation = var_10241_equation_0, values = (var_10127_cast, var_10217_cast))[name = tensor<string, []>("op_10241_cast")];
            tensor<string, []> var_10243_equation_0 = const()[name = tensor<string, []>("op_10243_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_10243_cast = einsum(equation = var_10243_equation_0, values = (var_10131_cast, var_10218_cast))[name = tensor<string, []>("op_10243_cast")];
            tensor<string, []> var_10245_equation_0 = const()[name = tensor<string, []>("op_10245_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_10245_cast = einsum(equation = var_10245_equation_0, values = (var_10131_cast, var_10219_cast))[name = tensor<string, []>("op_10245_cast")];
            tensor<string, []> var_10247_equation_0 = const()[name = tensor<string, []>("op_10247_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_10247_cast = einsum(equation = var_10247_equation_0, values = (var_10135_cast, var_10220_cast))[name = tensor<string, []>("op_10247_cast")];
            tensor<string, []> var_10249_equation_0 = const()[name = tensor<string, []>("op_10249_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_10249_cast = einsum(equation = var_10249_equation_0, values = (var_10135_cast, var_10221_cast))[name = tensor<string, []>("op_10249_cast")];
            tensor<string, []> var_10251_equation_0 = const()[name = tensor<string, []>("op_10251_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_10251_cast = einsum(equation = var_10251_equation_0, values = (var_10139_cast, var_10222_cast))[name = tensor<string, []>("op_10251_cast")];
            tensor<string, []> var_10253_equation_0 = const()[name = tensor<string, []>("op_10253_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_10253_cast = einsum(equation = var_10253_equation_0, values = (var_10139_cast, var_10223_cast))[name = tensor<string, []>("op_10253_cast")];
            tensor<string, []> var_10255_equation_0 = const()[name = tensor<string, []>("op_10255_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_10255_cast = einsum(equation = var_10255_equation_0, values = (var_10143_cast, var_10224_cast))[name = tensor<string, []>("op_10255_cast")];
            tensor<string, []> var_10257_equation_0 = const()[name = tensor<string, []>("op_10257_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 80, 1, 512]> var_10257_cast = einsum(equation = var_10257_equation_0, values = (var_10143_cast, var_10225_cast))[name = tensor<string, []>("op_10257_cast")];
            tensor<bool, []> var_10259_interleave_0 = const()[name = tensor<string, []>("op_10259_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_10259_cast = concat(axis = var_8160, interleave = var_10259_interleave_0, values = (var_10227_cast, var_10229_cast))[name = tensor<string, []>("op_10259_cast")];
            tensor<bool, []> var_10261_interleave_0 = const()[name = tensor<string, []>("op_10261_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_10261_cast = concat(axis = var_8160, interleave = var_10261_interleave_0, values = (var_10231_cast, var_10233_cast))[name = tensor<string, []>("op_10261_cast")];
            tensor<bool, []> var_10263_interleave_0 = const()[name = tensor<string, []>("op_10263_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_10263_cast = concat(axis = var_8160, interleave = var_10263_interleave_0, values = (var_10235_cast, var_10237_cast))[name = tensor<string, []>("op_10263_cast")];
            tensor<bool, []> var_10265_interleave_0 = const()[name = tensor<string, []>("op_10265_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_10265_cast = concat(axis = var_8160, interleave = var_10265_interleave_0, values = (var_10239_cast, var_10241_cast))[name = tensor<string, []>("op_10265_cast")];
            tensor<bool, []> var_10267_interleave_0 = const()[name = tensor<string, []>("op_10267_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_10267_cast = concat(axis = var_8160, interleave = var_10267_interleave_0, values = (var_10243_cast, var_10245_cast))[name = tensor<string, []>("op_10267_cast")];
            tensor<bool, []> var_10269_interleave_0 = const()[name = tensor<string, []>("op_10269_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_10269_cast = concat(axis = var_8160, interleave = var_10269_interleave_0, values = (var_10247_cast, var_10249_cast))[name = tensor<string, []>("op_10269_cast")];
            tensor<bool, []> var_10271_interleave_0 = const()[name = tensor<string, []>("op_10271_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_10271_cast = concat(axis = var_8160, interleave = var_10271_interleave_0, values = (var_10251_cast, var_10253_cast))[name = tensor<string, []>("op_10271_cast")];
            tensor<bool, []> var_10273_interleave_0 = const()[name = tensor<string, []>("op_10273_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 80, 1, 1024]> var_10273_cast = concat(axis = var_8160, interleave = var_10273_interleave_0, values = (var_10255_cast, var_10257_cast))[name = tensor<string, []>("op_10273_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_8182, interleave = input_437_interleave_0, values = (var_10259_cast, var_10261_cast, var_10263_cast, var_10265_cast, var_10267_cast, var_10269_cast, var_10271_cast, var_10273_cast))[name = tensor<string, []>("input_437_cast")];
            tensor<int32, [2]> var_10279 = const()[name = tensor<string, []>("op_10279"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10281 = const()[name = tensor<string, []>("op_10281"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_10283_pad_type_0 = const()[name = tensor<string, []>("op_10283_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_10283_pad_0 = const()[name = tensor<string, []>("op_10283_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(623924864))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(624232128))), name = tensor<string, []>("up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            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, []>(624232320)))];
            tensor<fp16, [2, 640, 1, 1024]> var_10283_cast = conv(bias = up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_10281, groups = var_8182, pad = var_10283_pad_0, pad_type = var_10283_pad_type_0, strides = var_10279, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_437_cast)[name = tensor<string, []>("op_10283_cast")];
            tensor<fp16, [2, 640, 1, 1024]> inputs_77_cast = add(x = var_10283_cast, y = inputs_75_cast)[name = tensor<string, []>("inputs_77_cast")];
            tensor<int32, [1]> var_10287 = const()[name = tensor<string, []>("op_10287"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_77_cast = reduce_mean(axes = var_10287, keep_dims = var_8177, x = inputs_77_cast)[name = tensor<string, []>("channels_mean_77_cast")];
            tensor<fp16, [2, 640, 1, 1024]> zero_mean_77_cast = sub(x = inputs_77_cast, y = channels_mean_77_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_77_cast")];
            tensor<int32, [1]> var_10291 = const()[name = tensor<string, []>("op_10291"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_10292_cast = reduce_mean(axes = var_10291, keep_dims = var_8177, x = zero_mean_sq_77_cast)[name = tensor<string, []>("op_10292_cast")];
            tensor<fp16, []> var_10293_to_fp16 = const()[name = tensor<string, []>("op_10293_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_10294_cast = add(x = var_10292_cast, y = var_10293_to_fp16)[name = tensor<string, []>("op_10294_cast")];
            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_10294_cast)[name = tensor<string, []>("denom_77_cast")];
            tensor<fp16, [2, 640, 1, 1024]> out_77_cast = mul(x = zero_mean_77_cast, y = denom_77_cast)[name = tensor<string, []>("out_77_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_10298_to_fp16 = const()[name = tensor<string, []>("op_10298_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(624233664)))];
            tensor<fp16, [2, 640, 1, 1024]> var_10299_cast = add(x = out_77_cast, y = var_10298_to_fp16)[name = tensor<string, []>("op_10299_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_10301_to_fp16 = const()[name = tensor<string, []>("op_10301_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(624235008)))];
            tensor<fp16, [2, 640, 1, 1024]> input_439_cast = mul(x = var_10299_cast, y = var_10301_to_fp16)[name = tensor<string, []>("input_439_cast")];
            tensor<int32, [2]> var_10309 = const()[name = tensor<string, []>("op_10309"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10311 = const()[name = tensor<string, []>("op_10311"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_10313_pad_type_0 = const()[name = tensor<string, []>("op_10313_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_10313_pad_0 = const()[name = tensor<string, []>("op_10313_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2457600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(624236352))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(626694016))), name = tensor<string, []>("up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([5120, 640, 1, 1])];
            tensor<fp16, [5120]> up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3840]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(626694208))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(626698112))), name = tensor<string, []>("up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([5120])];
            tensor<fp16, [2, 5120, 1, 1024]> var_10313_cast = conv(bias = up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_10311, groups = var_8182, pad = var_10313_pad_0, pad_type = var_10313_pad_type_0, strides = var_10309, weight = up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_439_cast)[name = tensor<string, []>("op_10313_cast")];
            tensor<int32, [2]> var_10314_split_sizes_0 = const()[name = tensor<string, []>("op_10314_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_10314_axis_0 = const()[name = tensor<string, []>("op_10314_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 1024]> var_10314_cast_0, tensor<fp16, [2, 2560, 1, 1024]> var_10314_cast_1 = split(axis = var_10314_axis_0, split_sizes = var_10314_split_sizes_0, x = var_10313_cast)[name = tensor<string, []>("op_10314_cast")];
            tensor<string, []> var_10316_mode_0 = const()[name = tensor<string, []>("op_10316_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 1024]> var_10316_cast = gelu(mode = var_10316_mode_0, x = var_10314_cast_1)[name = tensor<string, []>("op_10316_cast")];
            tensor<fp16, [2, 2560, 1, 1024]> input_441_cast = mul(x = var_10314_cast_0, y = var_10316_cast)[name = tensor<string, []>("input_441_cast")];
            tensor<int32, [2]> var_10320 = const()[name = tensor<string, []>("op_10320"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10322 = const()[name = tensor<string, []>("op_10322"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_10324_pad_type_0 = const()[name = tensor<string, []>("op_10324_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_10324_pad_0 = const()[name = tensor<string, []>("op_10324_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1228800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(626698304))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(627927168))), name = tensor<string, []>("up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 2560, 1, 1])];
            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, []>(627927360)))];
            tensor<fp16, [2, 640, 1, 1024]> var_10324_cast = conv(bias = up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_10322, groups = var_8182, pad = var_10324_pad_0, pad_type = var_10324_pad_type_0, strides = var_10320, weight = up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_441_cast)[name = tensor<string, []>("op_10324_cast")];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_273_cast = add(x = var_10324_cast, y = inputs_77_cast)[name = tensor<string, []>("hidden_states_273_cast")];
            tensor<int32, [4]> var_10326 = const()[name = tensor<string, []>("op_10326"), val = tensor<int32, [4]>([2, 640, 32, 32])];
            tensor<fp16, [2, 640, 32, 32]> input_443_cast = reshape(shape = var_10326, x = hidden_states_273_cast)[name = tensor<string, []>("input_443_cast")];
            tensor<int32, [2]> var_10330 = const()[name = tensor<string, []>("op_10330"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10332 = const()[name = tensor<string, []>("op_10332"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(627928704))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(628235968))), name = tensor<string, []>("up_blocks_2_attentions_2_proj_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 1, 1])];
            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, []>(628236160)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_275_cast = conv(bias = up_blocks_2_attentions_2_proj_out_bias_to_fp16, dilations = var_10332, groups = var_8182, pad = hidden_states_275_pad_0, pad_type = hidden_states_275_pad_type_0, strides = var_10330, weight = up_blocks_2_attentions_2_proj_out_weight_to_fp16_palettized, x = input_443_cast)[name = tensor<string, []>("hidden_states_275_cast")];
            tensor<fp16, [2, 640, 32, 32]> input_445_cast = add(x = hidden_states_275_cast, y = hidden_states_263_cast)[name = tensor<string, []>("input_445_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)[name = tensor<string, []>("input_447_cast")];
            tensor<int32, [2]> var_10341 = const()[name = tensor<string, []>("op_10341"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10343 = const()[name = tensor<string, []>("op_10343"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2764800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(628237504))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(631002368))), name = tensor<string, []>("up_blocks_2_upsamplers_0_conv_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 640, 3, 3])];
            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, []>(631002560)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_277_cast = conv(bias = up_blocks_2_upsamplers_0_conv_bias_to_fp16, dilations = var_10343, groups = var_8182, pad = hidden_states_277_pad_0, pad_type = hidden_states_277_pad_type_0, strides = var_10341, weight = up_blocks_2_upsamplers_0_conv_weight_to_fp16_palettized, x = input_447_cast)[name = tensor<string, []>("hidden_states_277_cast")];
            tensor<int32, []> var_10347 = const()[name = tensor<string, []>("op_10347"), val = tensor<int32, []>(3)];
            tensor<bool, []> var_10370 = const()[name = tensor<string, []>("op_10370"), val = tensor<bool, []>(true)];
            tensor<int32, []> var_10375 = const()[name = tensor<string, []>("op_10375"), 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_10375, interleave = input_449_interleave_0, values = (hidden_states_277_cast, input_61_cast))[name = tensor<string, []>("input_449_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)[name = tensor<string, []>("reshape_204_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)[name = tensor<string, []>("reduce_mean_153_cast")];
            tensor<fp16, [2, 32, 30, 64, 64]> sub_102_cast = sub(x = reshape_204_cast, y = reduce_mean_153_cast)[name = tensor<string, []>("sub_102_cast")];
            tensor<fp16, [2, 32, 30, 64, 64]> square_51_cast = square(x = sub_102_cast)[name = tensor<string, []>("square_51_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)[name = tensor<string, []>("reduce_mean_155_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)[name = tensor<string, []>("add_102_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_51_cast = sqrt(x = add_102_cast)[name = tensor<string, []>("sqrt_51_cast")];
            tensor<fp16, [2, 32, 30, 64, 64]> real_div_51_cast = real_div(x = sub_102_cast, y = sqrt_51_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_205_cast")];
            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, []>(631003904)))];
            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, []>(631005888)))];
            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_97_mean_0_to_fp16, variance = add_97_variance_0_to_fp16, x = reshape_205_cast)[name = tensor<string, []>("add_103_cast")];
            tensor<fp16, [2, 960, 64, 64]> input_453_cast = silu(x = add_103_cast)[name = tensor<string, []>("input_453_cast")];
            tensor<int32, [2]> var_10402 = const()[name = tensor<string, []>("op_10402"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10404 = const()[name = tensor<string, []>("op_10404"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2073600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(631007872))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(633081536))), name = tensor<string, []>("up_blocks_3_resnets_0_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 960, 3, 3])];
            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, []>(633081728)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_279_cast = conv(bias = up_blocks_3_resnets_0_conv1_bias_to_fp16, dilations = var_10404, groups = var_10375, pad = hidden_states_279_pad_0, pad_type = hidden_states_279_pad_type_0, strides = var_10402, weight = up_blocks_3_resnets_0_conv1_weight_to_fp16_palettized, x = input_453_cast)[name = tensor<string, []>("hidden_states_279_cast")];
            tensor<int32, [2]> var_10410 = const()[name = tensor<string, []>("op_10410"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10412 = const()[name = tensor<string, []>("op_10412"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(633082432))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(633389696))), name = tensor<string, []>("up_blocks_3_resnets_0_time_emb_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 1280, 1, 1])];
            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, []>(633389888)))];
            tensor<fp16, [2, 320, 1, 1]> temb_39_cast = conv(bias = up_blocks_3_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_10412, groups = var_10375, pad = temb_39_pad_0, pad_type = temb_39_pad_type_0, strides = var_10410, weight = up_blocks_3_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor<string, []>("temb_39_cast")];
            tensor<fp16, [2, 320, 64, 64]> input_457_cast = add(x = hidden_states_279_cast, y = temb_39_cast)[name = tensor<string, []>("input_457_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)[name = tensor<string, []>("reshape_208_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)[name = tensor<string, []>("reduce_mean_156_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> sub_104_cast = sub(x = reshape_208_cast, y = reduce_mean_156_cast)[name = tensor<string, []>("sub_104_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> square_52_cast = square(x = sub_104_cast)[name = tensor<string, []>("square_52_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)[name = tensor<string, []>("reduce_mean_158_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)[name = tensor<string, []>("add_104_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_52_cast = sqrt(x = add_104_cast)[name = tensor<string, []>("sqrt_52_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> real_div_52_cast = real_div(x = sub_104_cast, y = sqrt_52_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_209_cast")];
            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, []>(633390592)))];
            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, []>(633391296)))];
            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_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_209_cast)[name = tensor<string, []>("add_105_cast")];
            tensor<fp16, [2, 320, 64, 64]> input_461_cast = silu(x = add_105_cast)[name = tensor<string, []>("input_461_cast")];
            tensor<int32, [2]> var_10422 = const()[name = tensor<string, []>("op_10422"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10424 = const()[name = tensor<string, []>("op_10424"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [691200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(633392000))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(634083264))), name = tensor<string, []>("up_blocks_3_resnets_0_conv2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 3, 3])];
            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, []>(634083456)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_281_cast = conv(bias = up_blocks_3_resnets_0_conv2_bias_to_fp16, dilations = var_10424, groups = var_10375, pad = hidden_states_281_pad_0, pad_type = hidden_states_281_pad_type_0, strides = var_10422, weight = up_blocks_3_resnets_0_conv2_weight_to_fp16_palettized, x = input_461_cast)[name = tensor<string, []>("hidden_states_281_cast")];
            tensor<int32, [2]> var_10429 = const()[name = tensor<string, []>("op_10429"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10431 = const()[name = tensor<string, []>("op_10431"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [230400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(634084160))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(634314624))), name = tensor<string, []>("up_blocks_3_resnets_0_conv_shortcut_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 960, 1, 1])];
            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, []>(634314816)))];
            tensor<fp16, [2, 320, 64, 64]> x_23_cast = conv(bias = up_blocks_3_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_10431, groups = var_10375, pad = x_23_pad_0, pad_type = x_23_pad_type_0, strides = var_10429, weight = up_blocks_3_resnets_0_conv_shortcut_weight_to_fp16_palettized, x = input_449_cast)[name = tensor<string, []>("x_23_cast")];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_283_cast = add(x = x_23_cast, y = hidden_states_281_cast)[name = tensor<string, []>("hidden_states_283_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)[name = tensor<string, []>("reshape_212_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)[name = tensor<string, []>("reduce_mean_159_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> sub_106_cast = sub(x = reshape_212_cast, y = reduce_mean_159_cast)[name = tensor<string, []>("sub_106_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> square_53_cast = square(x = sub_106_cast)[name = tensor<string, []>("square_53_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)[name = tensor<string, []>("reduce_mean_161_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)[name = tensor<string, []>("add_106_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_53_cast = sqrt(x = add_106_cast)[name = tensor<string, []>("sqrt_53_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> real_div_53_cast = real_div(x = sub_106_cast, y = sqrt_53_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_213_cast")];
            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, []>(634315520)))];
            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, []>(634316224)))];
            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_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_213_cast)[name = tensor<string, []>("add_107_cast")];
            tensor<int32, [2]> var_10451 = const()[name = tensor<string, []>("op_10451"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10453 = const()[name = tensor<string, []>("op_10453"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(634316928))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(634393792))), name = tensor<string, []>("up_blocks_3_attentions_0_proj_in_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            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, []>(634393984)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_285_cast = conv(bias = up_blocks_3_attentions_0_proj_in_bias_to_fp16, dilations = var_10453, groups = var_10375, pad = hidden_states_285_pad_0, pad_type = hidden_states_285_pad_type_0, strides = var_10451, weight = up_blocks_3_attentions_0_proj_in_weight_to_fp16_palettized, x = add_107_cast)[name = tensor<string, []>("hidden_states_285_cast")];
            tensor<int32, [4]> var_10458 = const()[name = tensor<string, []>("op_10458"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<fp16, [2, 320, 1, 4096]> inputs_79_cast = reshape(shape = var_10458, x = hidden_states_285_cast)[name = tensor<string, []>("inputs_79_cast")];
            tensor<int32, [1]> var_10468 = const()[name = tensor<string, []>("op_10468"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_79_cast = reduce_mean(axes = var_10468, keep_dims = var_10370, x = inputs_79_cast)[name = tensor<string, []>("channels_mean_79_cast")];
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_79_cast = sub(x = inputs_79_cast, y = channels_mean_79_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_79_cast")];
            tensor<int32, [1]> var_10472 = const()[name = tensor<string, []>("op_10472"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_10473_cast = reduce_mean(axes = var_10472, keep_dims = var_10370, x = zero_mean_sq_79_cast)[name = tensor<string, []>("op_10473_cast")];
            tensor<fp16, []> var_10474_to_fp16 = const()[name = tensor<string, []>("op_10474_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_10475_cast = add(x = var_10473_cast, y = var_10474_to_fp16)[name = tensor<string, []>("op_10475_cast")];
            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_10475_cast)[name = tensor<string, []>("denom_79_cast")];
            tensor<fp16, [2, 320, 1, 4096]> out_79_cast = mul(x = zero_mean_79_cast, y = denom_79_cast)[name = tensor<string, []>("out_79_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_10479_to_fp16 = const()[name = tensor<string, []>("op_10479_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(634394688)))];
            tensor<fp16, [2, 320, 1, 4096]> var_10480_cast = add(x = out_79_cast, y = var_10479_to_fp16)[name = tensor<string, []>("op_10480_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_10482_to_fp16 = const()[name = tensor<string, []>("op_10482_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(634395392)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_287_cast = mul(x = var_10480_cast, y = var_10482_to_fp16)[name = tensor<string, []>("hidden_states_287_cast")];
            tensor<int32, [2]> var_10489 = const()[name = tensor<string, []>("op_10489"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10491 = const()[name = tensor<string, []>("op_10491"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(634396096))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(634472960))), name = tensor<string, []>("up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            tensor<fp16, [2, 320, 1, 4096]> q_53_cast = conv(dilations = var_10491, groups = var_10375, pad = q_53_pad_0, pad_type = q_53_pad_type_0, strides = var_10489, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_287_cast)[name = tensor<string, []>("q_53_cast")];
            tensor<int32, [2]> var_10495 = const()[name = tensor<string, []>("op_10495"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10497 = const()[name = tensor<string, []>("op_10497"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(634473152))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(634550016))), name = tensor<string, []>("up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            tensor<fp16, [2, 320, 1, 4096]> k_105_cast = conv(dilations = var_10497, groups = var_10375, pad = k_105_pad_0, pad_type = k_105_pad_type_0, strides = var_10495, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_287_cast)[name = tensor<string, []>("k_105_cast")];
            tensor<int32, [2]> var_10501 = const()[name = tensor<string, []>("op_10501"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10503 = const()[name = tensor<string, []>("op_10503"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(634550208))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(634627072))), name = tensor<string, []>("up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            tensor<fp16, [2, 320, 1, 4096]> v_53_cast = conv(dilations = var_10503, groups = var_10375, pad = v_53_pad_0, pad_type = v_53_pad_type_0, strides = var_10501, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_287_cast)[name = tensor<string, []>("v_53_cast")];
            tensor<int32, [4]> var_10507_begin_0 = const()[name = tensor<string, []>("op_10507_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10507_end_0 = const()[name = tensor<string, []>("op_10507_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_10507_end_mask_0 = const()[name = tensor<string, []>("op_10507_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_10507_cast = slice_by_index(begin = var_10507_begin_0, end = var_10507_end_0, end_mask = var_10507_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_10507_cast")];
            tensor<int32, [4]> var_10511_begin_0 = const()[name = tensor<string, []>("op_10511_begin_0"), val = tensor<int32, [4]>([0, 40, 0, 0])];
            tensor<int32, [4]> var_10511_end_0 = const()[name = tensor<string, []>("op_10511_end_0"), val = tensor<int32, [4]>([2, 80, 1, 4096])];
            tensor<bool, [4]> var_10511_end_mask_0 = const()[name = tensor<string, []>("op_10511_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_10511_cast = slice_by_index(begin = var_10511_begin_0, end = var_10511_end_0, end_mask = var_10511_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_10511_cast")];
            tensor<int32, [4]> var_10515_begin_0 = const()[name = tensor<string, []>("op_10515_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_10515_end_0 = const()[name = tensor<string, []>("op_10515_end_0"), val = tensor<int32, [4]>([2, 120, 1, 4096])];
            tensor<bool, [4]> var_10515_end_mask_0 = const()[name = tensor<string, []>("op_10515_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_10515_cast = slice_by_index(begin = var_10515_begin_0, end = var_10515_end_0, end_mask = var_10515_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_10515_cast")];
            tensor<int32, [4]> var_10519_begin_0 = const()[name = tensor<string, []>("op_10519_begin_0"), val = tensor<int32, [4]>([0, 120, 0, 0])];
            tensor<int32, [4]> var_10519_end_0 = const()[name = tensor<string, []>("op_10519_end_0"), val = tensor<int32, [4]>([2, 160, 1, 4096])];
            tensor<bool, [4]> var_10519_end_mask_0 = const()[name = tensor<string, []>("op_10519_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_10519_cast = slice_by_index(begin = var_10519_begin_0, end = var_10519_end_0, end_mask = var_10519_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_10519_cast")];
            tensor<int32, [4]> var_10523_begin_0 = const()[name = tensor<string, []>("op_10523_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_10523_end_0 = const()[name = tensor<string, []>("op_10523_end_0"), val = tensor<int32, [4]>([2, 200, 1, 4096])];
            tensor<bool, [4]> var_10523_end_mask_0 = const()[name = tensor<string, []>("op_10523_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_10523_cast = slice_by_index(begin = var_10523_begin_0, end = var_10523_end_0, end_mask = var_10523_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_10523_cast")];
            tensor<int32, [4]> var_10527_begin_0 = const()[name = tensor<string, []>("op_10527_begin_0"), val = tensor<int32, [4]>([0, 200, 0, 0])];
            tensor<int32, [4]> var_10527_end_0 = const()[name = tensor<string, []>("op_10527_end_0"), val = tensor<int32, [4]>([2, 240, 1, 4096])];
            tensor<bool, [4]> var_10527_end_mask_0 = const()[name = tensor<string, []>("op_10527_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_10527_cast = slice_by_index(begin = var_10527_begin_0, end = var_10527_end_0, end_mask = var_10527_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_10527_cast")];
            tensor<int32, [4]> var_10531_begin_0 = const()[name = tensor<string, []>("op_10531_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_10531_end_0 = const()[name = tensor<string, []>("op_10531_end_0"), val = tensor<int32, [4]>([2, 280, 1, 4096])];
            tensor<bool, [4]> var_10531_end_mask_0 = const()[name = tensor<string, []>("op_10531_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_10531_cast = slice_by_index(begin = var_10531_begin_0, end = var_10531_end_0, end_mask = var_10531_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_10531_cast")];
            tensor<int32, [4]> var_10535_begin_0 = const()[name = tensor<string, []>("op_10535_begin_0"), val = tensor<int32, [4]>([0, 280, 0, 0])];
            tensor<int32, [4]> var_10535_end_0 = const()[name = tensor<string, []>("op_10535_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_10535_end_mask_0 = const()[name = tensor<string, []>("op_10535_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_10535_cast = slice_by_index(begin = var_10535_begin_0, end = var_10535_end_0, end_mask = var_10535_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_10535_cast")];
            tensor<int32, [4]> var_10538_begin_0 = const()[name = tensor<string, []>("op_10538_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10538_end_0 = const()[name = tensor<string, []>("op_10538_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_10538_end_mask_0 = const()[name = tensor<string, []>("op_10538_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10538_cast = slice_by_index(begin = var_10538_begin_0, end = var_10538_end_0, end_mask = var_10538_end_mask_0, x = var_10507_cast)[name = tensor<string, []>("op_10538_cast")];
            tensor<int32, [4]> var_10539_begin_0 = const()[name = tensor<string, []>("op_10539_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10539_end_0 = const()[name = tensor<string, []>("op_10539_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_10539_end_mask_0 = const()[name = tensor<string, []>("op_10539_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10539_cast = slice_by_index(begin = var_10539_begin_0, end = var_10539_end_0, end_mask = var_10539_end_mask_0, x = var_10507_cast)[name = tensor<string, []>("op_10539_cast")];
            tensor<int32, [4]> var_10540_begin_0 = const()[name = tensor<string, []>("op_10540_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_10540_end_0 = const()[name = tensor<string, []>("op_10540_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_10540_end_mask_0 = const()[name = tensor<string, []>("op_10540_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10540_cast = slice_by_index(begin = var_10540_begin_0, end = var_10540_end_0, end_mask = var_10540_end_mask_0, x = var_10507_cast)[name = tensor<string, []>("op_10540_cast")];
            tensor<int32, [4]> var_10541_begin_0 = const()[name = tensor<string, []>("op_10541_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_10541_end_0 = const()[name = tensor<string, []>("op_10541_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_10541_end_mask_0 = const()[name = tensor<string, []>("op_10541_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10541_cast = slice_by_index(begin = var_10541_begin_0, end = var_10541_end_0, end_mask = var_10541_end_mask_0, x = var_10507_cast)[name = tensor<string, []>("op_10541_cast")];
            tensor<int32, [4]> var_10542_begin_0 = const()[name = tensor<string, []>("op_10542_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_10542_end_0 = const()[name = tensor<string, []>("op_10542_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_10542_end_mask_0 = const()[name = tensor<string, []>("op_10542_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10542_cast = slice_by_index(begin = var_10542_begin_0, end = var_10542_end_0, end_mask = var_10542_end_mask_0, x = var_10507_cast)[name = tensor<string, []>("op_10542_cast")];
            tensor<int32, [4]> var_10543_begin_0 = const()[name = tensor<string, []>("op_10543_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_10543_end_0 = const()[name = tensor<string, []>("op_10543_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_10543_end_mask_0 = const()[name = tensor<string, []>("op_10543_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10543_cast = slice_by_index(begin = var_10543_begin_0, end = var_10543_end_0, end_mask = var_10543_end_mask_0, x = var_10507_cast)[name = tensor<string, []>("op_10543_cast")];
            tensor<int32, [4]> var_10544_begin_0 = const()[name = tensor<string, []>("op_10544_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_10544_end_0 = const()[name = tensor<string, []>("op_10544_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_10544_end_mask_0 = const()[name = tensor<string, []>("op_10544_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10544_cast = slice_by_index(begin = var_10544_begin_0, end = var_10544_end_0, end_mask = var_10544_end_mask_0, x = var_10507_cast)[name = tensor<string, []>("op_10544_cast")];
            tensor<int32, [4]> var_10545_begin_0 = const()[name = tensor<string, []>("op_10545_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_10545_end_0 = const()[name = tensor<string, []>("op_10545_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_10545_end_mask_0 = const()[name = tensor<string, []>("op_10545_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10545_cast = slice_by_index(begin = var_10545_begin_0, end = var_10545_end_0, end_mask = var_10545_end_mask_0, x = var_10507_cast)[name = tensor<string, []>("op_10545_cast")];
            tensor<int32, [4]> var_10546_begin_0 = const()[name = tensor<string, []>("op_10546_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10546_end_0 = const()[name = tensor<string, []>("op_10546_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_10546_end_mask_0 = const()[name = tensor<string, []>("op_10546_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10546_cast = slice_by_index(begin = var_10546_begin_0, end = var_10546_end_0, end_mask = var_10546_end_mask_0, x = var_10511_cast)[name = tensor<string, []>("op_10546_cast")];
            tensor<int32, [4]> var_10547_begin_0 = const()[name = tensor<string, []>("op_10547_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10547_end_0 = const()[name = tensor<string, []>("op_10547_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_10547_end_mask_0 = const()[name = tensor<string, []>("op_10547_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10547_cast = slice_by_index(begin = var_10547_begin_0, end = var_10547_end_0, end_mask = var_10547_end_mask_0, x = var_10511_cast)[name = tensor<string, []>("op_10547_cast")];
            tensor<int32, [4]> var_10548_begin_0 = const()[name = tensor<string, []>("op_10548_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_10548_end_0 = const()[name = tensor<string, []>("op_10548_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_10548_end_mask_0 = const()[name = tensor<string, []>("op_10548_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10548_cast = slice_by_index(begin = var_10548_begin_0, end = var_10548_end_0, end_mask = var_10548_end_mask_0, x = var_10511_cast)[name = tensor<string, []>("op_10548_cast")];
            tensor<int32, [4]> var_10549_begin_0 = const()[name = tensor<string, []>("op_10549_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_10549_end_0 = const()[name = tensor<string, []>("op_10549_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_10549_end_mask_0 = const()[name = tensor<string, []>("op_10549_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10549_cast = slice_by_index(begin = var_10549_begin_0, end = var_10549_end_0, end_mask = var_10549_end_mask_0, x = var_10511_cast)[name = tensor<string, []>("op_10549_cast")];
            tensor<int32, [4]> var_10550_begin_0 = const()[name = tensor<string, []>("op_10550_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_10550_end_0 = const()[name = tensor<string, []>("op_10550_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_10550_end_mask_0 = const()[name = tensor<string, []>("op_10550_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10550_cast = slice_by_index(begin = var_10550_begin_0, end = var_10550_end_0, end_mask = var_10550_end_mask_0, x = var_10511_cast)[name = tensor<string, []>("op_10550_cast")];
            tensor<int32, [4]> var_10551_begin_0 = const()[name = tensor<string, []>("op_10551_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_10551_end_0 = const()[name = tensor<string, []>("op_10551_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_10551_end_mask_0 = const()[name = tensor<string, []>("op_10551_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10551_cast = slice_by_index(begin = var_10551_begin_0, end = var_10551_end_0, end_mask = var_10551_end_mask_0, x = var_10511_cast)[name = tensor<string, []>("op_10551_cast")];
            tensor<int32, [4]> var_10552_begin_0 = const()[name = tensor<string, []>("op_10552_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_10552_end_0 = const()[name = tensor<string, []>("op_10552_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_10552_end_mask_0 = const()[name = tensor<string, []>("op_10552_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10552_cast = slice_by_index(begin = var_10552_begin_0, end = var_10552_end_0, end_mask = var_10552_end_mask_0, x = var_10511_cast)[name = tensor<string, []>("op_10552_cast")];
            tensor<int32, [4]> var_10553_begin_0 = const()[name = tensor<string, []>("op_10553_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_10553_end_0 = const()[name = tensor<string, []>("op_10553_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_10553_end_mask_0 = const()[name = tensor<string, []>("op_10553_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10553_cast = slice_by_index(begin = var_10553_begin_0, end = var_10553_end_0, end_mask = var_10553_end_mask_0, x = var_10511_cast)[name = tensor<string, []>("op_10553_cast")];
            tensor<int32, [4]> var_10554_begin_0 = const()[name = tensor<string, []>("op_10554_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10554_end_0 = const()[name = tensor<string, []>("op_10554_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_10554_end_mask_0 = const()[name = tensor<string, []>("op_10554_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10554_cast = slice_by_index(begin = var_10554_begin_0, end = var_10554_end_0, end_mask = var_10554_end_mask_0, x = var_10515_cast)[name = tensor<string, []>("op_10554_cast")];
            tensor<int32, [4]> var_10555_begin_0 = const()[name = tensor<string, []>("op_10555_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10555_end_0 = const()[name = tensor<string, []>("op_10555_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_10555_end_mask_0 = const()[name = tensor<string, []>("op_10555_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10555_cast = slice_by_index(begin = var_10555_begin_0, end = var_10555_end_0, end_mask = var_10555_end_mask_0, x = var_10515_cast)[name = tensor<string, []>("op_10555_cast")];
            tensor<int32, [4]> var_10556_begin_0 = const()[name = tensor<string, []>("op_10556_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_10556_end_0 = const()[name = tensor<string, []>("op_10556_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_10556_end_mask_0 = const()[name = tensor<string, []>("op_10556_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10556_cast = slice_by_index(begin = var_10556_begin_0, end = var_10556_end_0, end_mask = var_10556_end_mask_0, x = var_10515_cast)[name = tensor<string, []>("op_10556_cast")];
            tensor<int32, [4]> var_10557_begin_0 = const()[name = tensor<string, []>("op_10557_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_10557_end_0 = const()[name = tensor<string, []>("op_10557_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_10557_end_mask_0 = const()[name = tensor<string, []>("op_10557_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10557_cast = slice_by_index(begin = var_10557_begin_0, end = var_10557_end_0, end_mask = var_10557_end_mask_0, x = var_10515_cast)[name = tensor<string, []>("op_10557_cast")];
            tensor<int32, [4]> var_10558_begin_0 = const()[name = tensor<string, []>("op_10558_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_10558_end_0 = const()[name = tensor<string, []>("op_10558_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_10558_end_mask_0 = const()[name = tensor<string, []>("op_10558_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10558_cast = slice_by_index(begin = var_10558_begin_0, end = var_10558_end_0, end_mask = var_10558_end_mask_0, x = var_10515_cast)[name = tensor<string, []>("op_10558_cast")];
            tensor<int32, [4]> var_10559_begin_0 = const()[name = tensor<string, []>("op_10559_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_10559_end_0 = const()[name = tensor<string, []>("op_10559_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_10559_end_mask_0 = const()[name = tensor<string, []>("op_10559_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10559_cast = slice_by_index(begin = var_10559_begin_0, end = var_10559_end_0, end_mask = var_10559_end_mask_0, x = var_10515_cast)[name = tensor<string, []>("op_10559_cast")];
            tensor<int32, [4]> var_10560_begin_0 = const()[name = tensor<string, []>("op_10560_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_10560_end_0 = const()[name = tensor<string, []>("op_10560_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_10560_end_mask_0 = const()[name = tensor<string, []>("op_10560_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10560_cast = slice_by_index(begin = var_10560_begin_0, end = var_10560_end_0, end_mask = var_10560_end_mask_0, x = var_10515_cast)[name = tensor<string, []>("op_10560_cast")];
            tensor<int32, [4]> var_10561_begin_0 = const()[name = tensor<string, []>("op_10561_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_10561_end_0 = const()[name = tensor<string, []>("op_10561_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_10561_end_mask_0 = const()[name = tensor<string, []>("op_10561_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10561_cast = slice_by_index(begin = var_10561_begin_0, end = var_10561_end_0, end_mask = var_10561_end_mask_0, x = var_10515_cast)[name = tensor<string, []>("op_10561_cast")];
            tensor<int32, [4]> var_10562_begin_0 = const()[name = tensor<string, []>("op_10562_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10562_end_0 = const()[name = tensor<string, []>("op_10562_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_10562_end_mask_0 = const()[name = tensor<string, []>("op_10562_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10562_cast = slice_by_index(begin = var_10562_begin_0, end = var_10562_end_0, end_mask = var_10562_end_mask_0, x = var_10519_cast)[name = tensor<string, []>("op_10562_cast")];
            tensor<int32, [4]> var_10563_begin_0 = const()[name = tensor<string, []>("op_10563_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10563_end_0 = const()[name = tensor<string, []>("op_10563_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_10563_end_mask_0 = const()[name = tensor<string, []>("op_10563_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10563_cast = slice_by_index(begin = var_10563_begin_0, end = var_10563_end_0, end_mask = var_10563_end_mask_0, x = var_10519_cast)[name = tensor<string, []>("op_10563_cast")];
            tensor<int32, [4]> var_10564_begin_0 = const()[name = tensor<string, []>("op_10564_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_10564_end_0 = const()[name = tensor<string, []>("op_10564_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_10564_end_mask_0 = const()[name = tensor<string, []>("op_10564_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10564_cast = slice_by_index(begin = var_10564_begin_0, end = var_10564_end_0, end_mask = var_10564_end_mask_0, x = var_10519_cast)[name = tensor<string, []>("op_10564_cast")];
            tensor<int32, [4]> var_10565_begin_0 = const()[name = tensor<string, []>("op_10565_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_10565_end_0 = const()[name = tensor<string, []>("op_10565_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_10565_end_mask_0 = const()[name = tensor<string, []>("op_10565_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10565_cast = slice_by_index(begin = var_10565_begin_0, end = var_10565_end_0, end_mask = var_10565_end_mask_0, x = var_10519_cast)[name = tensor<string, []>("op_10565_cast")];
            tensor<int32, [4]> var_10566_begin_0 = const()[name = tensor<string, []>("op_10566_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_10566_end_0 = const()[name = tensor<string, []>("op_10566_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_10566_end_mask_0 = const()[name = tensor<string, []>("op_10566_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10566_cast = slice_by_index(begin = var_10566_begin_0, end = var_10566_end_0, end_mask = var_10566_end_mask_0, x = var_10519_cast)[name = tensor<string, []>("op_10566_cast")];
            tensor<int32, [4]> var_10567_begin_0 = const()[name = tensor<string, []>("op_10567_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_10567_end_0 = const()[name = tensor<string, []>("op_10567_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_10567_end_mask_0 = const()[name = tensor<string, []>("op_10567_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10567_cast = slice_by_index(begin = var_10567_begin_0, end = var_10567_end_0, end_mask = var_10567_end_mask_0, x = var_10519_cast)[name = tensor<string, []>("op_10567_cast")];
            tensor<int32, [4]> var_10568_begin_0 = const()[name = tensor<string, []>("op_10568_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_10568_end_0 = const()[name = tensor<string, []>("op_10568_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_10568_end_mask_0 = const()[name = tensor<string, []>("op_10568_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10568_cast = slice_by_index(begin = var_10568_begin_0, end = var_10568_end_0, end_mask = var_10568_end_mask_0, x = var_10519_cast)[name = tensor<string, []>("op_10568_cast")];
            tensor<int32, [4]> var_10569_begin_0 = const()[name = tensor<string, []>("op_10569_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_10569_end_0 = const()[name = tensor<string, []>("op_10569_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_10569_end_mask_0 = const()[name = tensor<string, []>("op_10569_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10569_cast = slice_by_index(begin = var_10569_begin_0, end = var_10569_end_0, end_mask = var_10569_end_mask_0, x = var_10519_cast)[name = tensor<string, []>("op_10569_cast")];
            tensor<int32, [4]> var_10570_begin_0 = const()[name = tensor<string, []>("op_10570_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10570_end_0 = const()[name = tensor<string, []>("op_10570_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_10570_end_mask_0 = const()[name = tensor<string, []>("op_10570_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10570_cast = slice_by_index(begin = var_10570_begin_0, end = var_10570_end_0, end_mask = var_10570_end_mask_0, x = var_10523_cast)[name = tensor<string, []>("op_10570_cast")];
            tensor<int32, [4]> var_10571_begin_0 = const()[name = tensor<string, []>("op_10571_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10571_end_0 = const()[name = tensor<string, []>("op_10571_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_10571_end_mask_0 = const()[name = tensor<string, []>("op_10571_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10571_cast = slice_by_index(begin = var_10571_begin_0, end = var_10571_end_0, end_mask = var_10571_end_mask_0, x = var_10523_cast)[name = tensor<string, []>("op_10571_cast")];
            tensor<int32, [4]> var_10572_begin_0 = const()[name = tensor<string, []>("op_10572_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_10572_end_0 = const()[name = tensor<string, []>("op_10572_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_10572_end_mask_0 = const()[name = tensor<string, []>("op_10572_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10572_cast = slice_by_index(begin = var_10572_begin_0, end = var_10572_end_0, end_mask = var_10572_end_mask_0, x = var_10523_cast)[name = tensor<string, []>("op_10572_cast")];
            tensor<int32, [4]> var_10573_begin_0 = const()[name = tensor<string, []>("op_10573_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_10573_end_0 = const()[name = tensor<string, []>("op_10573_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_10573_end_mask_0 = const()[name = tensor<string, []>("op_10573_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10573_cast = slice_by_index(begin = var_10573_begin_0, end = var_10573_end_0, end_mask = var_10573_end_mask_0, x = var_10523_cast)[name = tensor<string, []>("op_10573_cast")];
            tensor<int32, [4]> var_10574_begin_0 = const()[name = tensor<string, []>("op_10574_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_10574_end_0 = const()[name = tensor<string, []>("op_10574_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_10574_end_mask_0 = const()[name = tensor<string, []>("op_10574_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10574_cast = slice_by_index(begin = var_10574_begin_0, end = var_10574_end_0, end_mask = var_10574_end_mask_0, x = var_10523_cast)[name = tensor<string, []>("op_10574_cast")];
            tensor<int32, [4]> var_10575_begin_0 = const()[name = tensor<string, []>("op_10575_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_10575_end_0 = const()[name = tensor<string, []>("op_10575_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_10575_end_mask_0 = const()[name = tensor<string, []>("op_10575_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10575_cast = slice_by_index(begin = var_10575_begin_0, end = var_10575_end_0, end_mask = var_10575_end_mask_0, x = var_10523_cast)[name = tensor<string, []>("op_10575_cast")];
            tensor<int32, [4]> var_10576_begin_0 = const()[name = tensor<string, []>("op_10576_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_10576_end_0 = const()[name = tensor<string, []>("op_10576_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_10576_end_mask_0 = const()[name = tensor<string, []>("op_10576_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10576_cast = slice_by_index(begin = var_10576_begin_0, end = var_10576_end_0, end_mask = var_10576_end_mask_0, x = var_10523_cast)[name = tensor<string, []>("op_10576_cast")];
            tensor<int32, [4]> var_10577_begin_0 = const()[name = tensor<string, []>("op_10577_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_10577_end_0 = const()[name = tensor<string, []>("op_10577_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_10577_end_mask_0 = const()[name = tensor<string, []>("op_10577_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10577_cast = slice_by_index(begin = var_10577_begin_0, end = var_10577_end_0, end_mask = var_10577_end_mask_0, x = var_10523_cast)[name = tensor<string, []>("op_10577_cast")];
            tensor<int32, [4]> var_10578_begin_0 = const()[name = tensor<string, []>("op_10578_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10578_end_0 = const()[name = tensor<string, []>("op_10578_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_10578_end_mask_0 = const()[name = tensor<string, []>("op_10578_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10578_cast = slice_by_index(begin = var_10578_begin_0, end = var_10578_end_0, end_mask = var_10578_end_mask_0, x = var_10527_cast)[name = tensor<string, []>("op_10578_cast")];
            tensor<int32, [4]> var_10579_begin_0 = const()[name = tensor<string, []>("op_10579_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10579_end_0 = const()[name = tensor<string, []>("op_10579_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_10579_end_mask_0 = const()[name = tensor<string, []>("op_10579_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10579_cast = slice_by_index(begin = var_10579_begin_0, end = var_10579_end_0, end_mask = var_10579_end_mask_0, x = var_10527_cast)[name = tensor<string, []>("op_10579_cast")];
            tensor<int32, [4]> var_10580_begin_0 = const()[name = tensor<string, []>("op_10580_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_10580_end_0 = const()[name = tensor<string, []>("op_10580_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_10580_end_mask_0 = const()[name = tensor<string, []>("op_10580_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10580_cast = slice_by_index(begin = var_10580_begin_0, end = var_10580_end_0, end_mask = var_10580_end_mask_0, x = var_10527_cast)[name = tensor<string, []>("op_10580_cast")];
            tensor<int32, [4]> var_10581_begin_0 = const()[name = tensor<string, []>("op_10581_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_10581_end_0 = const()[name = tensor<string, []>("op_10581_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_10581_end_mask_0 = const()[name = tensor<string, []>("op_10581_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10581_cast = slice_by_index(begin = var_10581_begin_0, end = var_10581_end_0, end_mask = var_10581_end_mask_0, x = var_10527_cast)[name = tensor<string, []>("op_10581_cast")];
            tensor<int32, [4]> var_10582_begin_0 = const()[name = tensor<string, []>("op_10582_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_10582_end_0 = const()[name = tensor<string, []>("op_10582_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_10582_end_mask_0 = const()[name = tensor<string, []>("op_10582_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10582_cast = slice_by_index(begin = var_10582_begin_0, end = var_10582_end_0, end_mask = var_10582_end_mask_0, x = var_10527_cast)[name = tensor<string, []>("op_10582_cast")];
            tensor<int32, [4]> var_10583_begin_0 = const()[name = tensor<string, []>("op_10583_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_10583_end_0 = const()[name = tensor<string, []>("op_10583_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_10583_end_mask_0 = const()[name = tensor<string, []>("op_10583_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10583_cast = slice_by_index(begin = var_10583_begin_0, end = var_10583_end_0, end_mask = var_10583_end_mask_0, x = var_10527_cast)[name = tensor<string, []>("op_10583_cast")];
            tensor<int32, [4]> var_10584_begin_0 = const()[name = tensor<string, []>("op_10584_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_10584_end_0 = const()[name = tensor<string, []>("op_10584_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_10584_end_mask_0 = const()[name = tensor<string, []>("op_10584_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10584_cast = slice_by_index(begin = var_10584_begin_0, end = var_10584_end_0, end_mask = var_10584_end_mask_0, x = var_10527_cast)[name = tensor<string, []>("op_10584_cast")];
            tensor<int32, [4]> var_10585_begin_0 = const()[name = tensor<string, []>("op_10585_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_10585_end_0 = const()[name = tensor<string, []>("op_10585_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_10585_end_mask_0 = const()[name = tensor<string, []>("op_10585_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10585_cast = slice_by_index(begin = var_10585_begin_0, end = var_10585_end_0, end_mask = var_10585_end_mask_0, x = var_10527_cast)[name = tensor<string, []>("op_10585_cast")];
            tensor<int32, [4]> var_10586_begin_0 = const()[name = tensor<string, []>("op_10586_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10586_end_0 = const()[name = tensor<string, []>("op_10586_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_10586_end_mask_0 = const()[name = tensor<string, []>("op_10586_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10586_cast = slice_by_index(begin = var_10586_begin_0, end = var_10586_end_0, end_mask = var_10586_end_mask_0, x = var_10531_cast)[name = tensor<string, []>("op_10586_cast")];
            tensor<int32, [4]> var_10587_begin_0 = const()[name = tensor<string, []>("op_10587_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10587_end_0 = const()[name = tensor<string, []>("op_10587_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_10587_end_mask_0 = const()[name = tensor<string, []>("op_10587_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10587_cast = slice_by_index(begin = var_10587_begin_0, end = var_10587_end_0, end_mask = var_10587_end_mask_0, x = var_10531_cast)[name = tensor<string, []>("op_10587_cast")];
            tensor<int32, [4]> var_10588_begin_0 = const()[name = tensor<string, []>("op_10588_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_10588_end_0 = const()[name = tensor<string, []>("op_10588_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_10588_end_mask_0 = const()[name = tensor<string, []>("op_10588_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10588_cast = slice_by_index(begin = var_10588_begin_0, end = var_10588_end_0, end_mask = var_10588_end_mask_0, x = var_10531_cast)[name = tensor<string, []>("op_10588_cast")];
            tensor<int32, [4]> var_10589_begin_0 = const()[name = tensor<string, []>("op_10589_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_10589_end_0 = const()[name = tensor<string, []>("op_10589_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_10589_end_mask_0 = const()[name = tensor<string, []>("op_10589_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10589_cast = slice_by_index(begin = var_10589_begin_0, end = var_10589_end_0, end_mask = var_10589_end_mask_0, x = var_10531_cast)[name = tensor<string, []>("op_10589_cast")];
            tensor<int32, [4]> var_10590_begin_0 = const()[name = tensor<string, []>("op_10590_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_10590_end_0 = const()[name = tensor<string, []>("op_10590_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_10590_end_mask_0 = const()[name = tensor<string, []>("op_10590_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10590_cast = slice_by_index(begin = var_10590_begin_0, end = var_10590_end_0, end_mask = var_10590_end_mask_0, x = var_10531_cast)[name = tensor<string, []>("op_10590_cast")];
            tensor<int32, [4]> var_10591_begin_0 = const()[name = tensor<string, []>("op_10591_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_10591_end_0 = const()[name = tensor<string, []>("op_10591_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_10591_end_mask_0 = const()[name = tensor<string, []>("op_10591_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10591_cast = slice_by_index(begin = var_10591_begin_0, end = var_10591_end_0, end_mask = var_10591_end_mask_0, x = var_10531_cast)[name = tensor<string, []>("op_10591_cast")];
            tensor<int32, [4]> var_10592_begin_0 = const()[name = tensor<string, []>("op_10592_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_10592_end_0 = const()[name = tensor<string, []>("op_10592_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_10592_end_mask_0 = const()[name = tensor<string, []>("op_10592_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10592_cast = slice_by_index(begin = var_10592_begin_0, end = var_10592_end_0, end_mask = var_10592_end_mask_0, x = var_10531_cast)[name = tensor<string, []>("op_10592_cast")];
            tensor<int32, [4]> var_10593_begin_0 = const()[name = tensor<string, []>("op_10593_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_10593_end_0 = const()[name = tensor<string, []>("op_10593_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_10593_end_mask_0 = const()[name = tensor<string, []>("op_10593_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10593_cast = slice_by_index(begin = var_10593_begin_0, end = var_10593_end_0, end_mask = var_10593_end_mask_0, x = var_10531_cast)[name = tensor<string, []>("op_10593_cast")];
            tensor<int32, [4]> var_10594_begin_0 = const()[name = tensor<string, []>("op_10594_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10594_end_0 = const()[name = tensor<string, []>("op_10594_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_10594_end_mask_0 = const()[name = tensor<string, []>("op_10594_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10594_cast = slice_by_index(begin = var_10594_begin_0, end = var_10594_end_0, end_mask = var_10594_end_mask_0, x = var_10535_cast)[name = tensor<string, []>("op_10594_cast")];
            tensor<int32, [4]> var_10595_begin_0 = const()[name = tensor<string, []>("op_10595_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10595_end_0 = const()[name = tensor<string, []>("op_10595_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_10595_end_mask_0 = const()[name = tensor<string, []>("op_10595_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10595_cast = slice_by_index(begin = var_10595_begin_0, end = var_10595_end_0, end_mask = var_10595_end_mask_0, x = var_10535_cast)[name = tensor<string, []>("op_10595_cast")];
            tensor<int32, [4]> var_10596_begin_0 = const()[name = tensor<string, []>("op_10596_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_10596_end_0 = const()[name = tensor<string, []>("op_10596_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_10596_end_mask_0 = const()[name = tensor<string, []>("op_10596_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10596_cast = slice_by_index(begin = var_10596_begin_0, end = var_10596_end_0, end_mask = var_10596_end_mask_0, x = var_10535_cast)[name = tensor<string, []>("op_10596_cast")];
            tensor<int32, [4]> var_10597_begin_0 = const()[name = tensor<string, []>("op_10597_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_10597_end_0 = const()[name = tensor<string, []>("op_10597_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_10597_end_mask_0 = const()[name = tensor<string, []>("op_10597_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10597_cast = slice_by_index(begin = var_10597_begin_0, end = var_10597_end_0, end_mask = var_10597_end_mask_0, x = var_10535_cast)[name = tensor<string, []>("op_10597_cast")];
            tensor<int32, [4]> var_10598_begin_0 = const()[name = tensor<string, []>("op_10598_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_10598_end_0 = const()[name = tensor<string, []>("op_10598_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_10598_end_mask_0 = const()[name = tensor<string, []>("op_10598_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10598_cast = slice_by_index(begin = var_10598_begin_0, end = var_10598_end_0, end_mask = var_10598_end_mask_0, x = var_10535_cast)[name = tensor<string, []>("op_10598_cast")];
            tensor<int32, [4]> var_10599_begin_0 = const()[name = tensor<string, []>("op_10599_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_10599_end_0 = const()[name = tensor<string, []>("op_10599_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_10599_end_mask_0 = const()[name = tensor<string, []>("op_10599_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10599_cast = slice_by_index(begin = var_10599_begin_0, end = var_10599_end_0, end_mask = var_10599_end_mask_0, x = var_10535_cast)[name = tensor<string, []>("op_10599_cast")];
            tensor<int32, [4]> var_10600_begin_0 = const()[name = tensor<string, []>("op_10600_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_10600_end_0 = const()[name = tensor<string, []>("op_10600_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_10600_end_mask_0 = const()[name = tensor<string, []>("op_10600_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10600_cast = slice_by_index(begin = var_10600_begin_0, end = var_10600_end_0, end_mask = var_10600_end_mask_0, x = var_10535_cast)[name = tensor<string, []>("op_10600_cast")];
            tensor<int32, [4]> var_10601_begin_0 = const()[name = tensor<string, []>("op_10601_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_10601_end_0 = const()[name = tensor<string, []>("op_10601_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_10601_end_mask_0 = const()[name = tensor<string, []>("op_10601_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_10601_cast = slice_by_index(begin = var_10601_begin_0, end = var_10601_end_0, end_mask = var_10601_end_mask_0, x = var_10535_cast)[name = tensor<string, []>("op_10601_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_10606_begin_0 = const()[name = tensor<string, []>("op_10606_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10606_end_0 = const()[name = tensor<string, []>("op_10606_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 40])];
            tensor<bool, [4]> var_10606_end_mask_0 = const()[name = tensor<string, []>("op_10606_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)[name = tensor<string, []>("transpose_5")];
            tensor<fp16, [2, 4096, 1, 40]> var_10606_cast = slice_by_index(begin = var_10606_begin_0, end = var_10606_end_0, end_mask = var_10606_end_mask_0, x = transpose_5)[name = tensor<string, []>("op_10606_cast")];
            tensor<int32, [4]> var_10610_begin_0 = const()[name = tensor<string, []>("op_10610_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 40])];
            tensor<int32, [4]> var_10610_end_0 = const()[name = tensor<string, []>("op_10610_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 80])];
            tensor<bool, [4]> var_10610_end_mask_0 = const()[name = tensor<string, []>("op_10610_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_10610_cast = slice_by_index(begin = var_10610_begin_0, end = var_10610_end_0, end_mask = var_10610_end_mask_0, x = transpose_5)[name = tensor<string, []>("op_10610_cast")];
            tensor<int32, [4]> var_10614_begin_0 = const()[name = tensor<string, []>("op_10614_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 80])];
            tensor<int32, [4]> var_10614_end_0 = const()[name = tensor<string, []>("op_10614_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 120])];
            tensor<bool, [4]> var_10614_end_mask_0 = const()[name = tensor<string, []>("op_10614_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_10614_cast = slice_by_index(begin = var_10614_begin_0, end = var_10614_end_0, end_mask = var_10614_end_mask_0, x = transpose_5)[name = tensor<string, []>("op_10614_cast")];
            tensor<int32, [4]> var_10618_begin_0 = const()[name = tensor<string, []>("op_10618_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 120])];
            tensor<int32, [4]> var_10618_end_0 = const()[name = tensor<string, []>("op_10618_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 160])];
            tensor<bool, [4]> var_10618_end_mask_0 = const()[name = tensor<string, []>("op_10618_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_10618_cast = slice_by_index(begin = var_10618_begin_0, end = var_10618_end_0, end_mask = var_10618_end_mask_0, x = transpose_5)[name = tensor<string, []>("op_10618_cast")];
            tensor<int32, [4]> var_10622_begin_0 = const()[name = tensor<string, []>("op_10622_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_10622_end_0 = const()[name = tensor<string, []>("op_10622_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 200])];
            tensor<bool, [4]> var_10622_end_mask_0 = const()[name = tensor<string, []>("op_10622_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_10622_cast = slice_by_index(begin = var_10622_begin_0, end = var_10622_end_0, end_mask = var_10622_end_mask_0, x = transpose_5)[name = tensor<string, []>("op_10622_cast")];
            tensor<int32, [4]> var_10626_begin_0 = const()[name = tensor<string, []>("op_10626_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 200])];
            tensor<int32, [4]> var_10626_end_0 = const()[name = tensor<string, []>("op_10626_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 240])];
            tensor<bool, [4]> var_10626_end_mask_0 = const()[name = tensor<string, []>("op_10626_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_10626_cast = slice_by_index(begin = var_10626_begin_0, end = var_10626_end_0, end_mask = var_10626_end_mask_0, x = transpose_5)[name = tensor<string, []>("op_10626_cast")];
            tensor<int32, [4]> var_10630_begin_0 = const()[name = tensor<string, []>("op_10630_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 240])];
            tensor<int32, [4]> var_10630_end_0 = const()[name = tensor<string, []>("op_10630_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 280])];
            tensor<bool, [4]> var_10630_end_mask_0 = const()[name = tensor<string, []>("op_10630_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_10630_cast = slice_by_index(begin = var_10630_begin_0, end = var_10630_end_0, end_mask = var_10630_end_mask_0, x = transpose_5)[name = tensor<string, []>("op_10630_cast")];
            tensor<int32, [4]> var_10634_begin_0 = const()[name = tensor<string, []>("op_10634_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 280])];
            tensor<int32, [4]> var_10634_end_0 = const()[name = tensor<string, []>("op_10634_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_10634_end_mask_0 = const()[name = tensor<string, []>("op_10634_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_10634_cast = slice_by_index(begin = var_10634_begin_0, end = var_10634_end_0, end_mask = var_10634_end_mask_0, x = transpose_5)[name = tensor<string, []>("op_10634_cast")];
            tensor<int32, [4]> var_10636_begin_0 = const()[name = tensor<string, []>("op_10636_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10636_end_0 = const()[name = tensor<string, []>("op_10636_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_10636_end_mask_0 = const()[name = tensor<string, []>("op_10636_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_10636_cast = slice_by_index(begin = var_10636_begin_0, end = var_10636_end_0, end_mask = var_10636_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_10636_cast")];
            tensor<int32, [4]> var_10640_begin_0 = const()[name = tensor<string, []>("op_10640_begin_0"), val = tensor<int32, [4]>([0, 40, 0, 0])];
            tensor<int32, [4]> var_10640_end_0 = const()[name = tensor<string, []>("op_10640_end_0"), val = tensor<int32, [4]>([2, 80, 1, 4096])];
            tensor<bool, [4]> var_10640_end_mask_0 = const()[name = tensor<string, []>("op_10640_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_10640_cast = slice_by_index(begin = var_10640_begin_0, end = var_10640_end_0, end_mask = var_10640_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_10640_cast")];
            tensor<int32, [4]> var_10644_begin_0 = const()[name = tensor<string, []>("op_10644_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_10644_end_0 = const()[name = tensor<string, []>("op_10644_end_0"), val = tensor<int32, [4]>([2, 120, 1, 4096])];
            tensor<bool, [4]> var_10644_end_mask_0 = const()[name = tensor<string, []>("op_10644_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_10644_cast = slice_by_index(begin = var_10644_begin_0, end = var_10644_end_0, end_mask = var_10644_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_10644_cast")];
            tensor<int32, [4]> var_10648_begin_0 = const()[name = tensor<string, []>("op_10648_begin_0"), val = tensor<int32, [4]>([0, 120, 0, 0])];
            tensor<int32, [4]> var_10648_end_0 = const()[name = tensor<string, []>("op_10648_end_0"), val = tensor<int32, [4]>([2, 160, 1, 4096])];
            tensor<bool, [4]> var_10648_end_mask_0 = const()[name = tensor<string, []>("op_10648_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_10648_cast = slice_by_index(begin = var_10648_begin_0, end = var_10648_end_0, end_mask = var_10648_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_10648_cast")];
            tensor<int32, [4]> var_10652_begin_0 = const()[name = tensor<string, []>("op_10652_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_10652_end_0 = const()[name = tensor<string, []>("op_10652_end_0"), val = tensor<int32, [4]>([2, 200, 1, 4096])];
            tensor<bool, [4]> var_10652_end_mask_0 = const()[name = tensor<string, []>("op_10652_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_10652_cast = slice_by_index(begin = var_10652_begin_0, end = var_10652_end_0, end_mask = var_10652_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_10652_cast")];
            tensor<int32, [4]> var_10656_begin_0 = const()[name = tensor<string, []>("op_10656_begin_0"), val = tensor<int32, [4]>([0, 200, 0, 0])];
            tensor<int32, [4]> var_10656_end_0 = const()[name = tensor<string, []>("op_10656_end_0"), val = tensor<int32, [4]>([2, 240, 1, 4096])];
            tensor<bool, [4]> var_10656_end_mask_0 = const()[name = tensor<string, []>("op_10656_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_10656_cast = slice_by_index(begin = var_10656_begin_0, end = var_10656_end_0, end_mask = var_10656_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_10656_cast")];
            tensor<int32, [4]> var_10660_begin_0 = const()[name = tensor<string, []>("op_10660_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_10660_end_0 = const()[name = tensor<string, []>("op_10660_end_0"), val = tensor<int32, [4]>([2, 280, 1, 4096])];
            tensor<bool, [4]> var_10660_end_mask_0 = const()[name = tensor<string, []>("op_10660_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_10660_cast = slice_by_index(begin = var_10660_begin_0, end = var_10660_end_0, end_mask = var_10660_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_10660_cast")];
            tensor<int32, [4]> var_10664_begin_0 = const()[name = tensor<string, []>("op_10664_begin_0"), val = tensor<int32, [4]>([0, 280, 0, 0])];
            tensor<int32, [4]> var_10664_end_0 = const()[name = tensor<string, []>("op_10664_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_10664_end_mask_0 = const()[name = tensor<string, []>("op_10664_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_10664_cast = slice_by_index(begin = var_10664_begin_0, end = var_10664_end_0, end_mask = var_10664_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_10664_cast")];
            tensor<string, []> var_10668_equation_0 = const()[name = tensor<string, []>("op_10668_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10668_cast = einsum(equation = var_10668_equation_0, values = (var_10606_cast, var_10538_cast))[name = tensor<string, []>("op_10668_cast")];
            tensor<fp16, []> var_10669_to_fp16 = const()[name = tensor<string, []>("op_10669_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_833_cast = mul(x = var_10668_cast, y = var_10669_to_fp16)[name = tensor<string, []>("aw_chunk_833_cast")];
            tensor<string, []> var_10672_equation_0 = const()[name = tensor<string, []>("op_10672_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10672_cast = einsum(equation = var_10672_equation_0, values = (var_10606_cast, var_10539_cast))[name = tensor<string, []>("op_10672_cast")];
            tensor<fp16, []> var_10673_to_fp16 = const()[name = tensor<string, []>("op_10673_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_835_cast = mul(x = var_10672_cast, y = var_10673_to_fp16)[name = tensor<string, []>("aw_chunk_835_cast")];
            tensor<string, []> var_10676_equation_0 = const()[name = tensor<string, []>("op_10676_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10676_cast = einsum(equation = var_10676_equation_0, values = (var_10606_cast, var_10540_cast))[name = tensor<string, []>("op_10676_cast")];
            tensor<fp16, []> var_10677_to_fp16 = const()[name = tensor<string, []>("op_10677_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_837_cast = mul(x = var_10676_cast, y = var_10677_to_fp16)[name = tensor<string, []>("aw_chunk_837_cast")];
            tensor<string, []> var_10680_equation_0 = const()[name = tensor<string, []>("op_10680_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10680_cast = einsum(equation = var_10680_equation_0, values = (var_10606_cast, var_10541_cast))[name = tensor<string, []>("op_10680_cast")];
            tensor<fp16, []> var_10681_to_fp16 = const()[name = tensor<string, []>("op_10681_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_839_cast = mul(x = var_10680_cast, y = var_10681_to_fp16)[name = tensor<string, []>("aw_chunk_839_cast")];
            tensor<string, []> var_10684_equation_0 = const()[name = tensor<string, []>("op_10684_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10684_cast = einsum(equation = var_10684_equation_0, values = (var_10606_cast, var_10542_cast))[name = tensor<string, []>("op_10684_cast")];
            tensor<fp16, []> var_10685_to_fp16 = const()[name = tensor<string, []>("op_10685_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_841_cast = mul(x = var_10684_cast, y = var_10685_to_fp16)[name = tensor<string, []>("aw_chunk_841_cast")];
            tensor<string, []> var_10688_equation_0 = const()[name = tensor<string, []>("op_10688_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10688_cast = einsum(equation = var_10688_equation_0, values = (var_10606_cast, var_10543_cast))[name = tensor<string, []>("op_10688_cast")];
            tensor<fp16, []> var_10689_to_fp16 = const()[name = tensor<string, []>("op_10689_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_843_cast = mul(x = var_10688_cast, y = var_10689_to_fp16)[name = tensor<string, []>("aw_chunk_843_cast")];
            tensor<string, []> var_10692_equation_0 = const()[name = tensor<string, []>("op_10692_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10692_cast = einsum(equation = var_10692_equation_0, values = (var_10606_cast, var_10544_cast))[name = tensor<string, []>("op_10692_cast")];
            tensor<fp16, []> var_10693_to_fp16 = const()[name = tensor<string, []>("op_10693_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_845_cast = mul(x = var_10692_cast, y = var_10693_to_fp16)[name = tensor<string, []>("aw_chunk_845_cast")];
            tensor<string, []> var_10696_equation_0 = const()[name = tensor<string, []>("op_10696_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10696_cast = einsum(equation = var_10696_equation_0, values = (var_10606_cast, var_10545_cast))[name = tensor<string, []>("op_10696_cast")];
            tensor<fp16, []> var_10697_to_fp16 = const()[name = tensor<string, []>("op_10697_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_847_cast = mul(x = var_10696_cast, y = var_10697_to_fp16)[name = tensor<string, []>("aw_chunk_847_cast")];
            tensor<string, []> var_10700_equation_0 = const()[name = tensor<string, []>("op_10700_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10700_cast = einsum(equation = var_10700_equation_0, values = (var_10610_cast, var_10546_cast))[name = tensor<string, []>("op_10700_cast")];
            tensor<fp16, []> var_10701_to_fp16 = const()[name = tensor<string, []>("op_10701_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_849_cast = mul(x = var_10700_cast, y = var_10701_to_fp16)[name = tensor<string, []>("aw_chunk_849_cast")];
            tensor<string, []> var_10704_equation_0 = const()[name = tensor<string, []>("op_10704_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10704_cast = einsum(equation = var_10704_equation_0, values = (var_10610_cast, var_10547_cast))[name = tensor<string, []>("op_10704_cast")];
            tensor<fp16, []> var_10705_to_fp16 = const()[name = tensor<string, []>("op_10705_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_851_cast = mul(x = var_10704_cast, y = var_10705_to_fp16)[name = tensor<string, []>("aw_chunk_851_cast")];
            tensor<string, []> var_10708_equation_0 = const()[name = tensor<string, []>("op_10708_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10708_cast = einsum(equation = var_10708_equation_0, values = (var_10610_cast, var_10548_cast))[name = tensor<string, []>("op_10708_cast")];
            tensor<fp16, []> var_10709_to_fp16 = const()[name = tensor<string, []>("op_10709_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_853_cast = mul(x = var_10708_cast, y = var_10709_to_fp16)[name = tensor<string, []>("aw_chunk_853_cast")];
            tensor<string, []> var_10712_equation_0 = const()[name = tensor<string, []>("op_10712_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10712_cast = einsum(equation = var_10712_equation_0, values = (var_10610_cast, var_10549_cast))[name = tensor<string, []>("op_10712_cast")];
            tensor<fp16, []> var_10713_to_fp16 = const()[name = tensor<string, []>("op_10713_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_855_cast = mul(x = var_10712_cast, y = var_10713_to_fp16)[name = tensor<string, []>("aw_chunk_855_cast")];
            tensor<string, []> var_10716_equation_0 = const()[name = tensor<string, []>("op_10716_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10716_cast = einsum(equation = var_10716_equation_0, values = (var_10610_cast, var_10550_cast))[name = tensor<string, []>("op_10716_cast")];
            tensor<fp16, []> var_10717_to_fp16 = const()[name = tensor<string, []>("op_10717_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_857_cast = mul(x = var_10716_cast, y = var_10717_to_fp16)[name = tensor<string, []>("aw_chunk_857_cast")];
            tensor<string, []> var_10720_equation_0 = const()[name = tensor<string, []>("op_10720_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10720_cast = einsum(equation = var_10720_equation_0, values = (var_10610_cast, var_10551_cast))[name = tensor<string, []>("op_10720_cast")];
            tensor<fp16, []> var_10721_to_fp16 = const()[name = tensor<string, []>("op_10721_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_859_cast = mul(x = var_10720_cast, y = var_10721_to_fp16)[name = tensor<string, []>("aw_chunk_859_cast")];
            tensor<string, []> var_10724_equation_0 = const()[name = tensor<string, []>("op_10724_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10724_cast = einsum(equation = var_10724_equation_0, values = (var_10610_cast, var_10552_cast))[name = tensor<string, []>("op_10724_cast")];
            tensor<fp16, []> var_10725_to_fp16 = const()[name = tensor<string, []>("op_10725_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_861_cast = mul(x = var_10724_cast, y = var_10725_to_fp16)[name = tensor<string, []>("aw_chunk_861_cast")];
            tensor<string, []> var_10728_equation_0 = const()[name = tensor<string, []>("op_10728_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10728_cast = einsum(equation = var_10728_equation_0, values = (var_10610_cast, var_10553_cast))[name = tensor<string, []>("op_10728_cast")];
            tensor<fp16, []> var_10729_to_fp16 = const()[name = tensor<string, []>("op_10729_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_863_cast = mul(x = var_10728_cast, y = var_10729_to_fp16)[name = tensor<string, []>("aw_chunk_863_cast")];
            tensor<string, []> var_10732_equation_0 = const()[name = tensor<string, []>("op_10732_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10732_cast = einsum(equation = var_10732_equation_0, values = (var_10614_cast, var_10554_cast))[name = tensor<string, []>("op_10732_cast")];
            tensor<fp16, []> var_10733_to_fp16 = const()[name = tensor<string, []>("op_10733_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_865_cast = mul(x = var_10732_cast, y = var_10733_to_fp16)[name = tensor<string, []>("aw_chunk_865_cast")];
            tensor<string, []> var_10736_equation_0 = const()[name = tensor<string, []>("op_10736_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10736_cast = einsum(equation = var_10736_equation_0, values = (var_10614_cast, var_10555_cast))[name = tensor<string, []>("op_10736_cast")];
            tensor<fp16, []> var_10737_to_fp16 = const()[name = tensor<string, []>("op_10737_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_867_cast = mul(x = var_10736_cast, y = var_10737_to_fp16)[name = tensor<string, []>("aw_chunk_867_cast")];
            tensor<string, []> var_10740_equation_0 = const()[name = tensor<string, []>("op_10740_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10740_cast = einsum(equation = var_10740_equation_0, values = (var_10614_cast, var_10556_cast))[name = tensor<string, []>("op_10740_cast")];
            tensor<fp16, []> var_10741_to_fp16 = const()[name = tensor<string, []>("op_10741_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_869_cast = mul(x = var_10740_cast, y = var_10741_to_fp16)[name = tensor<string, []>("aw_chunk_869_cast")];
            tensor<string, []> var_10744_equation_0 = const()[name = tensor<string, []>("op_10744_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10744_cast = einsum(equation = var_10744_equation_0, values = (var_10614_cast, var_10557_cast))[name = tensor<string, []>("op_10744_cast")];
            tensor<fp16, []> var_10745_to_fp16 = const()[name = tensor<string, []>("op_10745_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_871_cast = mul(x = var_10744_cast, y = var_10745_to_fp16)[name = tensor<string, []>("aw_chunk_871_cast")];
            tensor<string, []> var_10748_equation_0 = const()[name = tensor<string, []>("op_10748_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10748_cast = einsum(equation = var_10748_equation_0, values = (var_10614_cast, var_10558_cast))[name = tensor<string, []>("op_10748_cast")];
            tensor<fp16, []> var_10749_to_fp16 = const()[name = tensor<string, []>("op_10749_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_873_cast = mul(x = var_10748_cast, y = var_10749_to_fp16)[name = tensor<string, []>("aw_chunk_873_cast")];
            tensor<string, []> var_10752_equation_0 = const()[name = tensor<string, []>("op_10752_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10752_cast = einsum(equation = var_10752_equation_0, values = (var_10614_cast, var_10559_cast))[name = tensor<string, []>("op_10752_cast")];
            tensor<fp16, []> var_10753_to_fp16 = const()[name = tensor<string, []>("op_10753_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_875_cast = mul(x = var_10752_cast, y = var_10753_to_fp16)[name = tensor<string, []>("aw_chunk_875_cast")];
            tensor<string, []> var_10756_equation_0 = const()[name = tensor<string, []>("op_10756_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10756_cast = einsum(equation = var_10756_equation_0, values = (var_10614_cast, var_10560_cast))[name = tensor<string, []>("op_10756_cast")];
            tensor<fp16, []> var_10757_to_fp16 = const()[name = tensor<string, []>("op_10757_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_877_cast = mul(x = var_10756_cast, y = var_10757_to_fp16)[name = tensor<string, []>("aw_chunk_877_cast")];
            tensor<string, []> var_10760_equation_0 = const()[name = tensor<string, []>("op_10760_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10760_cast = einsum(equation = var_10760_equation_0, values = (var_10614_cast, var_10561_cast))[name = tensor<string, []>("op_10760_cast")];
            tensor<fp16, []> var_10761_to_fp16 = const()[name = tensor<string, []>("op_10761_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_879_cast = mul(x = var_10760_cast, y = var_10761_to_fp16)[name = tensor<string, []>("aw_chunk_879_cast")];
            tensor<string, []> var_10764_equation_0 = const()[name = tensor<string, []>("op_10764_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10764_cast = einsum(equation = var_10764_equation_0, values = (var_10618_cast, var_10562_cast))[name = tensor<string, []>("op_10764_cast")];
            tensor<fp16, []> var_10765_to_fp16 = const()[name = tensor<string, []>("op_10765_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_881_cast = mul(x = var_10764_cast, y = var_10765_to_fp16)[name = tensor<string, []>("aw_chunk_881_cast")];
            tensor<string, []> var_10768_equation_0 = const()[name = tensor<string, []>("op_10768_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10768_cast = einsum(equation = var_10768_equation_0, values = (var_10618_cast, var_10563_cast))[name = tensor<string, []>("op_10768_cast")];
            tensor<fp16, []> var_10769_to_fp16 = const()[name = tensor<string, []>("op_10769_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_883_cast = mul(x = var_10768_cast, y = var_10769_to_fp16)[name = tensor<string, []>("aw_chunk_883_cast")];
            tensor<string, []> var_10772_equation_0 = const()[name = tensor<string, []>("op_10772_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10772_cast = einsum(equation = var_10772_equation_0, values = (var_10618_cast, var_10564_cast))[name = tensor<string, []>("op_10772_cast")];
            tensor<fp16, []> var_10773_to_fp16 = const()[name = tensor<string, []>("op_10773_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_885_cast = mul(x = var_10772_cast, y = var_10773_to_fp16)[name = tensor<string, []>("aw_chunk_885_cast")];
            tensor<string, []> var_10776_equation_0 = const()[name = tensor<string, []>("op_10776_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10776_cast = einsum(equation = var_10776_equation_0, values = (var_10618_cast, var_10565_cast))[name = tensor<string, []>("op_10776_cast")];
            tensor<fp16, []> var_10777_to_fp16 = const()[name = tensor<string, []>("op_10777_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_887_cast = mul(x = var_10776_cast, y = var_10777_to_fp16)[name = tensor<string, []>("aw_chunk_887_cast")];
            tensor<string, []> var_10780_equation_0 = const()[name = tensor<string, []>("op_10780_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10780_cast = einsum(equation = var_10780_equation_0, values = (var_10618_cast, var_10566_cast))[name = tensor<string, []>("op_10780_cast")];
            tensor<fp16, []> var_10781_to_fp16 = const()[name = tensor<string, []>("op_10781_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_889_cast = mul(x = var_10780_cast, y = var_10781_to_fp16)[name = tensor<string, []>("aw_chunk_889_cast")];
            tensor<string, []> var_10784_equation_0 = const()[name = tensor<string, []>("op_10784_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10784_cast = einsum(equation = var_10784_equation_0, values = (var_10618_cast, var_10567_cast))[name = tensor<string, []>("op_10784_cast")];
            tensor<fp16, []> var_10785_to_fp16 = const()[name = tensor<string, []>("op_10785_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_891_cast = mul(x = var_10784_cast, y = var_10785_to_fp16)[name = tensor<string, []>("aw_chunk_891_cast")];
            tensor<string, []> var_10788_equation_0 = const()[name = tensor<string, []>("op_10788_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10788_cast = einsum(equation = var_10788_equation_0, values = (var_10618_cast, var_10568_cast))[name = tensor<string, []>("op_10788_cast")];
            tensor<fp16, []> var_10789_to_fp16 = const()[name = tensor<string, []>("op_10789_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_893_cast = mul(x = var_10788_cast, y = var_10789_to_fp16)[name = tensor<string, []>("aw_chunk_893_cast")];
            tensor<string, []> var_10792_equation_0 = const()[name = tensor<string, []>("op_10792_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10792_cast = einsum(equation = var_10792_equation_0, values = (var_10618_cast, var_10569_cast))[name = tensor<string, []>("op_10792_cast")];
            tensor<fp16, []> var_10793_to_fp16 = const()[name = tensor<string, []>("op_10793_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_895_cast = mul(x = var_10792_cast, y = var_10793_to_fp16)[name = tensor<string, []>("aw_chunk_895_cast")];
            tensor<string, []> var_10796_equation_0 = const()[name = tensor<string, []>("op_10796_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10796_cast = einsum(equation = var_10796_equation_0, values = (var_10622_cast, var_10570_cast))[name = tensor<string, []>("op_10796_cast")];
            tensor<fp16, []> var_10797_to_fp16 = const()[name = tensor<string, []>("op_10797_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_897_cast = mul(x = var_10796_cast, y = var_10797_to_fp16)[name = tensor<string, []>("aw_chunk_897_cast")];
            tensor<string, []> var_10800_equation_0 = const()[name = tensor<string, []>("op_10800_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10800_cast = einsum(equation = var_10800_equation_0, values = (var_10622_cast, var_10571_cast))[name = tensor<string, []>("op_10800_cast")];
            tensor<fp16, []> var_10801_to_fp16 = const()[name = tensor<string, []>("op_10801_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_899_cast = mul(x = var_10800_cast, y = var_10801_to_fp16)[name = tensor<string, []>("aw_chunk_899_cast")];
            tensor<string, []> var_10804_equation_0 = const()[name = tensor<string, []>("op_10804_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10804_cast = einsum(equation = var_10804_equation_0, values = (var_10622_cast, var_10572_cast))[name = tensor<string, []>("op_10804_cast")];
            tensor<fp16, []> var_10805_to_fp16 = const()[name = tensor<string, []>("op_10805_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_901_cast = mul(x = var_10804_cast, y = var_10805_to_fp16)[name = tensor<string, []>("aw_chunk_901_cast")];
            tensor<string, []> var_10808_equation_0 = const()[name = tensor<string, []>("op_10808_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10808_cast = einsum(equation = var_10808_equation_0, values = (var_10622_cast, var_10573_cast))[name = tensor<string, []>("op_10808_cast")];
            tensor<fp16, []> var_10809_to_fp16 = const()[name = tensor<string, []>("op_10809_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_903_cast = mul(x = var_10808_cast, y = var_10809_to_fp16)[name = tensor<string, []>("aw_chunk_903_cast")];
            tensor<string, []> var_10812_equation_0 = const()[name = tensor<string, []>("op_10812_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10812_cast = einsum(equation = var_10812_equation_0, values = (var_10622_cast, var_10574_cast))[name = tensor<string, []>("op_10812_cast")];
            tensor<fp16, []> var_10813_to_fp16 = const()[name = tensor<string, []>("op_10813_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_905_cast = mul(x = var_10812_cast, y = var_10813_to_fp16)[name = tensor<string, []>("aw_chunk_905_cast")];
            tensor<string, []> var_10816_equation_0 = const()[name = tensor<string, []>("op_10816_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10816_cast = einsum(equation = var_10816_equation_0, values = (var_10622_cast, var_10575_cast))[name = tensor<string, []>("op_10816_cast")];
            tensor<fp16, []> var_10817_to_fp16 = const()[name = tensor<string, []>("op_10817_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_907_cast = mul(x = var_10816_cast, y = var_10817_to_fp16)[name = tensor<string, []>("aw_chunk_907_cast")];
            tensor<string, []> var_10820_equation_0 = const()[name = tensor<string, []>("op_10820_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10820_cast = einsum(equation = var_10820_equation_0, values = (var_10622_cast, var_10576_cast))[name = tensor<string, []>("op_10820_cast")];
            tensor<fp16, []> var_10821_to_fp16 = const()[name = tensor<string, []>("op_10821_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_909_cast = mul(x = var_10820_cast, y = var_10821_to_fp16)[name = tensor<string, []>("aw_chunk_909_cast")];
            tensor<string, []> var_10824_equation_0 = const()[name = tensor<string, []>("op_10824_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10824_cast = einsum(equation = var_10824_equation_0, values = (var_10622_cast, var_10577_cast))[name = tensor<string, []>("op_10824_cast")];
            tensor<fp16, []> var_10825_to_fp16 = const()[name = tensor<string, []>("op_10825_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_911_cast = mul(x = var_10824_cast, y = var_10825_to_fp16)[name = tensor<string, []>("aw_chunk_911_cast")];
            tensor<string, []> var_10828_equation_0 = const()[name = tensor<string, []>("op_10828_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10828_cast = einsum(equation = var_10828_equation_0, values = (var_10626_cast, var_10578_cast))[name = tensor<string, []>("op_10828_cast")];
            tensor<fp16, []> var_10829_to_fp16 = const()[name = tensor<string, []>("op_10829_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_913_cast = mul(x = var_10828_cast, y = var_10829_to_fp16)[name = tensor<string, []>("aw_chunk_913_cast")];
            tensor<string, []> var_10832_equation_0 = const()[name = tensor<string, []>("op_10832_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10832_cast = einsum(equation = var_10832_equation_0, values = (var_10626_cast, var_10579_cast))[name = tensor<string, []>("op_10832_cast")];
            tensor<fp16, []> var_10833_to_fp16 = const()[name = tensor<string, []>("op_10833_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_915_cast = mul(x = var_10832_cast, y = var_10833_to_fp16)[name = tensor<string, []>("aw_chunk_915_cast")];
            tensor<string, []> var_10836_equation_0 = const()[name = tensor<string, []>("op_10836_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10836_cast = einsum(equation = var_10836_equation_0, values = (var_10626_cast, var_10580_cast))[name = tensor<string, []>("op_10836_cast")];
            tensor<fp16, []> var_10837_to_fp16 = const()[name = tensor<string, []>("op_10837_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_917_cast = mul(x = var_10836_cast, y = var_10837_to_fp16)[name = tensor<string, []>("aw_chunk_917_cast")];
            tensor<string, []> var_10840_equation_0 = const()[name = tensor<string, []>("op_10840_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10840_cast = einsum(equation = var_10840_equation_0, values = (var_10626_cast, var_10581_cast))[name = tensor<string, []>("op_10840_cast")];
            tensor<fp16, []> var_10841_to_fp16 = const()[name = tensor<string, []>("op_10841_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_919_cast = mul(x = var_10840_cast, y = var_10841_to_fp16)[name = tensor<string, []>("aw_chunk_919_cast")];
            tensor<string, []> var_10844_equation_0 = const()[name = tensor<string, []>("op_10844_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10844_cast = einsum(equation = var_10844_equation_0, values = (var_10626_cast, var_10582_cast))[name = tensor<string, []>("op_10844_cast")];
            tensor<fp16, []> var_10845_to_fp16 = const()[name = tensor<string, []>("op_10845_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_921_cast = mul(x = var_10844_cast, y = var_10845_to_fp16)[name = tensor<string, []>("aw_chunk_921_cast")];
            tensor<string, []> var_10848_equation_0 = const()[name = tensor<string, []>("op_10848_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10848_cast = einsum(equation = var_10848_equation_0, values = (var_10626_cast, var_10583_cast))[name = tensor<string, []>("op_10848_cast")];
            tensor<fp16, []> var_10849_to_fp16 = const()[name = tensor<string, []>("op_10849_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_923_cast = mul(x = var_10848_cast, y = var_10849_to_fp16)[name = tensor<string, []>("aw_chunk_923_cast")];
            tensor<string, []> var_10852_equation_0 = const()[name = tensor<string, []>("op_10852_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10852_cast = einsum(equation = var_10852_equation_0, values = (var_10626_cast, var_10584_cast))[name = tensor<string, []>("op_10852_cast")];
            tensor<fp16, []> var_10853_to_fp16 = const()[name = tensor<string, []>("op_10853_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_925_cast = mul(x = var_10852_cast, y = var_10853_to_fp16)[name = tensor<string, []>("aw_chunk_925_cast")];
            tensor<string, []> var_10856_equation_0 = const()[name = tensor<string, []>("op_10856_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10856_cast = einsum(equation = var_10856_equation_0, values = (var_10626_cast, var_10585_cast))[name = tensor<string, []>("op_10856_cast")];
            tensor<fp16, []> var_10857_to_fp16 = const()[name = tensor<string, []>("op_10857_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_927_cast = mul(x = var_10856_cast, y = var_10857_to_fp16)[name = tensor<string, []>("aw_chunk_927_cast")];
            tensor<string, []> var_10860_equation_0 = const()[name = tensor<string, []>("op_10860_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10860_cast = einsum(equation = var_10860_equation_0, values = (var_10630_cast, var_10586_cast))[name = tensor<string, []>("op_10860_cast")];
            tensor<fp16, []> var_10861_to_fp16 = const()[name = tensor<string, []>("op_10861_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_929_cast = mul(x = var_10860_cast, y = var_10861_to_fp16)[name = tensor<string, []>("aw_chunk_929_cast")];
            tensor<string, []> var_10864_equation_0 = const()[name = tensor<string, []>("op_10864_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10864_cast = einsum(equation = var_10864_equation_0, values = (var_10630_cast, var_10587_cast))[name = tensor<string, []>("op_10864_cast")];
            tensor<fp16, []> var_10865_to_fp16 = const()[name = tensor<string, []>("op_10865_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_931_cast = mul(x = var_10864_cast, y = var_10865_to_fp16)[name = tensor<string, []>("aw_chunk_931_cast")];
            tensor<string, []> var_10868_equation_0 = const()[name = tensor<string, []>("op_10868_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10868_cast = einsum(equation = var_10868_equation_0, values = (var_10630_cast, var_10588_cast))[name = tensor<string, []>("op_10868_cast")];
            tensor<fp16, []> var_10869_to_fp16 = const()[name = tensor<string, []>("op_10869_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_933_cast = mul(x = var_10868_cast, y = var_10869_to_fp16)[name = tensor<string, []>("aw_chunk_933_cast")];
            tensor<string, []> var_10872_equation_0 = const()[name = tensor<string, []>("op_10872_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10872_cast = einsum(equation = var_10872_equation_0, values = (var_10630_cast, var_10589_cast))[name = tensor<string, []>("op_10872_cast")];
            tensor<fp16, []> var_10873_to_fp16 = const()[name = tensor<string, []>("op_10873_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_935_cast = mul(x = var_10872_cast, y = var_10873_to_fp16)[name = tensor<string, []>("aw_chunk_935_cast")];
            tensor<string, []> var_10876_equation_0 = const()[name = tensor<string, []>("op_10876_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10876_cast = einsum(equation = var_10876_equation_0, values = (var_10630_cast, var_10590_cast))[name = tensor<string, []>("op_10876_cast")];
            tensor<fp16, []> var_10877_to_fp16 = const()[name = tensor<string, []>("op_10877_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_937_cast = mul(x = var_10876_cast, y = var_10877_to_fp16)[name = tensor<string, []>("aw_chunk_937_cast")];
            tensor<string, []> var_10880_equation_0 = const()[name = tensor<string, []>("op_10880_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10880_cast = einsum(equation = var_10880_equation_0, values = (var_10630_cast, var_10591_cast))[name = tensor<string, []>("op_10880_cast")];
            tensor<fp16, []> var_10881_to_fp16 = const()[name = tensor<string, []>("op_10881_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_939_cast = mul(x = var_10880_cast, y = var_10881_to_fp16)[name = tensor<string, []>("aw_chunk_939_cast")];
            tensor<string, []> var_10884_equation_0 = const()[name = tensor<string, []>("op_10884_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10884_cast = einsum(equation = var_10884_equation_0, values = (var_10630_cast, var_10592_cast))[name = tensor<string, []>("op_10884_cast")];
            tensor<fp16, []> var_10885_to_fp16 = const()[name = tensor<string, []>("op_10885_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_941_cast = mul(x = var_10884_cast, y = var_10885_to_fp16)[name = tensor<string, []>("aw_chunk_941_cast")];
            tensor<string, []> var_10888_equation_0 = const()[name = tensor<string, []>("op_10888_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10888_cast = einsum(equation = var_10888_equation_0, values = (var_10630_cast, var_10593_cast))[name = tensor<string, []>("op_10888_cast")];
            tensor<fp16, []> var_10889_to_fp16 = const()[name = tensor<string, []>("op_10889_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_943_cast = mul(x = var_10888_cast, y = var_10889_to_fp16)[name = tensor<string, []>("aw_chunk_943_cast")];
            tensor<string, []> var_10892_equation_0 = const()[name = tensor<string, []>("op_10892_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10892_cast = einsum(equation = var_10892_equation_0, values = (var_10634_cast, var_10594_cast))[name = tensor<string, []>("op_10892_cast")];
            tensor<fp16, []> var_10893_to_fp16 = const()[name = tensor<string, []>("op_10893_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_945_cast = mul(x = var_10892_cast, y = var_10893_to_fp16)[name = tensor<string, []>("aw_chunk_945_cast")];
            tensor<string, []> var_10896_equation_0 = const()[name = tensor<string, []>("op_10896_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10896_cast = einsum(equation = var_10896_equation_0, values = (var_10634_cast, var_10595_cast))[name = tensor<string, []>("op_10896_cast")];
            tensor<fp16, []> var_10897_to_fp16 = const()[name = tensor<string, []>("op_10897_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_947_cast = mul(x = var_10896_cast, y = var_10897_to_fp16)[name = tensor<string, []>("aw_chunk_947_cast")];
            tensor<string, []> var_10900_equation_0 = const()[name = tensor<string, []>("op_10900_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10900_cast = einsum(equation = var_10900_equation_0, values = (var_10634_cast, var_10596_cast))[name = tensor<string, []>("op_10900_cast")];
            tensor<fp16, []> var_10901_to_fp16 = const()[name = tensor<string, []>("op_10901_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_949_cast = mul(x = var_10900_cast, y = var_10901_to_fp16)[name = tensor<string, []>("aw_chunk_949_cast")];
            tensor<string, []> var_10904_equation_0 = const()[name = tensor<string, []>("op_10904_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10904_cast = einsum(equation = var_10904_equation_0, values = (var_10634_cast, var_10597_cast))[name = tensor<string, []>("op_10904_cast")];
            tensor<fp16, []> var_10905_to_fp16 = const()[name = tensor<string, []>("op_10905_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_951_cast = mul(x = var_10904_cast, y = var_10905_to_fp16)[name = tensor<string, []>("aw_chunk_951_cast")];
            tensor<string, []> var_10908_equation_0 = const()[name = tensor<string, []>("op_10908_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10908_cast = einsum(equation = var_10908_equation_0, values = (var_10634_cast, var_10598_cast))[name = tensor<string, []>("op_10908_cast")];
            tensor<fp16, []> var_10909_to_fp16 = const()[name = tensor<string, []>("op_10909_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_953_cast = mul(x = var_10908_cast, y = var_10909_to_fp16)[name = tensor<string, []>("aw_chunk_953_cast")];
            tensor<string, []> var_10912_equation_0 = const()[name = tensor<string, []>("op_10912_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10912_cast = einsum(equation = var_10912_equation_0, values = (var_10634_cast, var_10599_cast))[name = tensor<string, []>("op_10912_cast")];
            tensor<fp16, []> var_10913_to_fp16 = const()[name = tensor<string, []>("op_10913_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_955_cast = mul(x = var_10912_cast, y = var_10913_to_fp16)[name = tensor<string, []>("aw_chunk_955_cast")];
            tensor<string, []> var_10916_equation_0 = const()[name = tensor<string, []>("op_10916_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10916_cast = einsum(equation = var_10916_equation_0, values = (var_10634_cast, var_10600_cast))[name = tensor<string, []>("op_10916_cast")];
            tensor<fp16, []> var_10917_to_fp16 = const()[name = tensor<string, []>("op_10917_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_957_cast = mul(x = var_10916_cast, y = var_10917_to_fp16)[name = tensor<string, []>("aw_chunk_957_cast")];
            tensor<string, []> var_10920_equation_0 = const()[name = tensor<string, []>("op_10920_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_10920_cast = einsum(equation = var_10920_equation_0, values = (var_10634_cast, var_10601_cast))[name = tensor<string, []>("op_10920_cast")];
            tensor<fp16, []> var_10921_to_fp16 = const()[name = tensor<string, []>("op_10921_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_959_cast = mul(x = var_10920_cast, y = var_10921_to_fp16)[name = tensor<string, []>("aw_chunk_959_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10923_cast = softmax(axis = var_10375, x = aw_chunk_833_cast)[name = tensor<string, []>("op_10923_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10924_cast = softmax(axis = var_10375, x = aw_chunk_835_cast)[name = tensor<string, []>("op_10924_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10925_cast = softmax(axis = var_10375, x = aw_chunk_837_cast)[name = tensor<string, []>("op_10925_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10926_cast = softmax(axis = var_10375, x = aw_chunk_839_cast)[name = tensor<string, []>("op_10926_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10927_cast = softmax(axis = var_10375, x = aw_chunk_841_cast)[name = tensor<string, []>("op_10927_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10928_cast = softmax(axis = var_10375, x = aw_chunk_843_cast)[name = tensor<string, []>("op_10928_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10929_cast = softmax(axis = var_10375, x = aw_chunk_845_cast)[name = tensor<string, []>("op_10929_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10930_cast = softmax(axis = var_10375, x = aw_chunk_847_cast)[name = tensor<string, []>("op_10930_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10931_cast = softmax(axis = var_10375, x = aw_chunk_849_cast)[name = tensor<string, []>("op_10931_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10932_cast = softmax(axis = var_10375, x = aw_chunk_851_cast)[name = tensor<string, []>("op_10932_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10933_cast = softmax(axis = var_10375, x = aw_chunk_853_cast)[name = tensor<string, []>("op_10933_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10934_cast = softmax(axis = var_10375, x = aw_chunk_855_cast)[name = tensor<string, []>("op_10934_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10935_cast = softmax(axis = var_10375, x = aw_chunk_857_cast)[name = tensor<string, []>("op_10935_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10936_cast = softmax(axis = var_10375, x = aw_chunk_859_cast)[name = tensor<string, []>("op_10936_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10937_cast = softmax(axis = var_10375, x = aw_chunk_861_cast)[name = tensor<string, []>("op_10937_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10938_cast = softmax(axis = var_10375, x = aw_chunk_863_cast)[name = tensor<string, []>("op_10938_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10939_cast = softmax(axis = var_10375, x = aw_chunk_865_cast)[name = tensor<string, []>("op_10939_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10940_cast = softmax(axis = var_10375, x = aw_chunk_867_cast)[name = tensor<string, []>("op_10940_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10941_cast = softmax(axis = var_10375, x = aw_chunk_869_cast)[name = tensor<string, []>("op_10941_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10942_cast = softmax(axis = var_10375, x = aw_chunk_871_cast)[name = tensor<string, []>("op_10942_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10943_cast = softmax(axis = var_10375, x = aw_chunk_873_cast)[name = tensor<string, []>("op_10943_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10944_cast = softmax(axis = var_10375, x = aw_chunk_875_cast)[name = tensor<string, []>("op_10944_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10945_cast = softmax(axis = var_10375, x = aw_chunk_877_cast)[name = tensor<string, []>("op_10945_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10946_cast = softmax(axis = var_10375, x = aw_chunk_879_cast)[name = tensor<string, []>("op_10946_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10947_cast = softmax(axis = var_10375, x = aw_chunk_881_cast)[name = tensor<string, []>("op_10947_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10948_cast = softmax(axis = var_10375, x = aw_chunk_883_cast)[name = tensor<string, []>("op_10948_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10949_cast = softmax(axis = var_10375, x = aw_chunk_885_cast)[name = tensor<string, []>("op_10949_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10950_cast = softmax(axis = var_10375, x = aw_chunk_887_cast)[name = tensor<string, []>("op_10950_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10951_cast = softmax(axis = var_10375, x = aw_chunk_889_cast)[name = tensor<string, []>("op_10951_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10952_cast = softmax(axis = var_10375, x = aw_chunk_891_cast)[name = tensor<string, []>("op_10952_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10953_cast = softmax(axis = var_10375, x = aw_chunk_893_cast)[name = tensor<string, []>("op_10953_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10954_cast = softmax(axis = var_10375, x = aw_chunk_895_cast)[name = tensor<string, []>("op_10954_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10955_cast = softmax(axis = var_10375, x = aw_chunk_897_cast)[name = tensor<string, []>("op_10955_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10956_cast = softmax(axis = var_10375, x = aw_chunk_899_cast)[name = tensor<string, []>("op_10956_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10957_cast = softmax(axis = var_10375, x = aw_chunk_901_cast)[name = tensor<string, []>("op_10957_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10958_cast = softmax(axis = var_10375, x = aw_chunk_903_cast)[name = tensor<string, []>("op_10958_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10959_cast = softmax(axis = var_10375, x = aw_chunk_905_cast)[name = tensor<string, []>("op_10959_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10960_cast = softmax(axis = var_10375, x = aw_chunk_907_cast)[name = tensor<string, []>("op_10960_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10961_cast = softmax(axis = var_10375, x = aw_chunk_909_cast)[name = tensor<string, []>("op_10961_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10962_cast = softmax(axis = var_10375, x = aw_chunk_911_cast)[name = tensor<string, []>("op_10962_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10963_cast = softmax(axis = var_10375, x = aw_chunk_913_cast)[name = tensor<string, []>("op_10963_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10964_cast = softmax(axis = var_10375, x = aw_chunk_915_cast)[name = tensor<string, []>("op_10964_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10965_cast = softmax(axis = var_10375, x = aw_chunk_917_cast)[name = tensor<string, []>("op_10965_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10966_cast = softmax(axis = var_10375, x = aw_chunk_919_cast)[name = tensor<string, []>("op_10966_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10967_cast = softmax(axis = var_10375, x = aw_chunk_921_cast)[name = tensor<string, []>("op_10967_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10968_cast = softmax(axis = var_10375, x = aw_chunk_923_cast)[name = tensor<string, []>("op_10968_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10969_cast = softmax(axis = var_10375, x = aw_chunk_925_cast)[name = tensor<string, []>("op_10969_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10970_cast = softmax(axis = var_10375, x = aw_chunk_927_cast)[name = tensor<string, []>("op_10970_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10971_cast = softmax(axis = var_10375, x = aw_chunk_929_cast)[name = tensor<string, []>("op_10971_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10972_cast = softmax(axis = var_10375, x = aw_chunk_931_cast)[name = tensor<string, []>("op_10972_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10973_cast = softmax(axis = var_10375, x = aw_chunk_933_cast)[name = tensor<string, []>("op_10973_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10974_cast = softmax(axis = var_10375, x = aw_chunk_935_cast)[name = tensor<string, []>("op_10974_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10975_cast = softmax(axis = var_10375, x = aw_chunk_937_cast)[name = tensor<string, []>("op_10975_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10976_cast = softmax(axis = var_10375, x = aw_chunk_939_cast)[name = tensor<string, []>("op_10976_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10977_cast = softmax(axis = var_10375, x = aw_chunk_941_cast)[name = tensor<string, []>("op_10977_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10978_cast = softmax(axis = var_10375, x = aw_chunk_943_cast)[name = tensor<string, []>("op_10978_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10979_cast = softmax(axis = var_10375, x = aw_chunk_945_cast)[name = tensor<string, []>("op_10979_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10980_cast = softmax(axis = var_10375, x = aw_chunk_947_cast)[name = tensor<string, []>("op_10980_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10981_cast = softmax(axis = var_10375, x = aw_chunk_949_cast)[name = tensor<string, []>("op_10981_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10982_cast = softmax(axis = var_10375, x = aw_chunk_951_cast)[name = tensor<string, []>("op_10982_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10983_cast = softmax(axis = var_10375, x = aw_chunk_953_cast)[name = tensor<string, []>("op_10983_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10984_cast = softmax(axis = var_10375, x = aw_chunk_955_cast)[name = tensor<string, []>("op_10984_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10985_cast = softmax(axis = var_10375, x = aw_chunk_957_cast)[name = tensor<string, []>("op_10985_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_10986_cast = softmax(axis = var_10375, x = aw_chunk_959_cast)[name = tensor<string, []>("op_10986_cast")];
            tensor<string, []> var_10988_equation_0 = const()[name = tensor<string, []>("op_10988_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_10988_cast = einsum(equation = var_10988_equation_0, values = (var_10636_cast, var_10923_cast))[name = tensor<string, []>("op_10988_cast")];
            tensor<string, []> var_10990_equation_0 = const()[name = tensor<string, []>("op_10990_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_10990_cast = einsum(equation = var_10990_equation_0, values = (var_10636_cast, var_10924_cast))[name = tensor<string, []>("op_10990_cast")];
            tensor<string, []> var_10992_equation_0 = const()[name = tensor<string, []>("op_10992_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_10992_cast = einsum(equation = var_10992_equation_0, values = (var_10636_cast, var_10925_cast))[name = tensor<string, []>("op_10992_cast")];
            tensor<string, []> var_10994_equation_0 = const()[name = tensor<string, []>("op_10994_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_10994_cast = einsum(equation = var_10994_equation_0, values = (var_10636_cast, var_10926_cast))[name = tensor<string, []>("op_10994_cast")];
            tensor<string, []> var_10996_equation_0 = const()[name = tensor<string, []>("op_10996_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_10996_cast = einsum(equation = var_10996_equation_0, values = (var_10636_cast, var_10927_cast))[name = tensor<string, []>("op_10996_cast")];
            tensor<string, []> var_10998_equation_0 = const()[name = tensor<string, []>("op_10998_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_10998_cast = einsum(equation = var_10998_equation_0, values = (var_10636_cast, var_10928_cast))[name = tensor<string, []>("op_10998_cast")];
            tensor<string, []> var_11000_equation_0 = const()[name = tensor<string, []>("op_11000_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11000_cast = einsum(equation = var_11000_equation_0, values = (var_10636_cast, var_10929_cast))[name = tensor<string, []>("op_11000_cast")];
            tensor<string, []> var_11002_equation_0 = const()[name = tensor<string, []>("op_11002_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11002_cast = einsum(equation = var_11002_equation_0, values = (var_10636_cast, var_10930_cast))[name = tensor<string, []>("op_11002_cast")];
            tensor<string, []> var_11004_equation_0 = const()[name = tensor<string, []>("op_11004_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11004_cast = einsum(equation = var_11004_equation_0, values = (var_10640_cast, var_10931_cast))[name = tensor<string, []>("op_11004_cast")];
            tensor<string, []> var_11006_equation_0 = const()[name = tensor<string, []>("op_11006_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11006_cast = einsum(equation = var_11006_equation_0, values = (var_10640_cast, var_10932_cast))[name = tensor<string, []>("op_11006_cast")];
            tensor<string, []> var_11008_equation_0 = const()[name = tensor<string, []>("op_11008_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11008_cast = einsum(equation = var_11008_equation_0, values = (var_10640_cast, var_10933_cast))[name = tensor<string, []>("op_11008_cast")];
            tensor<string, []> var_11010_equation_0 = const()[name = tensor<string, []>("op_11010_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11010_cast = einsum(equation = var_11010_equation_0, values = (var_10640_cast, var_10934_cast))[name = tensor<string, []>("op_11010_cast")];
            tensor<string, []> var_11012_equation_0 = const()[name = tensor<string, []>("op_11012_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11012_cast = einsum(equation = var_11012_equation_0, values = (var_10640_cast, var_10935_cast))[name = tensor<string, []>("op_11012_cast")];
            tensor<string, []> var_11014_equation_0 = const()[name = tensor<string, []>("op_11014_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11014_cast = einsum(equation = var_11014_equation_0, values = (var_10640_cast, var_10936_cast))[name = tensor<string, []>("op_11014_cast")];
            tensor<string, []> var_11016_equation_0 = const()[name = tensor<string, []>("op_11016_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11016_cast = einsum(equation = var_11016_equation_0, values = (var_10640_cast, var_10937_cast))[name = tensor<string, []>("op_11016_cast")];
            tensor<string, []> var_11018_equation_0 = const()[name = tensor<string, []>("op_11018_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11018_cast = einsum(equation = var_11018_equation_0, values = (var_10640_cast, var_10938_cast))[name = tensor<string, []>("op_11018_cast")];
            tensor<string, []> var_11020_equation_0 = const()[name = tensor<string, []>("op_11020_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11020_cast = einsum(equation = var_11020_equation_0, values = (var_10644_cast, var_10939_cast))[name = tensor<string, []>("op_11020_cast")];
            tensor<string, []> var_11022_equation_0 = const()[name = tensor<string, []>("op_11022_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11022_cast = einsum(equation = var_11022_equation_0, values = (var_10644_cast, var_10940_cast))[name = tensor<string, []>("op_11022_cast")];
            tensor<string, []> var_11024_equation_0 = const()[name = tensor<string, []>("op_11024_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11024_cast = einsum(equation = var_11024_equation_0, values = (var_10644_cast, var_10941_cast))[name = tensor<string, []>("op_11024_cast")];
            tensor<string, []> var_11026_equation_0 = const()[name = tensor<string, []>("op_11026_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11026_cast = einsum(equation = var_11026_equation_0, values = (var_10644_cast, var_10942_cast))[name = tensor<string, []>("op_11026_cast")];
            tensor<string, []> var_11028_equation_0 = const()[name = tensor<string, []>("op_11028_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11028_cast = einsum(equation = var_11028_equation_0, values = (var_10644_cast, var_10943_cast))[name = tensor<string, []>("op_11028_cast")];
            tensor<string, []> var_11030_equation_0 = const()[name = tensor<string, []>("op_11030_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11030_cast = einsum(equation = var_11030_equation_0, values = (var_10644_cast, var_10944_cast))[name = tensor<string, []>("op_11030_cast")];
            tensor<string, []> var_11032_equation_0 = const()[name = tensor<string, []>("op_11032_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11032_cast = einsum(equation = var_11032_equation_0, values = (var_10644_cast, var_10945_cast))[name = tensor<string, []>("op_11032_cast")];
            tensor<string, []> var_11034_equation_0 = const()[name = tensor<string, []>("op_11034_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11034_cast = einsum(equation = var_11034_equation_0, values = (var_10644_cast, var_10946_cast))[name = tensor<string, []>("op_11034_cast")];
            tensor<string, []> var_11036_equation_0 = const()[name = tensor<string, []>("op_11036_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11036_cast = einsum(equation = var_11036_equation_0, values = (var_10648_cast, var_10947_cast))[name = tensor<string, []>("op_11036_cast")];
            tensor<string, []> var_11038_equation_0 = const()[name = tensor<string, []>("op_11038_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11038_cast = einsum(equation = var_11038_equation_0, values = (var_10648_cast, var_10948_cast))[name = tensor<string, []>("op_11038_cast")];
            tensor<string, []> var_11040_equation_0 = const()[name = tensor<string, []>("op_11040_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11040_cast = einsum(equation = var_11040_equation_0, values = (var_10648_cast, var_10949_cast))[name = tensor<string, []>("op_11040_cast")];
            tensor<string, []> var_11042_equation_0 = const()[name = tensor<string, []>("op_11042_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11042_cast = einsum(equation = var_11042_equation_0, values = (var_10648_cast, var_10950_cast))[name = tensor<string, []>("op_11042_cast")];
            tensor<string, []> var_11044_equation_0 = const()[name = tensor<string, []>("op_11044_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11044_cast = einsum(equation = var_11044_equation_0, values = (var_10648_cast, var_10951_cast))[name = tensor<string, []>("op_11044_cast")];
            tensor<string, []> var_11046_equation_0 = const()[name = tensor<string, []>("op_11046_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11046_cast = einsum(equation = var_11046_equation_0, values = (var_10648_cast, var_10952_cast))[name = tensor<string, []>("op_11046_cast")];
            tensor<string, []> var_11048_equation_0 = const()[name = tensor<string, []>("op_11048_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11048_cast = einsum(equation = var_11048_equation_0, values = (var_10648_cast, var_10953_cast))[name = tensor<string, []>("op_11048_cast")];
            tensor<string, []> var_11050_equation_0 = const()[name = tensor<string, []>("op_11050_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11050_cast = einsum(equation = var_11050_equation_0, values = (var_10648_cast, var_10954_cast))[name = tensor<string, []>("op_11050_cast")];
            tensor<string, []> var_11052_equation_0 = const()[name = tensor<string, []>("op_11052_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11052_cast = einsum(equation = var_11052_equation_0, values = (var_10652_cast, var_10955_cast))[name = tensor<string, []>("op_11052_cast")];
            tensor<string, []> var_11054_equation_0 = const()[name = tensor<string, []>("op_11054_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11054_cast = einsum(equation = var_11054_equation_0, values = (var_10652_cast, var_10956_cast))[name = tensor<string, []>("op_11054_cast")];
            tensor<string, []> var_11056_equation_0 = const()[name = tensor<string, []>("op_11056_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11056_cast = einsum(equation = var_11056_equation_0, values = (var_10652_cast, var_10957_cast))[name = tensor<string, []>("op_11056_cast")];
            tensor<string, []> var_11058_equation_0 = const()[name = tensor<string, []>("op_11058_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11058_cast = einsum(equation = var_11058_equation_0, values = (var_10652_cast, var_10958_cast))[name = tensor<string, []>("op_11058_cast")];
            tensor<string, []> var_11060_equation_0 = const()[name = tensor<string, []>("op_11060_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11060_cast = einsum(equation = var_11060_equation_0, values = (var_10652_cast, var_10959_cast))[name = tensor<string, []>("op_11060_cast")];
            tensor<string, []> var_11062_equation_0 = const()[name = tensor<string, []>("op_11062_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11062_cast = einsum(equation = var_11062_equation_0, values = (var_10652_cast, var_10960_cast))[name = tensor<string, []>("op_11062_cast")];
            tensor<string, []> var_11064_equation_0 = const()[name = tensor<string, []>("op_11064_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11064_cast = einsum(equation = var_11064_equation_0, values = (var_10652_cast, var_10961_cast))[name = tensor<string, []>("op_11064_cast")];
            tensor<string, []> var_11066_equation_0 = const()[name = tensor<string, []>("op_11066_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11066_cast = einsum(equation = var_11066_equation_0, values = (var_10652_cast, var_10962_cast))[name = tensor<string, []>("op_11066_cast")];
            tensor<string, []> var_11068_equation_0 = const()[name = tensor<string, []>("op_11068_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11068_cast = einsum(equation = var_11068_equation_0, values = (var_10656_cast, var_10963_cast))[name = tensor<string, []>("op_11068_cast")];
            tensor<string, []> var_11070_equation_0 = const()[name = tensor<string, []>("op_11070_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11070_cast = einsum(equation = var_11070_equation_0, values = (var_10656_cast, var_10964_cast))[name = tensor<string, []>("op_11070_cast")];
            tensor<string, []> var_11072_equation_0 = const()[name = tensor<string, []>("op_11072_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11072_cast = einsum(equation = var_11072_equation_0, values = (var_10656_cast, var_10965_cast))[name = tensor<string, []>("op_11072_cast")];
            tensor<string, []> var_11074_equation_0 = const()[name = tensor<string, []>("op_11074_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11074_cast = einsum(equation = var_11074_equation_0, values = (var_10656_cast, var_10966_cast))[name = tensor<string, []>("op_11074_cast")];
            tensor<string, []> var_11076_equation_0 = const()[name = tensor<string, []>("op_11076_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11076_cast = einsum(equation = var_11076_equation_0, values = (var_10656_cast, var_10967_cast))[name = tensor<string, []>("op_11076_cast")];
            tensor<string, []> var_11078_equation_0 = const()[name = tensor<string, []>("op_11078_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11078_cast = einsum(equation = var_11078_equation_0, values = (var_10656_cast, var_10968_cast))[name = tensor<string, []>("op_11078_cast")];
            tensor<string, []> var_11080_equation_0 = const()[name = tensor<string, []>("op_11080_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11080_cast = einsum(equation = var_11080_equation_0, values = (var_10656_cast, var_10969_cast))[name = tensor<string, []>("op_11080_cast")];
            tensor<string, []> var_11082_equation_0 = const()[name = tensor<string, []>("op_11082_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11082_cast = einsum(equation = var_11082_equation_0, values = (var_10656_cast, var_10970_cast))[name = tensor<string, []>("op_11082_cast")];
            tensor<string, []> var_11084_equation_0 = const()[name = tensor<string, []>("op_11084_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11084_cast = einsum(equation = var_11084_equation_0, values = (var_10660_cast, var_10971_cast))[name = tensor<string, []>("op_11084_cast")];
            tensor<string, []> var_11086_equation_0 = const()[name = tensor<string, []>("op_11086_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11086_cast = einsum(equation = var_11086_equation_0, values = (var_10660_cast, var_10972_cast))[name = tensor<string, []>("op_11086_cast")];
            tensor<string, []> var_11088_equation_0 = const()[name = tensor<string, []>("op_11088_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11088_cast = einsum(equation = var_11088_equation_0, values = (var_10660_cast, var_10973_cast))[name = tensor<string, []>("op_11088_cast")];
            tensor<string, []> var_11090_equation_0 = const()[name = tensor<string, []>("op_11090_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11090_cast = einsum(equation = var_11090_equation_0, values = (var_10660_cast, var_10974_cast))[name = tensor<string, []>("op_11090_cast")];
            tensor<string, []> var_11092_equation_0 = const()[name = tensor<string, []>("op_11092_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11092_cast = einsum(equation = var_11092_equation_0, values = (var_10660_cast, var_10975_cast))[name = tensor<string, []>("op_11092_cast")];
            tensor<string, []> var_11094_equation_0 = const()[name = tensor<string, []>("op_11094_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11094_cast = einsum(equation = var_11094_equation_0, values = (var_10660_cast, var_10976_cast))[name = tensor<string, []>("op_11094_cast")];
            tensor<string, []> var_11096_equation_0 = const()[name = tensor<string, []>("op_11096_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11096_cast = einsum(equation = var_11096_equation_0, values = (var_10660_cast, var_10977_cast))[name = tensor<string, []>("op_11096_cast")];
            tensor<string, []> var_11098_equation_0 = const()[name = tensor<string, []>("op_11098_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11098_cast = einsum(equation = var_11098_equation_0, values = (var_10660_cast, var_10978_cast))[name = tensor<string, []>("op_11098_cast")];
            tensor<string, []> var_11100_equation_0 = const()[name = tensor<string, []>("op_11100_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11100_cast = einsum(equation = var_11100_equation_0, values = (var_10664_cast, var_10979_cast))[name = tensor<string, []>("op_11100_cast")];
            tensor<string, []> var_11102_equation_0 = const()[name = tensor<string, []>("op_11102_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11102_cast = einsum(equation = var_11102_equation_0, values = (var_10664_cast, var_10980_cast))[name = tensor<string, []>("op_11102_cast")];
            tensor<string, []> var_11104_equation_0 = const()[name = tensor<string, []>("op_11104_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11104_cast = einsum(equation = var_11104_equation_0, values = (var_10664_cast, var_10981_cast))[name = tensor<string, []>("op_11104_cast")];
            tensor<string, []> var_11106_equation_0 = const()[name = tensor<string, []>("op_11106_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11106_cast = einsum(equation = var_11106_equation_0, values = (var_10664_cast, var_10982_cast))[name = tensor<string, []>("op_11106_cast")];
            tensor<string, []> var_11108_equation_0 = const()[name = tensor<string, []>("op_11108_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11108_cast = einsum(equation = var_11108_equation_0, values = (var_10664_cast, var_10983_cast))[name = tensor<string, []>("op_11108_cast")];
            tensor<string, []> var_11110_equation_0 = const()[name = tensor<string, []>("op_11110_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11110_cast = einsum(equation = var_11110_equation_0, values = (var_10664_cast, var_10984_cast))[name = tensor<string, []>("op_11110_cast")];
            tensor<string, []> var_11112_equation_0 = const()[name = tensor<string, []>("op_11112_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11112_cast = einsum(equation = var_11112_equation_0, values = (var_10664_cast, var_10985_cast))[name = tensor<string, []>("op_11112_cast")];
            tensor<string, []> var_11114_equation_0 = const()[name = tensor<string, []>("op_11114_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11114_cast = einsum(equation = var_11114_equation_0, values = (var_10664_cast, var_10986_cast))[name = tensor<string, []>("op_11114_cast")];
            tensor<bool, []> var_11116_interleave_0 = const()[name = tensor<string, []>("op_11116_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_11116_cast = concat(axis = var_10347, interleave = var_11116_interleave_0, values = (var_10988_cast, var_10990_cast, var_10992_cast, var_10994_cast, var_10996_cast, var_10998_cast, var_11000_cast, var_11002_cast))[name = tensor<string, []>("op_11116_cast")];
            tensor<bool, []> var_11118_interleave_0 = const()[name = tensor<string, []>("op_11118_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_11118_cast = concat(axis = var_10347, interleave = var_11118_interleave_0, values = (var_11004_cast, var_11006_cast, var_11008_cast, var_11010_cast, var_11012_cast, var_11014_cast, var_11016_cast, var_11018_cast))[name = tensor<string, []>("op_11118_cast")];
            tensor<bool, []> var_11120_interleave_0 = const()[name = tensor<string, []>("op_11120_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_11120_cast = concat(axis = var_10347, interleave = var_11120_interleave_0, values = (var_11020_cast, var_11022_cast, var_11024_cast, var_11026_cast, var_11028_cast, var_11030_cast, var_11032_cast, var_11034_cast))[name = tensor<string, []>("op_11120_cast")];
            tensor<bool, []> var_11122_interleave_0 = const()[name = tensor<string, []>("op_11122_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_11122_cast = concat(axis = var_10347, interleave = var_11122_interleave_0, values = (var_11036_cast, var_11038_cast, var_11040_cast, var_11042_cast, var_11044_cast, var_11046_cast, var_11048_cast, var_11050_cast))[name = tensor<string, []>("op_11122_cast")];
            tensor<bool, []> var_11124_interleave_0 = const()[name = tensor<string, []>("op_11124_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_11124_cast = concat(axis = var_10347, interleave = var_11124_interleave_0, values = (var_11052_cast, var_11054_cast, var_11056_cast, var_11058_cast, var_11060_cast, var_11062_cast, var_11064_cast, var_11066_cast))[name = tensor<string, []>("op_11124_cast")];
            tensor<bool, []> var_11126_interleave_0 = const()[name = tensor<string, []>("op_11126_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_11126_cast = concat(axis = var_10347, interleave = var_11126_interleave_0, values = (var_11068_cast, var_11070_cast, var_11072_cast, var_11074_cast, var_11076_cast, var_11078_cast, var_11080_cast, var_11082_cast))[name = tensor<string, []>("op_11126_cast")];
            tensor<bool, []> var_11128_interleave_0 = const()[name = tensor<string, []>("op_11128_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_11128_cast = concat(axis = var_10347, interleave = var_11128_interleave_0, values = (var_11084_cast, var_11086_cast, var_11088_cast, var_11090_cast, var_11092_cast, var_11094_cast, var_11096_cast, var_11098_cast))[name = tensor<string, []>("op_11128_cast")];
            tensor<bool, []> var_11130_interleave_0 = const()[name = tensor<string, []>("op_11130_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_11130_cast = concat(axis = var_10347, interleave = var_11130_interleave_0, values = (var_11100_cast, var_11102_cast, var_11104_cast, var_11106_cast, var_11108_cast, var_11110_cast, var_11112_cast, var_11114_cast))[name = tensor<string, []>("op_11130_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_10375, interleave = input_465_interleave_0, values = (var_11116_cast, var_11118_cast, var_11120_cast, var_11122_cast, var_11124_cast, var_11126_cast, var_11128_cast, var_11130_cast))[name = tensor<string, []>("input_465_cast")];
            tensor<int32, [2]> var_11136 = const()[name = tensor<string, []>("op_11136"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11138 = const()[name = tensor<string, []>("op_11138"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_11140_pad_type_0 = const()[name = tensor<string, []>("op_11140_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_11140_pad_0 = const()[name = tensor<string, []>("op_11140_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(634627264))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(634704128))), name = tensor<string, []>("up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            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, []>(634704320)))];
            tensor<fp16, [2, 320, 1, 4096]> var_11140_cast = conv(bias = up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_11138, groups = var_10375, pad = var_11140_pad_0, pad_type = var_11140_pad_type_0, strides = var_11136, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_465_cast)[name = tensor<string, []>("op_11140_cast")];
            tensor<fp16, [2, 320, 1, 4096]> inputs_81_cast = add(x = var_11140_cast, y = inputs_79_cast)[name = tensor<string, []>("inputs_81_cast")];
            tensor<int32, [1]> var_11144 = const()[name = tensor<string, []>("op_11144"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_81_cast = reduce_mean(axes = var_11144, keep_dims = var_10370, x = inputs_81_cast)[name = tensor<string, []>("channels_mean_81_cast")];
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_81_cast = sub(x = inputs_81_cast, y = channels_mean_81_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_81_cast")];
            tensor<int32, [1]> var_11148 = const()[name = tensor<string, []>("op_11148"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_11149_cast = reduce_mean(axes = var_11148, keep_dims = var_10370, x = zero_mean_sq_81_cast)[name = tensor<string, []>("op_11149_cast")];
            tensor<fp16, []> var_11150_to_fp16 = const()[name = tensor<string, []>("op_11150_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_11151_cast = add(x = var_11149_cast, y = var_11150_to_fp16)[name = tensor<string, []>("op_11151_cast")];
            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_11151_cast)[name = tensor<string, []>("denom_81_cast")];
            tensor<fp16, [2, 320, 1, 4096]> out_81_cast = mul(x = zero_mean_81_cast, y = denom_81_cast)[name = tensor<string, []>("out_81_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_11155_to_fp16 = const()[name = tensor<string, []>("op_11155_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(634705024)))];
            tensor<fp16, [2, 320, 1, 4096]> var_11156_cast = add(x = out_81_cast, y = var_11155_to_fp16)[name = tensor<string, []>("op_11156_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_11158_to_fp16 = const()[name = tensor<string, []>("op_11158_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(634705728)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_289_cast = mul(x = var_11156_cast, y = var_11158_to_fp16)[name = tensor<string, []>("hidden_states_289_cast")];
            tensor<int32, [2]> var_11165 = const()[name = tensor<string, []>("op_11165"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11167 = const()[name = tensor<string, []>("op_11167"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(634706432))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(634783296))), name = tensor<string, []>("up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            tensor<fp16, [2, 320, 1, 4096]> q_55_cast = conv(dilations = var_11167, groups = var_10375, pad = q_55_pad_0, pad_type = q_55_pad_type_0, strides = var_11165, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_289_cast)[name = tensor<string, []>("q_55_cast")];
            tensor<int32, [2]> var_11171 = const()[name = tensor<string, []>("op_11171"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11173 = const()[name = tensor<string, []>("op_11173"), 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, 768, 1, 1]> up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [184320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(634783488))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(634967872))), name = tensor<string, []>("up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 768, 1, 1])];
            tensor<fp16, [2, 320, 1, 77]> k_109_cast = conv(dilations = var_11173, groups = var_10375, pad = k_109_pad_0, pad_type = k_109_pad_type_0, strides = var_11171, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_109_cast")];
            tensor<int32, [2]> var_11177 = const()[name = tensor<string, []>("op_11177"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11179 = const()[name = tensor<string, []>("op_11179"), 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, 768, 1, 1]> up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [184320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(634968064))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(635152448))), name = tensor<string, []>("up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 768, 1, 1])];
            tensor<fp16, [2, 320, 1, 77]> v_55_cast = conv(dilations = var_11179, groups = var_10375, pad = v_55_pad_0, pad_type = v_55_pad_type_0, strides = var_11177, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_55_cast")];
            tensor<int32, [4]> var_11183_begin_0 = const()[name = tensor<string, []>("op_11183_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11183_end_0 = const()[name = tensor<string, []>("op_11183_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_11183_end_mask_0 = const()[name = tensor<string, []>("op_11183_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_11183_cast = slice_by_index(begin = var_11183_begin_0, end = var_11183_end_0, end_mask = var_11183_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_11183_cast")];
            tensor<int32, [4]> var_11187_begin_0 = const()[name = tensor<string, []>("op_11187_begin_0"), val = tensor<int32, [4]>([0, 40, 0, 0])];
            tensor<int32, [4]> var_11187_end_0 = const()[name = tensor<string, []>("op_11187_end_0"), val = tensor<int32, [4]>([2, 80, 1, 4096])];
            tensor<bool, [4]> var_11187_end_mask_0 = const()[name = tensor<string, []>("op_11187_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_11187_cast = slice_by_index(begin = var_11187_begin_0, end = var_11187_end_0, end_mask = var_11187_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_11187_cast")];
            tensor<int32, [4]> var_11191_begin_0 = const()[name = tensor<string, []>("op_11191_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_11191_end_0 = const()[name = tensor<string, []>("op_11191_end_0"), val = tensor<int32, [4]>([2, 120, 1, 4096])];
            tensor<bool, [4]> var_11191_end_mask_0 = const()[name = tensor<string, []>("op_11191_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_11191_cast = slice_by_index(begin = var_11191_begin_0, end = var_11191_end_0, end_mask = var_11191_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_11191_cast")];
            tensor<int32, [4]> var_11195_begin_0 = const()[name = tensor<string, []>("op_11195_begin_0"), val = tensor<int32, [4]>([0, 120, 0, 0])];
            tensor<int32, [4]> var_11195_end_0 = const()[name = tensor<string, []>("op_11195_end_0"), val = tensor<int32, [4]>([2, 160, 1, 4096])];
            tensor<bool, [4]> var_11195_end_mask_0 = const()[name = tensor<string, []>("op_11195_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_11195_cast = slice_by_index(begin = var_11195_begin_0, end = var_11195_end_0, end_mask = var_11195_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_11195_cast")];
            tensor<int32, [4]> var_11199_begin_0 = const()[name = tensor<string, []>("op_11199_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_11199_end_0 = const()[name = tensor<string, []>("op_11199_end_0"), val = tensor<int32, [4]>([2, 200, 1, 4096])];
            tensor<bool, [4]> var_11199_end_mask_0 = const()[name = tensor<string, []>("op_11199_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_11199_cast = slice_by_index(begin = var_11199_begin_0, end = var_11199_end_0, end_mask = var_11199_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_11199_cast")];
            tensor<int32, [4]> var_11203_begin_0 = const()[name = tensor<string, []>("op_11203_begin_0"), val = tensor<int32, [4]>([0, 200, 0, 0])];
            tensor<int32, [4]> var_11203_end_0 = const()[name = tensor<string, []>("op_11203_end_0"), val = tensor<int32, [4]>([2, 240, 1, 4096])];
            tensor<bool, [4]> var_11203_end_mask_0 = const()[name = tensor<string, []>("op_11203_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_11203_cast = slice_by_index(begin = var_11203_begin_0, end = var_11203_end_0, end_mask = var_11203_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_11203_cast")];
            tensor<int32, [4]> var_11207_begin_0 = const()[name = tensor<string, []>("op_11207_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_11207_end_0 = const()[name = tensor<string, []>("op_11207_end_0"), val = tensor<int32, [4]>([2, 280, 1, 4096])];
            tensor<bool, [4]> var_11207_end_mask_0 = const()[name = tensor<string, []>("op_11207_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_11207_cast = slice_by_index(begin = var_11207_begin_0, end = var_11207_end_0, end_mask = var_11207_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_11207_cast")];
            tensor<int32, [4]> var_11211_begin_0 = const()[name = tensor<string, []>("op_11211_begin_0"), val = tensor<int32, [4]>([0, 280, 0, 0])];
            tensor<int32, [4]> var_11211_end_0 = const()[name = tensor<string, []>("op_11211_end_0"), val = tensor<int32, [4]>([2, 320, 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, 40, 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_55_cast)[name = tensor<string, []>("op_11211_cast")];
            tensor<int32, [4]> var_11214_begin_0 = const()[name = tensor<string, []>("op_11214_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11214_end_0 = const()[name = tensor<string, []>("op_11214_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_11214_end_mask_0 = const()[name = tensor<string, []>("op_11214_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11214_cast = slice_by_index(begin = var_11214_begin_0, end = var_11214_end_0, end_mask = var_11214_end_mask_0, x = var_11183_cast)[name = tensor<string, []>("op_11214_cast")];
            tensor<int32, [4]> var_11215_begin_0 = const()[name = tensor<string, []>("op_11215_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_11215_end_0 = const()[name = tensor<string, []>("op_11215_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_11215_end_mask_0 = const()[name = tensor<string, []>("op_11215_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11215_cast = slice_by_index(begin = var_11215_begin_0, end = var_11215_end_0, end_mask = var_11215_end_mask_0, x = var_11183_cast)[name = tensor<string, []>("op_11215_cast")];
            tensor<int32, [4]> var_11216_begin_0 = const()[name = tensor<string, []>("op_11216_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_11216_end_0 = const()[name = tensor<string, []>("op_11216_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_11216_end_mask_0 = const()[name = tensor<string, []>("op_11216_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11216_cast = slice_by_index(begin = var_11216_begin_0, end = var_11216_end_0, end_mask = var_11216_end_mask_0, x = var_11183_cast)[name = tensor<string, []>("op_11216_cast")];
            tensor<int32, [4]> var_11217_begin_0 = const()[name = tensor<string, []>("op_11217_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_11217_end_0 = const()[name = tensor<string, []>("op_11217_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_11217_end_mask_0 = const()[name = tensor<string, []>("op_11217_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11217_cast = slice_by_index(begin = var_11217_begin_0, end = var_11217_end_0, end_mask = var_11217_end_mask_0, x = var_11183_cast)[name = tensor<string, []>("op_11217_cast")];
            tensor<int32, [4]> var_11218_begin_0 = const()[name = tensor<string, []>("op_11218_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_11218_end_0 = const()[name = tensor<string, []>("op_11218_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_11218_end_mask_0 = const()[name = tensor<string, []>("op_11218_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11218_cast = slice_by_index(begin = var_11218_begin_0, end = var_11218_end_0, end_mask = var_11218_end_mask_0, x = var_11183_cast)[name = tensor<string, []>("op_11218_cast")];
            tensor<int32, [4]> var_11219_begin_0 = const()[name = tensor<string, []>("op_11219_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_11219_end_0 = const()[name = tensor<string, []>("op_11219_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_11219_end_mask_0 = const()[name = tensor<string, []>("op_11219_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11219_cast = slice_by_index(begin = var_11219_begin_0, end = var_11219_end_0, end_mask = var_11219_end_mask_0, x = var_11183_cast)[name = tensor<string, []>("op_11219_cast")];
            tensor<int32, [4]> var_11220_begin_0 = const()[name = tensor<string, []>("op_11220_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_11220_end_0 = const()[name = tensor<string, []>("op_11220_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_11220_end_mask_0 = const()[name = tensor<string, []>("op_11220_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11220_cast = slice_by_index(begin = var_11220_begin_0, end = var_11220_end_0, end_mask = var_11220_end_mask_0, x = var_11183_cast)[name = tensor<string, []>("op_11220_cast")];
            tensor<int32, [4]> var_11221_begin_0 = const()[name = tensor<string, []>("op_11221_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_11221_end_0 = const()[name = tensor<string, []>("op_11221_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_11221_end_mask_0 = const()[name = tensor<string, []>("op_11221_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11221_cast = slice_by_index(begin = var_11221_begin_0, end = var_11221_end_0, end_mask = var_11221_end_mask_0, x = var_11183_cast)[name = tensor<string, []>("op_11221_cast")];
            tensor<int32, [4]> var_11222_begin_0 = const()[name = tensor<string, []>("op_11222_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11222_end_0 = const()[name = tensor<string, []>("op_11222_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_11222_end_mask_0 = const()[name = tensor<string, []>("op_11222_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11222_cast = slice_by_index(begin = var_11222_begin_0, end = var_11222_end_0, end_mask = var_11222_end_mask_0, x = var_11187_cast)[name = tensor<string, []>("op_11222_cast")];
            tensor<int32, [4]> var_11223_begin_0 = const()[name = tensor<string, []>("op_11223_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_11223_end_0 = const()[name = tensor<string, []>("op_11223_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_11223_end_mask_0 = const()[name = tensor<string, []>("op_11223_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11223_cast = slice_by_index(begin = var_11223_begin_0, end = var_11223_end_0, end_mask = var_11223_end_mask_0, x = var_11187_cast)[name = tensor<string, []>("op_11223_cast")];
            tensor<int32, [4]> var_11224_begin_0 = const()[name = tensor<string, []>("op_11224_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_11224_end_0 = const()[name = tensor<string, []>("op_11224_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_11224_end_mask_0 = const()[name = tensor<string, []>("op_11224_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11224_cast = slice_by_index(begin = var_11224_begin_0, end = var_11224_end_0, end_mask = var_11224_end_mask_0, x = var_11187_cast)[name = tensor<string, []>("op_11224_cast")];
            tensor<int32, [4]> var_11225_begin_0 = const()[name = tensor<string, []>("op_11225_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_11225_end_0 = const()[name = tensor<string, []>("op_11225_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_11225_end_mask_0 = const()[name = tensor<string, []>("op_11225_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11225_cast = slice_by_index(begin = var_11225_begin_0, end = var_11225_end_0, end_mask = var_11225_end_mask_0, x = var_11187_cast)[name = tensor<string, []>("op_11225_cast")];
            tensor<int32, [4]> var_11226_begin_0 = const()[name = tensor<string, []>("op_11226_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_11226_end_0 = const()[name = tensor<string, []>("op_11226_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_11226_end_mask_0 = const()[name = tensor<string, []>("op_11226_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11226_cast = slice_by_index(begin = var_11226_begin_0, end = var_11226_end_0, end_mask = var_11226_end_mask_0, x = var_11187_cast)[name = tensor<string, []>("op_11226_cast")];
            tensor<int32, [4]> var_11227_begin_0 = const()[name = tensor<string, []>("op_11227_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_11227_end_0 = const()[name = tensor<string, []>("op_11227_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_11227_end_mask_0 = const()[name = tensor<string, []>("op_11227_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11227_cast = slice_by_index(begin = var_11227_begin_0, end = var_11227_end_0, end_mask = var_11227_end_mask_0, x = var_11187_cast)[name = tensor<string, []>("op_11227_cast")];
            tensor<int32, [4]> var_11228_begin_0 = const()[name = tensor<string, []>("op_11228_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_11228_end_0 = const()[name = tensor<string, []>("op_11228_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_11228_end_mask_0 = const()[name = tensor<string, []>("op_11228_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11228_cast = slice_by_index(begin = var_11228_begin_0, end = var_11228_end_0, end_mask = var_11228_end_mask_0, x = var_11187_cast)[name = tensor<string, []>("op_11228_cast")];
            tensor<int32, [4]> var_11229_begin_0 = const()[name = tensor<string, []>("op_11229_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_11229_end_0 = const()[name = tensor<string, []>("op_11229_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_11229_end_mask_0 = const()[name = tensor<string, []>("op_11229_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11229_cast = slice_by_index(begin = var_11229_begin_0, end = var_11229_end_0, end_mask = var_11229_end_mask_0, x = var_11187_cast)[name = tensor<string, []>("op_11229_cast")];
            tensor<int32, [4]> var_11230_begin_0 = const()[name = tensor<string, []>("op_11230_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11230_end_0 = const()[name = tensor<string, []>("op_11230_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_11230_end_mask_0 = const()[name = tensor<string, []>("op_11230_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11230_cast = slice_by_index(begin = var_11230_begin_0, end = var_11230_end_0, end_mask = var_11230_end_mask_0, x = var_11191_cast)[name = tensor<string, []>("op_11230_cast")];
            tensor<int32, [4]> var_11231_begin_0 = const()[name = tensor<string, []>("op_11231_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_11231_end_0 = const()[name = tensor<string, []>("op_11231_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_11231_end_mask_0 = const()[name = tensor<string, []>("op_11231_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11231_cast = slice_by_index(begin = var_11231_begin_0, end = var_11231_end_0, end_mask = var_11231_end_mask_0, x = var_11191_cast)[name = tensor<string, []>("op_11231_cast")];
            tensor<int32, [4]> var_11232_begin_0 = const()[name = tensor<string, []>("op_11232_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_11232_end_0 = const()[name = tensor<string, []>("op_11232_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_11232_end_mask_0 = const()[name = tensor<string, []>("op_11232_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11232_cast = slice_by_index(begin = var_11232_begin_0, end = var_11232_end_0, end_mask = var_11232_end_mask_0, x = var_11191_cast)[name = tensor<string, []>("op_11232_cast")];
            tensor<int32, [4]> var_11233_begin_0 = const()[name = tensor<string, []>("op_11233_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_11233_end_0 = const()[name = tensor<string, []>("op_11233_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_11233_end_mask_0 = const()[name = tensor<string, []>("op_11233_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11233_cast = slice_by_index(begin = var_11233_begin_0, end = var_11233_end_0, end_mask = var_11233_end_mask_0, x = var_11191_cast)[name = tensor<string, []>("op_11233_cast")];
            tensor<int32, [4]> var_11234_begin_0 = const()[name = tensor<string, []>("op_11234_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_11234_end_0 = const()[name = tensor<string, []>("op_11234_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            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, 40, 1, 512]> var_11234_cast = slice_by_index(begin = var_11234_begin_0, end = var_11234_end_0, end_mask = var_11234_end_mask_0, x = var_11191_cast)[name = tensor<string, []>("op_11234_cast")];
            tensor<int32, [4]> var_11235_begin_0 = const()[name = tensor<string, []>("op_11235_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_11235_end_0 = const()[name = tensor<string, []>("op_11235_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_11235_end_mask_0 = const()[name = tensor<string, []>("op_11235_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11235_cast = slice_by_index(begin = var_11235_begin_0, end = var_11235_end_0, end_mask = var_11235_end_mask_0, x = var_11191_cast)[name = tensor<string, []>("op_11235_cast")];
            tensor<int32, [4]> var_11236_begin_0 = const()[name = tensor<string, []>("op_11236_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_11236_end_0 = const()[name = tensor<string, []>("op_11236_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_11236_end_mask_0 = const()[name = tensor<string, []>("op_11236_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11236_cast = slice_by_index(begin = var_11236_begin_0, end = var_11236_end_0, end_mask = var_11236_end_mask_0, x = var_11191_cast)[name = tensor<string, []>("op_11236_cast")];
            tensor<int32, [4]> var_11237_begin_0 = const()[name = tensor<string, []>("op_11237_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_11237_end_0 = const()[name = tensor<string, []>("op_11237_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_11237_end_mask_0 = const()[name = tensor<string, []>("op_11237_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11237_cast = slice_by_index(begin = var_11237_begin_0, end = var_11237_end_0, end_mask = var_11237_end_mask_0, x = var_11191_cast)[name = tensor<string, []>("op_11237_cast")];
            tensor<int32, [4]> var_11238_begin_0 = const()[name = tensor<string, []>("op_11238_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11238_end_0 = const()[name = tensor<string, []>("op_11238_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            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, 40, 1, 512]> var_11238_cast = slice_by_index(begin = var_11238_begin_0, end = var_11238_end_0, end_mask = var_11238_end_mask_0, x = var_11195_cast)[name = tensor<string, []>("op_11238_cast")];
            tensor<int32, [4]> var_11239_begin_0 = const()[name = tensor<string, []>("op_11239_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_11239_end_0 = const()[name = tensor<string, []>("op_11239_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_11239_end_mask_0 = const()[name = tensor<string, []>("op_11239_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11239_cast = slice_by_index(begin = var_11239_begin_0, end = var_11239_end_0, end_mask = var_11239_end_mask_0, x = var_11195_cast)[name = tensor<string, []>("op_11239_cast")];
            tensor<int32, [4]> var_11240_begin_0 = const()[name = tensor<string, []>("op_11240_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_11240_end_0 = const()[name = tensor<string, []>("op_11240_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_11240_end_mask_0 = const()[name = tensor<string, []>("op_11240_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11240_cast = slice_by_index(begin = var_11240_begin_0, end = var_11240_end_0, end_mask = var_11240_end_mask_0, x = var_11195_cast)[name = tensor<string, []>("op_11240_cast")];
            tensor<int32, [4]> var_11241_begin_0 = const()[name = tensor<string, []>("op_11241_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_11241_end_0 = const()[name = tensor<string, []>("op_11241_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_11241_end_mask_0 = const()[name = tensor<string, []>("op_11241_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11241_cast = slice_by_index(begin = var_11241_begin_0, end = var_11241_end_0, end_mask = var_11241_end_mask_0, x = var_11195_cast)[name = tensor<string, []>("op_11241_cast")];
            tensor<int32, [4]> var_11242_begin_0 = const()[name = tensor<string, []>("op_11242_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_11242_end_0 = const()[name = tensor<string, []>("op_11242_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            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, 40, 1, 512]> var_11242_cast = slice_by_index(begin = var_11242_begin_0, end = var_11242_end_0, end_mask = var_11242_end_mask_0, x = var_11195_cast)[name = tensor<string, []>("op_11242_cast")];
            tensor<int32, [4]> var_11243_begin_0 = const()[name = tensor<string, []>("op_11243_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_11243_end_0 = const()[name = tensor<string, []>("op_11243_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_11243_end_mask_0 = const()[name = tensor<string, []>("op_11243_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11243_cast = slice_by_index(begin = var_11243_begin_0, end = var_11243_end_0, end_mask = var_11243_end_mask_0, x = var_11195_cast)[name = tensor<string, []>("op_11243_cast")];
            tensor<int32, [4]> var_11244_begin_0 = const()[name = tensor<string, []>("op_11244_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_11244_end_0 = const()[name = tensor<string, []>("op_11244_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_11244_end_mask_0 = const()[name = tensor<string, []>("op_11244_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11244_cast = slice_by_index(begin = var_11244_begin_0, end = var_11244_end_0, end_mask = var_11244_end_mask_0, x = var_11195_cast)[name = tensor<string, []>("op_11244_cast")];
            tensor<int32, [4]> var_11245_begin_0 = const()[name = tensor<string, []>("op_11245_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_11245_end_0 = const()[name = tensor<string, []>("op_11245_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_11245_end_mask_0 = const()[name = tensor<string, []>("op_11245_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11245_cast = slice_by_index(begin = var_11245_begin_0, end = var_11245_end_0, end_mask = var_11245_end_mask_0, x = var_11195_cast)[name = tensor<string, []>("op_11245_cast")];
            tensor<int32, [4]> var_11246_begin_0 = const()[name = tensor<string, []>("op_11246_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11246_end_0 = const()[name = tensor<string, []>("op_11246_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            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, 40, 1, 512]> var_11246_cast = slice_by_index(begin = var_11246_begin_0, end = var_11246_end_0, end_mask = var_11246_end_mask_0, x = var_11199_cast)[name = tensor<string, []>("op_11246_cast")];
            tensor<int32, [4]> var_11247_begin_0 = const()[name = tensor<string, []>("op_11247_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_11247_end_0 = const()[name = tensor<string, []>("op_11247_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_11247_end_mask_0 = const()[name = tensor<string, []>("op_11247_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11247_cast = slice_by_index(begin = var_11247_begin_0, end = var_11247_end_0, end_mask = var_11247_end_mask_0, x = var_11199_cast)[name = tensor<string, []>("op_11247_cast")];
            tensor<int32, [4]> var_11248_begin_0 = const()[name = tensor<string, []>("op_11248_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_11248_end_0 = const()[name = tensor<string, []>("op_11248_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_11248_end_mask_0 = const()[name = tensor<string, []>("op_11248_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11248_cast = slice_by_index(begin = var_11248_begin_0, end = var_11248_end_0, end_mask = var_11248_end_mask_0, x = var_11199_cast)[name = tensor<string, []>("op_11248_cast")];
            tensor<int32, [4]> var_11249_begin_0 = const()[name = tensor<string, []>("op_11249_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_11249_end_0 = const()[name = tensor<string, []>("op_11249_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_11249_end_mask_0 = const()[name = tensor<string, []>("op_11249_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11249_cast = slice_by_index(begin = var_11249_begin_0, end = var_11249_end_0, end_mask = var_11249_end_mask_0, x = var_11199_cast)[name = tensor<string, []>("op_11249_cast")];
            tensor<int32, [4]> var_11250_begin_0 = const()[name = tensor<string, []>("op_11250_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_11250_end_0 = const()[name = tensor<string, []>("op_11250_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            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, 40, 1, 512]> var_11250_cast = slice_by_index(begin = var_11250_begin_0, end = var_11250_end_0, end_mask = var_11250_end_mask_0, x = var_11199_cast)[name = tensor<string, []>("op_11250_cast")];
            tensor<int32, [4]> var_11251_begin_0 = const()[name = tensor<string, []>("op_11251_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_11251_end_0 = const()[name = tensor<string, []>("op_11251_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_11251_end_mask_0 = const()[name = tensor<string, []>("op_11251_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11251_cast = slice_by_index(begin = var_11251_begin_0, end = var_11251_end_0, end_mask = var_11251_end_mask_0, x = var_11199_cast)[name = tensor<string, []>("op_11251_cast")];
            tensor<int32, [4]> var_11252_begin_0 = const()[name = tensor<string, []>("op_11252_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_11252_end_0 = const()[name = tensor<string, []>("op_11252_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_11252_end_mask_0 = const()[name = tensor<string, []>("op_11252_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11252_cast = slice_by_index(begin = var_11252_begin_0, end = var_11252_end_0, end_mask = var_11252_end_mask_0, x = var_11199_cast)[name = tensor<string, []>("op_11252_cast")];
            tensor<int32, [4]> var_11253_begin_0 = const()[name = tensor<string, []>("op_11253_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_11253_end_0 = const()[name = tensor<string, []>("op_11253_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_11253_end_mask_0 = const()[name = tensor<string, []>("op_11253_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11253_cast = slice_by_index(begin = var_11253_begin_0, end = var_11253_end_0, end_mask = var_11253_end_mask_0, x = var_11199_cast)[name = tensor<string, []>("op_11253_cast")];
            tensor<int32, [4]> var_11254_begin_0 = const()[name = tensor<string, []>("op_11254_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11254_end_0 = const()[name = tensor<string, []>("op_11254_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_11254_end_mask_0 = const()[name = tensor<string, []>("op_11254_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11254_cast = slice_by_index(begin = var_11254_begin_0, end = var_11254_end_0, end_mask = var_11254_end_mask_0, x = var_11203_cast)[name = tensor<string, []>("op_11254_cast")];
            tensor<int32, [4]> var_11255_begin_0 = const()[name = tensor<string, []>("op_11255_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_11255_end_0 = const()[name = tensor<string, []>("op_11255_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_11255_end_mask_0 = const()[name = tensor<string, []>("op_11255_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11255_cast = slice_by_index(begin = var_11255_begin_0, end = var_11255_end_0, end_mask = var_11255_end_mask_0, x = var_11203_cast)[name = tensor<string, []>("op_11255_cast")];
            tensor<int32, [4]> var_11256_begin_0 = const()[name = tensor<string, []>("op_11256_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_11256_end_0 = const()[name = tensor<string, []>("op_11256_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_11256_end_mask_0 = const()[name = tensor<string, []>("op_11256_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11256_cast = slice_by_index(begin = var_11256_begin_0, end = var_11256_end_0, end_mask = var_11256_end_mask_0, x = var_11203_cast)[name = tensor<string, []>("op_11256_cast")];
            tensor<int32, [4]> var_11257_begin_0 = const()[name = tensor<string, []>("op_11257_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_11257_end_0 = const()[name = tensor<string, []>("op_11257_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_11257_end_mask_0 = const()[name = tensor<string, []>("op_11257_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11257_cast = slice_by_index(begin = var_11257_begin_0, end = var_11257_end_0, end_mask = var_11257_end_mask_0, x = var_11203_cast)[name = tensor<string, []>("op_11257_cast")];
            tensor<int32, [4]> var_11258_begin_0 = const()[name = tensor<string, []>("op_11258_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_11258_end_0 = const()[name = tensor<string, []>("op_11258_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_11258_end_mask_0 = const()[name = tensor<string, []>("op_11258_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11258_cast = slice_by_index(begin = var_11258_begin_0, end = var_11258_end_0, end_mask = var_11258_end_mask_0, x = var_11203_cast)[name = tensor<string, []>("op_11258_cast")];
            tensor<int32, [4]> var_11259_begin_0 = const()[name = tensor<string, []>("op_11259_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_11259_end_0 = const()[name = tensor<string, []>("op_11259_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_11259_end_mask_0 = const()[name = tensor<string, []>("op_11259_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11259_cast = slice_by_index(begin = var_11259_begin_0, end = var_11259_end_0, end_mask = var_11259_end_mask_0, x = var_11203_cast)[name = tensor<string, []>("op_11259_cast")];
            tensor<int32, [4]> var_11260_begin_0 = const()[name = tensor<string, []>("op_11260_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_11260_end_0 = const()[name = tensor<string, []>("op_11260_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_11260_end_mask_0 = const()[name = tensor<string, []>("op_11260_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11260_cast = slice_by_index(begin = var_11260_begin_0, end = var_11260_end_0, end_mask = var_11260_end_mask_0, x = var_11203_cast)[name = tensor<string, []>("op_11260_cast")];
            tensor<int32, [4]> var_11261_begin_0 = const()[name = tensor<string, []>("op_11261_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_11261_end_0 = const()[name = tensor<string, []>("op_11261_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_11261_end_mask_0 = const()[name = tensor<string, []>("op_11261_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11261_cast = slice_by_index(begin = var_11261_begin_0, end = var_11261_end_0, end_mask = var_11261_end_mask_0, x = var_11203_cast)[name = tensor<string, []>("op_11261_cast")];
            tensor<int32, [4]> var_11262_begin_0 = const()[name = tensor<string, []>("op_11262_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11262_end_0 = const()[name = tensor<string, []>("op_11262_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_11262_end_mask_0 = const()[name = tensor<string, []>("op_11262_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11262_cast = slice_by_index(begin = var_11262_begin_0, end = var_11262_end_0, end_mask = var_11262_end_mask_0, x = var_11207_cast)[name = tensor<string, []>("op_11262_cast")];
            tensor<int32, [4]> var_11263_begin_0 = const()[name = tensor<string, []>("op_11263_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_11263_end_0 = const()[name = tensor<string, []>("op_11263_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_11263_end_mask_0 = const()[name = tensor<string, []>("op_11263_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11263_cast = slice_by_index(begin = var_11263_begin_0, end = var_11263_end_0, end_mask = var_11263_end_mask_0, x = var_11207_cast)[name = tensor<string, []>("op_11263_cast")];
            tensor<int32, [4]> var_11264_begin_0 = const()[name = tensor<string, []>("op_11264_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_11264_end_0 = const()[name = tensor<string, []>("op_11264_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_11264_end_mask_0 = const()[name = tensor<string, []>("op_11264_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11264_cast = slice_by_index(begin = var_11264_begin_0, end = var_11264_end_0, end_mask = var_11264_end_mask_0, x = var_11207_cast)[name = tensor<string, []>("op_11264_cast")];
            tensor<int32, [4]> var_11265_begin_0 = const()[name = tensor<string, []>("op_11265_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_11265_end_0 = const()[name = tensor<string, []>("op_11265_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_11265_end_mask_0 = const()[name = tensor<string, []>("op_11265_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11265_cast = slice_by_index(begin = var_11265_begin_0, end = var_11265_end_0, end_mask = var_11265_end_mask_0, x = var_11207_cast)[name = tensor<string, []>("op_11265_cast")];
            tensor<int32, [4]> var_11266_begin_0 = const()[name = tensor<string, []>("op_11266_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_11266_end_0 = const()[name = tensor<string, []>("op_11266_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_11266_end_mask_0 = const()[name = tensor<string, []>("op_11266_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11266_cast = slice_by_index(begin = var_11266_begin_0, end = var_11266_end_0, end_mask = var_11266_end_mask_0, x = var_11207_cast)[name = tensor<string, []>("op_11266_cast")];
            tensor<int32, [4]> var_11267_begin_0 = const()[name = tensor<string, []>("op_11267_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_11267_end_0 = const()[name = tensor<string, []>("op_11267_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_11267_end_mask_0 = const()[name = tensor<string, []>("op_11267_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11267_cast = slice_by_index(begin = var_11267_begin_0, end = var_11267_end_0, end_mask = var_11267_end_mask_0, x = var_11207_cast)[name = tensor<string, []>("op_11267_cast")];
            tensor<int32, [4]> var_11268_begin_0 = const()[name = tensor<string, []>("op_11268_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_11268_end_0 = const()[name = tensor<string, []>("op_11268_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_11268_end_mask_0 = const()[name = tensor<string, []>("op_11268_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11268_cast = slice_by_index(begin = var_11268_begin_0, end = var_11268_end_0, end_mask = var_11268_end_mask_0, x = var_11207_cast)[name = tensor<string, []>("op_11268_cast")];
            tensor<int32, [4]> var_11269_begin_0 = const()[name = tensor<string, []>("op_11269_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_11269_end_0 = const()[name = tensor<string, []>("op_11269_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_11269_end_mask_0 = const()[name = tensor<string, []>("op_11269_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11269_cast = slice_by_index(begin = var_11269_begin_0, end = var_11269_end_0, end_mask = var_11269_end_mask_0, x = var_11207_cast)[name = tensor<string, []>("op_11269_cast")];
            tensor<int32, [4]> var_11270_begin_0 = const()[name = tensor<string, []>("op_11270_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11270_end_0 = const()[name = tensor<string, []>("op_11270_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_11270_end_mask_0 = const()[name = tensor<string, []>("op_11270_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11270_cast = slice_by_index(begin = var_11270_begin_0, end = var_11270_end_0, end_mask = var_11270_end_mask_0, x = var_11211_cast)[name = tensor<string, []>("op_11270_cast")];
            tensor<int32, [4]> var_11271_begin_0 = const()[name = tensor<string, []>("op_11271_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_11271_end_0 = const()[name = tensor<string, []>("op_11271_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_11271_end_mask_0 = const()[name = tensor<string, []>("op_11271_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11271_cast = slice_by_index(begin = var_11271_begin_0, end = var_11271_end_0, end_mask = var_11271_end_mask_0, x = var_11211_cast)[name = tensor<string, []>("op_11271_cast")];
            tensor<int32, [4]> var_11272_begin_0 = const()[name = tensor<string, []>("op_11272_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_11272_end_0 = const()[name = tensor<string, []>("op_11272_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_11272_end_mask_0 = const()[name = tensor<string, []>("op_11272_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11272_cast = slice_by_index(begin = var_11272_begin_0, end = var_11272_end_0, end_mask = var_11272_end_mask_0, x = var_11211_cast)[name = tensor<string, []>("op_11272_cast")];
            tensor<int32, [4]> var_11273_begin_0 = const()[name = tensor<string, []>("op_11273_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_11273_end_0 = const()[name = tensor<string, []>("op_11273_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_11273_end_mask_0 = const()[name = tensor<string, []>("op_11273_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11273_cast = slice_by_index(begin = var_11273_begin_0, end = var_11273_end_0, end_mask = var_11273_end_mask_0, x = var_11211_cast)[name = tensor<string, []>("op_11273_cast")];
            tensor<int32, [4]> var_11274_begin_0 = const()[name = tensor<string, []>("op_11274_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_11274_end_0 = const()[name = tensor<string, []>("op_11274_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_11274_end_mask_0 = const()[name = tensor<string, []>("op_11274_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11274_cast = slice_by_index(begin = var_11274_begin_0, end = var_11274_end_0, end_mask = var_11274_end_mask_0, x = var_11211_cast)[name = tensor<string, []>("op_11274_cast")];
            tensor<int32, [4]> var_11275_begin_0 = const()[name = tensor<string, []>("op_11275_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_11275_end_0 = const()[name = tensor<string, []>("op_11275_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_11275_end_mask_0 = const()[name = tensor<string, []>("op_11275_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11275_cast = slice_by_index(begin = var_11275_begin_0, end = var_11275_end_0, end_mask = var_11275_end_mask_0, x = var_11211_cast)[name = tensor<string, []>("op_11275_cast")];
            tensor<int32, [4]> var_11276_begin_0 = const()[name = tensor<string, []>("op_11276_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_11276_end_0 = const()[name = tensor<string, []>("op_11276_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_11276_end_mask_0 = const()[name = tensor<string, []>("op_11276_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11276_cast = slice_by_index(begin = var_11276_begin_0, end = var_11276_end_0, end_mask = var_11276_end_mask_0, x = var_11211_cast)[name = tensor<string, []>("op_11276_cast")];
            tensor<int32, [4]> var_11277_begin_0 = const()[name = tensor<string, []>("op_11277_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_11277_end_0 = const()[name = tensor<string, []>("op_11277_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_11277_end_mask_0 = const()[name = tensor<string, []>("op_11277_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_11277_cast = slice_by_index(begin = var_11277_begin_0, end = var_11277_end_0, end_mask = var_11277_end_mask_0, x = var_11211_cast)[name = tensor<string, []>("op_11277_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_11282_begin_0 = const()[name = tensor<string, []>("op_11282_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11282_end_0 = const()[name = tensor<string, []>("op_11282_end_0"), val = tensor<int32, [4]>([2, 77, 1, 40])];
            tensor<bool, [4]> var_11282_end_mask_0 = const()[name = tensor<string, []>("op_11282_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)[name = tensor<string, []>("transpose_4")];
            tensor<fp16, [2, 77, 1, 40]> var_11282_cast = slice_by_index(begin = var_11282_begin_0, end = var_11282_end_0, end_mask = var_11282_end_mask_0, x = transpose_4)[name = tensor<string, []>("op_11282_cast")];
            tensor<int32, [4]> var_11286_begin_0 = const()[name = tensor<string, []>("op_11286_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 40])];
            tensor<int32, [4]> var_11286_end_0 = const()[name = tensor<string, []>("op_11286_end_0"), val = tensor<int32, [4]>([2, 77, 1, 80])];
            tensor<bool, [4]> var_11286_end_mask_0 = const()[name = tensor<string, []>("op_11286_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 40]> var_11286_cast = slice_by_index(begin = var_11286_begin_0, end = var_11286_end_0, end_mask = var_11286_end_mask_0, x = transpose_4)[name = tensor<string, []>("op_11286_cast")];
            tensor<int32, [4]> var_11290_begin_0 = const()[name = tensor<string, []>("op_11290_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 80])];
            tensor<int32, [4]> var_11290_end_0 = const()[name = tensor<string, []>("op_11290_end_0"), val = tensor<int32, [4]>([2, 77, 1, 120])];
            tensor<bool, [4]> var_11290_end_mask_0 = const()[name = tensor<string, []>("op_11290_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 40]> var_11290_cast = slice_by_index(begin = var_11290_begin_0, end = var_11290_end_0, end_mask = var_11290_end_mask_0, x = transpose_4)[name = tensor<string, []>("op_11290_cast")];
            tensor<int32, [4]> var_11294_begin_0 = const()[name = tensor<string, []>("op_11294_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 120])];
            tensor<int32, [4]> var_11294_end_0 = const()[name = tensor<string, []>("op_11294_end_0"), val = tensor<int32, [4]>([2, 77, 1, 160])];
            tensor<bool, [4]> var_11294_end_mask_0 = const()[name = tensor<string, []>("op_11294_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 40]> var_11294_cast = slice_by_index(begin = var_11294_begin_0, end = var_11294_end_0, end_mask = var_11294_end_mask_0, x = transpose_4)[name = tensor<string, []>("op_11294_cast")];
            tensor<int32, [4]> var_11298_begin_0 = const()[name = tensor<string, []>("op_11298_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_11298_end_0 = const()[name = tensor<string, []>("op_11298_end_0"), val = tensor<int32, [4]>([2, 77, 1, 200])];
            tensor<bool, [4]> var_11298_end_mask_0 = const()[name = tensor<string, []>("op_11298_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 40]> var_11298_cast = slice_by_index(begin = var_11298_begin_0, end = var_11298_end_0, end_mask = var_11298_end_mask_0, x = transpose_4)[name = tensor<string, []>("op_11298_cast")];
            tensor<int32, [4]> var_11302_begin_0 = const()[name = tensor<string, []>("op_11302_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 200])];
            tensor<int32, [4]> var_11302_end_0 = const()[name = tensor<string, []>("op_11302_end_0"), val = tensor<int32, [4]>([2, 77, 1, 240])];
            tensor<bool, [4]> var_11302_end_mask_0 = const()[name = tensor<string, []>("op_11302_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 40]> var_11302_cast = slice_by_index(begin = var_11302_begin_0, end = var_11302_end_0, end_mask = var_11302_end_mask_0, x = transpose_4)[name = tensor<string, []>("op_11302_cast")];
            tensor<int32, [4]> var_11306_begin_0 = const()[name = tensor<string, []>("op_11306_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 240])];
            tensor<int32, [4]> var_11306_end_0 = const()[name = tensor<string, []>("op_11306_end_0"), val = tensor<int32, [4]>([2, 77, 1, 280])];
            tensor<bool, [4]> var_11306_end_mask_0 = const()[name = tensor<string, []>("op_11306_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 40]> var_11306_cast = slice_by_index(begin = var_11306_begin_0, end = var_11306_end_0, end_mask = var_11306_end_mask_0, x = transpose_4)[name = tensor<string, []>("op_11306_cast")];
            tensor<int32, [4]> var_11310_begin_0 = const()[name = tensor<string, []>("op_11310_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 280])];
            tensor<int32, [4]> var_11310_end_0 = const()[name = tensor<string, []>("op_11310_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_11310_end_mask_0 = const()[name = tensor<string, []>("op_11310_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 40]> var_11310_cast = slice_by_index(begin = var_11310_begin_0, end = var_11310_end_0, end_mask = var_11310_end_mask_0, x = transpose_4)[name = tensor<string, []>("op_11310_cast")];
            tensor<int32, [4]> var_11312_begin_0 = const()[name = tensor<string, []>("op_11312_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11312_end_0 = const()[name = tensor<string, []>("op_11312_end_0"), val = tensor<int32, [4]>([2, 40, 1, 77])];
            tensor<bool, [4]> var_11312_end_mask_0 = const()[name = tensor<string, []>("op_11312_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_11312_cast = slice_by_index(begin = var_11312_begin_0, end = var_11312_end_0, end_mask = var_11312_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_11312_cast")];
            tensor<int32, [4]> var_11316_begin_0 = const()[name = tensor<string, []>("op_11316_begin_0"), val = tensor<int32, [4]>([0, 40, 0, 0])];
            tensor<int32, [4]> var_11316_end_0 = const()[name = tensor<string, []>("op_11316_end_0"), val = tensor<int32, [4]>([2, 80, 1, 77])];
            tensor<bool, [4]> var_11316_end_mask_0 = const()[name = tensor<string, []>("op_11316_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_11316_cast = slice_by_index(begin = var_11316_begin_0, end = var_11316_end_0, end_mask = var_11316_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_11316_cast")];
            tensor<int32, [4]> var_11320_begin_0 = const()[name = tensor<string, []>("op_11320_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_11320_end_0 = const()[name = tensor<string, []>("op_11320_end_0"), val = tensor<int32, [4]>([2, 120, 1, 77])];
            tensor<bool, [4]> var_11320_end_mask_0 = const()[name = tensor<string, []>("op_11320_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_11320_cast = slice_by_index(begin = var_11320_begin_0, end = var_11320_end_0, end_mask = var_11320_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_11320_cast")];
            tensor<int32, [4]> var_11324_begin_0 = const()[name = tensor<string, []>("op_11324_begin_0"), val = tensor<int32, [4]>([0, 120, 0, 0])];
            tensor<int32, [4]> var_11324_end_0 = const()[name = tensor<string, []>("op_11324_end_0"), val = tensor<int32, [4]>([2, 160, 1, 77])];
            tensor<bool, [4]> var_11324_end_mask_0 = const()[name = tensor<string, []>("op_11324_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_11324_cast = slice_by_index(begin = var_11324_begin_0, end = var_11324_end_0, end_mask = var_11324_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_11324_cast")];
            tensor<int32, [4]> var_11328_begin_0 = const()[name = tensor<string, []>("op_11328_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_11328_end_0 = const()[name = tensor<string, []>("op_11328_end_0"), val = tensor<int32, [4]>([2, 200, 1, 77])];
            tensor<bool, [4]> var_11328_end_mask_0 = const()[name = tensor<string, []>("op_11328_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_11328_cast = slice_by_index(begin = var_11328_begin_0, end = var_11328_end_0, end_mask = var_11328_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_11328_cast")];
            tensor<int32, [4]> var_11332_begin_0 = const()[name = tensor<string, []>("op_11332_begin_0"), val = tensor<int32, [4]>([0, 200, 0, 0])];
            tensor<int32, [4]> var_11332_end_0 = const()[name = tensor<string, []>("op_11332_end_0"), val = tensor<int32, [4]>([2, 240, 1, 77])];
            tensor<bool, [4]> var_11332_end_mask_0 = const()[name = tensor<string, []>("op_11332_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_11332_cast = slice_by_index(begin = var_11332_begin_0, end = var_11332_end_0, end_mask = var_11332_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_11332_cast")];
            tensor<int32, [4]> var_11336_begin_0 = const()[name = tensor<string, []>("op_11336_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_11336_end_0 = const()[name = tensor<string, []>("op_11336_end_0"), val = tensor<int32, [4]>([2, 280, 1, 77])];
            tensor<bool, [4]> var_11336_end_mask_0 = const()[name = tensor<string, []>("op_11336_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_11336_cast = slice_by_index(begin = var_11336_begin_0, end = var_11336_end_0, end_mask = var_11336_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_11336_cast")];
            tensor<int32, [4]> var_11340_begin_0 = const()[name = tensor<string, []>("op_11340_begin_0"), val = tensor<int32, [4]>([0, 280, 0, 0])];
            tensor<int32, [4]> var_11340_end_0 = const()[name = tensor<string, []>("op_11340_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_11340_end_mask_0 = const()[name = tensor<string, []>("op_11340_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_11340_cast = slice_by_index(begin = var_11340_begin_0, end = var_11340_end_0, end_mask = var_11340_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_11340_cast")];
            tensor<string, []> var_11344_equation_0 = const()[name = tensor<string, []>("op_11344_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11344_cast = einsum(equation = var_11344_equation_0, values = (var_11282_cast, var_11214_cast))[name = tensor<string, []>("op_11344_cast")];
            tensor<fp16, []> var_11345_to_fp16 = const()[name = tensor<string, []>("op_11345_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_961_cast = mul(x = var_11344_cast, y = var_11345_to_fp16)[name = tensor<string, []>("aw_chunk_961_cast")];
            tensor<string, []> var_11348_equation_0 = const()[name = tensor<string, []>("op_11348_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11348_cast = einsum(equation = var_11348_equation_0, values = (var_11282_cast, var_11215_cast))[name = tensor<string, []>("op_11348_cast")];
            tensor<fp16, []> var_11349_to_fp16 = const()[name = tensor<string, []>("op_11349_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_963_cast = mul(x = var_11348_cast, y = var_11349_to_fp16)[name = tensor<string, []>("aw_chunk_963_cast")];
            tensor<string, []> var_11352_equation_0 = const()[name = tensor<string, []>("op_11352_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11352_cast = einsum(equation = var_11352_equation_0, values = (var_11282_cast, var_11216_cast))[name = tensor<string, []>("op_11352_cast")];
            tensor<fp16, []> var_11353_to_fp16 = const()[name = tensor<string, []>("op_11353_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_965_cast = mul(x = var_11352_cast, y = var_11353_to_fp16)[name = tensor<string, []>("aw_chunk_965_cast")];
            tensor<string, []> var_11356_equation_0 = const()[name = tensor<string, []>("op_11356_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11356_cast = einsum(equation = var_11356_equation_0, values = (var_11282_cast, var_11217_cast))[name = tensor<string, []>("op_11356_cast")];
            tensor<fp16, []> var_11357_to_fp16 = const()[name = tensor<string, []>("op_11357_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_967_cast = mul(x = var_11356_cast, y = var_11357_to_fp16)[name = tensor<string, []>("aw_chunk_967_cast")];
            tensor<string, []> var_11360_equation_0 = const()[name = tensor<string, []>("op_11360_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11360_cast = einsum(equation = var_11360_equation_0, values = (var_11282_cast, var_11218_cast))[name = tensor<string, []>("op_11360_cast")];
            tensor<fp16, []> var_11361_to_fp16 = const()[name = tensor<string, []>("op_11361_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_969_cast = mul(x = var_11360_cast, y = var_11361_to_fp16)[name = tensor<string, []>("aw_chunk_969_cast")];
            tensor<string, []> var_11364_equation_0 = const()[name = tensor<string, []>("op_11364_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11364_cast = einsum(equation = var_11364_equation_0, values = (var_11282_cast, var_11219_cast))[name = tensor<string, []>("op_11364_cast")];
            tensor<fp16, []> var_11365_to_fp16 = const()[name = tensor<string, []>("op_11365_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_971_cast = mul(x = var_11364_cast, y = var_11365_to_fp16)[name = tensor<string, []>("aw_chunk_971_cast")];
            tensor<string, []> var_11368_equation_0 = const()[name = tensor<string, []>("op_11368_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11368_cast = einsum(equation = var_11368_equation_0, values = (var_11282_cast, var_11220_cast))[name = tensor<string, []>("op_11368_cast")];
            tensor<fp16, []> var_11369_to_fp16 = const()[name = tensor<string, []>("op_11369_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_973_cast = mul(x = var_11368_cast, y = var_11369_to_fp16)[name = tensor<string, []>("aw_chunk_973_cast")];
            tensor<string, []> var_11372_equation_0 = const()[name = tensor<string, []>("op_11372_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11372_cast = einsum(equation = var_11372_equation_0, values = (var_11282_cast, var_11221_cast))[name = tensor<string, []>("op_11372_cast")];
            tensor<fp16, []> var_11373_to_fp16 = const()[name = tensor<string, []>("op_11373_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_975_cast = mul(x = var_11372_cast, y = var_11373_to_fp16)[name = tensor<string, []>("aw_chunk_975_cast")];
            tensor<string, []> var_11376_equation_0 = const()[name = tensor<string, []>("op_11376_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11376_cast = einsum(equation = var_11376_equation_0, values = (var_11286_cast, var_11222_cast))[name = tensor<string, []>("op_11376_cast")];
            tensor<fp16, []> var_11377_to_fp16 = const()[name = tensor<string, []>("op_11377_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_977_cast = mul(x = var_11376_cast, y = var_11377_to_fp16)[name = tensor<string, []>("aw_chunk_977_cast")];
            tensor<string, []> var_11380_equation_0 = const()[name = tensor<string, []>("op_11380_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11380_cast = einsum(equation = var_11380_equation_0, values = (var_11286_cast, var_11223_cast))[name = tensor<string, []>("op_11380_cast")];
            tensor<fp16, []> var_11381_to_fp16 = const()[name = tensor<string, []>("op_11381_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_979_cast = mul(x = var_11380_cast, y = var_11381_to_fp16)[name = tensor<string, []>("aw_chunk_979_cast")];
            tensor<string, []> var_11384_equation_0 = const()[name = tensor<string, []>("op_11384_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11384_cast = einsum(equation = var_11384_equation_0, values = (var_11286_cast, var_11224_cast))[name = tensor<string, []>("op_11384_cast")];
            tensor<fp16, []> var_11385_to_fp16 = const()[name = tensor<string, []>("op_11385_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_981_cast = mul(x = var_11384_cast, y = var_11385_to_fp16)[name = tensor<string, []>("aw_chunk_981_cast")];
            tensor<string, []> var_11388_equation_0 = const()[name = tensor<string, []>("op_11388_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11388_cast = einsum(equation = var_11388_equation_0, values = (var_11286_cast, var_11225_cast))[name = tensor<string, []>("op_11388_cast")];
            tensor<fp16, []> var_11389_to_fp16 = const()[name = tensor<string, []>("op_11389_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_983_cast = mul(x = var_11388_cast, y = var_11389_to_fp16)[name = tensor<string, []>("aw_chunk_983_cast")];
            tensor<string, []> var_11392_equation_0 = const()[name = tensor<string, []>("op_11392_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11392_cast = einsum(equation = var_11392_equation_0, values = (var_11286_cast, var_11226_cast))[name = tensor<string, []>("op_11392_cast")];
            tensor<fp16, []> var_11393_to_fp16 = const()[name = tensor<string, []>("op_11393_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_985_cast = mul(x = var_11392_cast, y = var_11393_to_fp16)[name = tensor<string, []>("aw_chunk_985_cast")];
            tensor<string, []> var_11396_equation_0 = const()[name = tensor<string, []>("op_11396_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11396_cast = einsum(equation = var_11396_equation_0, values = (var_11286_cast, var_11227_cast))[name = tensor<string, []>("op_11396_cast")];
            tensor<fp16, []> var_11397_to_fp16 = const()[name = tensor<string, []>("op_11397_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_987_cast = mul(x = var_11396_cast, y = var_11397_to_fp16)[name = tensor<string, []>("aw_chunk_987_cast")];
            tensor<string, []> var_11400_equation_0 = const()[name = tensor<string, []>("op_11400_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11400_cast = einsum(equation = var_11400_equation_0, values = (var_11286_cast, var_11228_cast))[name = tensor<string, []>("op_11400_cast")];
            tensor<fp16, []> var_11401_to_fp16 = const()[name = tensor<string, []>("op_11401_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_989_cast = mul(x = var_11400_cast, y = var_11401_to_fp16)[name = tensor<string, []>("aw_chunk_989_cast")];
            tensor<string, []> var_11404_equation_0 = const()[name = tensor<string, []>("op_11404_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11404_cast = einsum(equation = var_11404_equation_0, values = (var_11286_cast, var_11229_cast))[name = tensor<string, []>("op_11404_cast")];
            tensor<fp16, []> var_11405_to_fp16 = const()[name = tensor<string, []>("op_11405_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_991_cast = mul(x = var_11404_cast, y = var_11405_to_fp16)[name = tensor<string, []>("aw_chunk_991_cast")];
            tensor<string, []> var_11408_equation_0 = const()[name = tensor<string, []>("op_11408_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11408_cast = einsum(equation = var_11408_equation_0, values = (var_11290_cast, var_11230_cast))[name = tensor<string, []>("op_11408_cast")];
            tensor<fp16, []> var_11409_to_fp16 = const()[name = tensor<string, []>("op_11409_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_993_cast = mul(x = var_11408_cast, y = var_11409_to_fp16)[name = tensor<string, []>("aw_chunk_993_cast")];
            tensor<string, []> var_11412_equation_0 = const()[name = tensor<string, []>("op_11412_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11412_cast = einsum(equation = var_11412_equation_0, values = (var_11290_cast, var_11231_cast))[name = tensor<string, []>("op_11412_cast")];
            tensor<fp16, []> var_11413_to_fp16 = const()[name = tensor<string, []>("op_11413_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_995_cast = mul(x = var_11412_cast, y = var_11413_to_fp16)[name = tensor<string, []>("aw_chunk_995_cast")];
            tensor<string, []> var_11416_equation_0 = const()[name = tensor<string, []>("op_11416_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11416_cast = einsum(equation = var_11416_equation_0, values = (var_11290_cast, var_11232_cast))[name = tensor<string, []>("op_11416_cast")];
            tensor<fp16, []> var_11417_to_fp16 = const()[name = tensor<string, []>("op_11417_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_997_cast = mul(x = var_11416_cast, y = var_11417_to_fp16)[name = tensor<string, []>("aw_chunk_997_cast")];
            tensor<string, []> var_11420_equation_0 = const()[name = tensor<string, []>("op_11420_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11420_cast = einsum(equation = var_11420_equation_0, values = (var_11290_cast, var_11233_cast))[name = tensor<string, []>("op_11420_cast")];
            tensor<fp16, []> var_11421_to_fp16 = const()[name = tensor<string, []>("op_11421_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_999_cast = mul(x = var_11420_cast, y = var_11421_to_fp16)[name = tensor<string, []>("aw_chunk_999_cast")];
            tensor<string, []> var_11424_equation_0 = const()[name = tensor<string, []>("op_11424_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11424_cast = einsum(equation = var_11424_equation_0, values = (var_11290_cast, var_11234_cast))[name = tensor<string, []>("op_11424_cast")];
            tensor<fp16, []> var_11425_to_fp16 = const()[name = tensor<string, []>("op_11425_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1001_cast = mul(x = var_11424_cast, y = var_11425_to_fp16)[name = tensor<string, []>("aw_chunk_1001_cast")];
            tensor<string, []> var_11428_equation_0 = const()[name = tensor<string, []>("op_11428_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11428_cast = einsum(equation = var_11428_equation_0, values = (var_11290_cast, var_11235_cast))[name = tensor<string, []>("op_11428_cast")];
            tensor<fp16, []> var_11429_to_fp16 = const()[name = tensor<string, []>("op_11429_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1003_cast = mul(x = var_11428_cast, y = var_11429_to_fp16)[name = tensor<string, []>("aw_chunk_1003_cast")];
            tensor<string, []> var_11432_equation_0 = const()[name = tensor<string, []>("op_11432_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11432_cast = einsum(equation = var_11432_equation_0, values = (var_11290_cast, var_11236_cast))[name = tensor<string, []>("op_11432_cast")];
            tensor<fp16, []> var_11433_to_fp16 = const()[name = tensor<string, []>("op_11433_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1005_cast = mul(x = var_11432_cast, y = var_11433_to_fp16)[name = tensor<string, []>("aw_chunk_1005_cast")];
            tensor<string, []> var_11436_equation_0 = const()[name = tensor<string, []>("op_11436_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11436_cast = einsum(equation = var_11436_equation_0, values = (var_11290_cast, var_11237_cast))[name = tensor<string, []>("op_11436_cast")];
            tensor<fp16, []> var_11437_to_fp16 = const()[name = tensor<string, []>("op_11437_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1007_cast = mul(x = var_11436_cast, y = var_11437_to_fp16)[name = tensor<string, []>("aw_chunk_1007_cast")];
            tensor<string, []> var_11440_equation_0 = const()[name = tensor<string, []>("op_11440_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11440_cast = einsum(equation = var_11440_equation_0, values = (var_11294_cast, var_11238_cast))[name = tensor<string, []>("op_11440_cast")];
            tensor<fp16, []> var_11441_to_fp16 = const()[name = tensor<string, []>("op_11441_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1009_cast = mul(x = var_11440_cast, y = var_11441_to_fp16)[name = tensor<string, []>("aw_chunk_1009_cast")];
            tensor<string, []> var_11444_equation_0 = const()[name = tensor<string, []>("op_11444_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11444_cast = einsum(equation = var_11444_equation_0, values = (var_11294_cast, var_11239_cast))[name = tensor<string, []>("op_11444_cast")];
            tensor<fp16, []> var_11445_to_fp16 = const()[name = tensor<string, []>("op_11445_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1011_cast = mul(x = var_11444_cast, y = var_11445_to_fp16)[name = tensor<string, []>("aw_chunk_1011_cast")];
            tensor<string, []> var_11448_equation_0 = const()[name = tensor<string, []>("op_11448_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11448_cast = einsum(equation = var_11448_equation_0, values = (var_11294_cast, var_11240_cast))[name = tensor<string, []>("op_11448_cast")];
            tensor<fp16, []> var_11449_to_fp16 = const()[name = tensor<string, []>("op_11449_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1013_cast = mul(x = var_11448_cast, y = var_11449_to_fp16)[name = tensor<string, []>("aw_chunk_1013_cast")];
            tensor<string, []> var_11452_equation_0 = const()[name = tensor<string, []>("op_11452_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11452_cast = einsum(equation = var_11452_equation_0, values = (var_11294_cast, var_11241_cast))[name = tensor<string, []>("op_11452_cast")];
            tensor<fp16, []> var_11453_to_fp16 = const()[name = tensor<string, []>("op_11453_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1015_cast = mul(x = var_11452_cast, y = var_11453_to_fp16)[name = tensor<string, []>("aw_chunk_1015_cast")];
            tensor<string, []> var_11456_equation_0 = const()[name = tensor<string, []>("op_11456_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11456_cast = einsum(equation = var_11456_equation_0, values = (var_11294_cast, var_11242_cast))[name = tensor<string, []>("op_11456_cast")];
            tensor<fp16, []> var_11457_to_fp16 = const()[name = tensor<string, []>("op_11457_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1017_cast = mul(x = var_11456_cast, y = var_11457_to_fp16)[name = tensor<string, []>("aw_chunk_1017_cast")];
            tensor<string, []> var_11460_equation_0 = const()[name = tensor<string, []>("op_11460_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11460_cast = einsum(equation = var_11460_equation_0, values = (var_11294_cast, var_11243_cast))[name = tensor<string, []>("op_11460_cast")];
            tensor<fp16, []> var_11461_to_fp16 = const()[name = tensor<string, []>("op_11461_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1019_cast = mul(x = var_11460_cast, y = var_11461_to_fp16)[name = tensor<string, []>("aw_chunk_1019_cast")];
            tensor<string, []> var_11464_equation_0 = const()[name = tensor<string, []>("op_11464_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11464_cast = einsum(equation = var_11464_equation_0, values = (var_11294_cast, var_11244_cast))[name = tensor<string, []>("op_11464_cast")];
            tensor<fp16, []> var_11465_to_fp16 = const()[name = tensor<string, []>("op_11465_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1021_cast = mul(x = var_11464_cast, y = var_11465_to_fp16)[name = tensor<string, []>("aw_chunk_1021_cast")];
            tensor<string, []> var_11468_equation_0 = const()[name = tensor<string, []>("op_11468_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11468_cast = einsum(equation = var_11468_equation_0, values = (var_11294_cast, var_11245_cast))[name = tensor<string, []>("op_11468_cast")];
            tensor<fp16, []> var_11469_to_fp16 = const()[name = tensor<string, []>("op_11469_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1023_cast = mul(x = var_11468_cast, y = var_11469_to_fp16)[name = tensor<string, []>("aw_chunk_1023_cast")];
            tensor<string, []> var_11472_equation_0 = const()[name = tensor<string, []>("op_11472_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11472_cast = einsum(equation = var_11472_equation_0, values = (var_11298_cast, var_11246_cast))[name = tensor<string, []>("op_11472_cast")];
            tensor<fp16, []> var_11473_to_fp16 = const()[name = tensor<string, []>("op_11473_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1025_cast = mul(x = var_11472_cast, y = var_11473_to_fp16)[name = tensor<string, []>("aw_chunk_1025_cast")];
            tensor<string, []> var_11476_equation_0 = const()[name = tensor<string, []>("op_11476_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11476_cast = einsum(equation = var_11476_equation_0, values = (var_11298_cast, var_11247_cast))[name = tensor<string, []>("op_11476_cast")];
            tensor<fp16, []> var_11477_to_fp16 = const()[name = tensor<string, []>("op_11477_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1027_cast = mul(x = var_11476_cast, y = var_11477_to_fp16)[name = tensor<string, []>("aw_chunk_1027_cast")];
            tensor<string, []> var_11480_equation_0 = const()[name = tensor<string, []>("op_11480_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11480_cast = einsum(equation = var_11480_equation_0, values = (var_11298_cast, var_11248_cast))[name = tensor<string, []>("op_11480_cast")];
            tensor<fp16, []> var_11481_to_fp16 = const()[name = tensor<string, []>("op_11481_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1029_cast = mul(x = var_11480_cast, y = var_11481_to_fp16)[name = tensor<string, []>("aw_chunk_1029_cast")];
            tensor<string, []> var_11484_equation_0 = const()[name = tensor<string, []>("op_11484_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11484_cast = einsum(equation = var_11484_equation_0, values = (var_11298_cast, var_11249_cast))[name = tensor<string, []>("op_11484_cast")];
            tensor<fp16, []> var_11485_to_fp16 = const()[name = tensor<string, []>("op_11485_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1031_cast = mul(x = var_11484_cast, y = var_11485_to_fp16)[name = tensor<string, []>("aw_chunk_1031_cast")];
            tensor<string, []> var_11488_equation_0 = const()[name = tensor<string, []>("op_11488_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11488_cast = einsum(equation = var_11488_equation_0, values = (var_11298_cast, var_11250_cast))[name = tensor<string, []>("op_11488_cast")];
            tensor<fp16, []> var_11489_to_fp16 = const()[name = tensor<string, []>("op_11489_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1033_cast = mul(x = var_11488_cast, y = var_11489_to_fp16)[name = tensor<string, []>("aw_chunk_1033_cast")];
            tensor<string, []> var_11492_equation_0 = const()[name = tensor<string, []>("op_11492_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11492_cast = einsum(equation = var_11492_equation_0, values = (var_11298_cast, var_11251_cast))[name = tensor<string, []>("op_11492_cast")];
            tensor<fp16, []> var_11493_to_fp16 = const()[name = tensor<string, []>("op_11493_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1035_cast = mul(x = var_11492_cast, y = var_11493_to_fp16)[name = tensor<string, []>("aw_chunk_1035_cast")];
            tensor<string, []> var_11496_equation_0 = const()[name = tensor<string, []>("op_11496_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11496_cast = einsum(equation = var_11496_equation_0, values = (var_11298_cast, var_11252_cast))[name = tensor<string, []>("op_11496_cast")];
            tensor<fp16, []> var_11497_to_fp16 = const()[name = tensor<string, []>("op_11497_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1037_cast = mul(x = var_11496_cast, y = var_11497_to_fp16)[name = tensor<string, []>("aw_chunk_1037_cast")];
            tensor<string, []> var_11500_equation_0 = const()[name = tensor<string, []>("op_11500_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11500_cast = einsum(equation = var_11500_equation_0, values = (var_11298_cast, var_11253_cast))[name = tensor<string, []>("op_11500_cast")];
            tensor<fp16, []> var_11501_to_fp16 = const()[name = tensor<string, []>("op_11501_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1039_cast = mul(x = var_11500_cast, y = var_11501_to_fp16)[name = tensor<string, []>("aw_chunk_1039_cast")];
            tensor<string, []> var_11504_equation_0 = const()[name = tensor<string, []>("op_11504_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11504_cast = einsum(equation = var_11504_equation_0, values = (var_11302_cast, var_11254_cast))[name = tensor<string, []>("op_11504_cast")];
            tensor<fp16, []> var_11505_to_fp16 = const()[name = tensor<string, []>("op_11505_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1041_cast = mul(x = var_11504_cast, y = var_11505_to_fp16)[name = tensor<string, []>("aw_chunk_1041_cast")];
            tensor<string, []> var_11508_equation_0 = const()[name = tensor<string, []>("op_11508_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11508_cast = einsum(equation = var_11508_equation_0, values = (var_11302_cast, var_11255_cast))[name = tensor<string, []>("op_11508_cast")];
            tensor<fp16, []> var_11509_to_fp16 = const()[name = tensor<string, []>("op_11509_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1043_cast = mul(x = var_11508_cast, y = var_11509_to_fp16)[name = tensor<string, []>("aw_chunk_1043_cast")];
            tensor<string, []> var_11512_equation_0 = const()[name = tensor<string, []>("op_11512_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11512_cast = einsum(equation = var_11512_equation_0, values = (var_11302_cast, var_11256_cast))[name = tensor<string, []>("op_11512_cast")];
            tensor<fp16, []> var_11513_to_fp16 = const()[name = tensor<string, []>("op_11513_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1045_cast = mul(x = var_11512_cast, y = var_11513_to_fp16)[name = tensor<string, []>("aw_chunk_1045_cast")];
            tensor<string, []> var_11516_equation_0 = const()[name = tensor<string, []>("op_11516_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11516_cast = einsum(equation = var_11516_equation_0, values = (var_11302_cast, var_11257_cast))[name = tensor<string, []>("op_11516_cast")];
            tensor<fp16, []> var_11517_to_fp16 = const()[name = tensor<string, []>("op_11517_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1047_cast = mul(x = var_11516_cast, y = var_11517_to_fp16)[name = tensor<string, []>("aw_chunk_1047_cast")];
            tensor<string, []> var_11520_equation_0 = const()[name = tensor<string, []>("op_11520_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11520_cast = einsum(equation = var_11520_equation_0, values = (var_11302_cast, var_11258_cast))[name = tensor<string, []>("op_11520_cast")];
            tensor<fp16, []> var_11521_to_fp16 = const()[name = tensor<string, []>("op_11521_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1049_cast = mul(x = var_11520_cast, y = var_11521_to_fp16)[name = tensor<string, []>("aw_chunk_1049_cast")];
            tensor<string, []> var_11524_equation_0 = const()[name = tensor<string, []>("op_11524_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11524_cast = einsum(equation = var_11524_equation_0, values = (var_11302_cast, var_11259_cast))[name = tensor<string, []>("op_11524_cast")];
            tensor<fp16, []> var_11525_to_fp16 = const()[name = tensor<string, []>("op_11525_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1051_cast = mul(x = var_11524_cast, y = var_11525_to_fp16)[name = tensor<string, []>("aw_chunk_1051_cast")];
            tensor<string, []> var_11528_equation_0 = const()[name = tensor<string, []>("op_11528_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11528_cast = einsum(equation = var_11528_equation_0, values = (var_11302_cast, var_11260_cast))[name = tensor<string, []>("op_11528_cast")];
            tensor<fp16, []> var_11529_to_fp16 = const()[name = tensor<string, []>("op_11529_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1053_cast = mul(x = var_11528_cast, y = var_11529_to_fp16)[name = tensor<string, []>("aw_chunk_1053_cast")];
            tensor<string, []> var_11532_equation_0 = const()[name = tensor<string, []>("op_11532_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11532_cast = einsum(equation = var_11532_equation_0, values = (var_11302_cast, var_11261_cast))[name = tensor<string, []>("op_11532_cast")];
            tensor<fp16, []> var_11533_to_fp16 = const()[name = tensor<string, []>("op_11533_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1055_cast = mul(x = var_11532_cast, y = var_11533_to_fp16)[name = tensor<string, []>("aw_chunk_1055_cast")];
            tensor<string, []> var_11536_equation_0 = const()[name = tensor<string, []>("op_11536_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11536_cast = einsum(equation = var_11536_equation_0, values = (var_11306_cast, var_11262_cast))[name = tensor<string, []>("op_11536_cast")];
            tensor<fp16, []> var_11537_to_fp16 = const()[name = tensor<string, []>("op_11537_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1057_cast = mul(x = var_11536_cast, y = var_11537_to_fp16)[name = tensor<string, []>("aw_chunk_1057_cast")];
            tensor<string, []> var_11540_equation_0 = const()[name = tensor<string, []>("op_11540_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11540_cast = einsum(equation = var_11540_equation_0, values = (var_11306_cast, var_11263_cast))[name = tensor<string, []>("op_11540_cast")];
            tensor<fp16, []> var_11541_to_fp16 = const()[name = tensor<string, []>("op_11541_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1059_cast = mul(x = var_11540_cast, y = var_11541_to_fp16)[name = tensor<string, []>("aw_chunk_1059_cast")];
            tensor<string, []> var_11544_equation_0 = const()[name = tensor<string, []>("op_11544_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11544_cast = einsum(equation = var_11544_equation_0, values = (var_11306_cast, var_11264_cast))[name = tensor<string, []>("op_11544_cast")];
            tensor<fp16, []> var_11545_to_fp16 = const()[name = tensor<string, []>("op_11545_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1061_cast = mul(x = var_11544_cast, y = var_11545_to_fp16)[name = tensor<string, []>("aw_chunk_1061_cast")];
            tensor<string, []> var_11548_equation_0 = const()[name = tensor<string, []>("op_11548_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11548_cast = einsum(equation = var_11548_equation_0, values = (var_11306_cast, var_11265_cast))[name = tensor<string, []>("op_11548_cast")];
            tensor<fp16, []> var_11549_to_fp16 = const()[name = tensor<string, []>("op_11549_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1063_cast = mul(x = var_11548_cast, y = var_11549_to_fp16)[name = tensor<string, []>("aw_chunk_1063_cast")];
            tensor<string, []> var_11552_equation_0 = const()[name = tensor<string, []>("op_11552_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11552_cast = einsum(equation = var_11552_equation_0, values = (var_11306_cast, var_11266_cast))[name = tensor<string, []>("op_11552_cast")];
            tensor<fp16, []> var_11553_to_fp16 = const()[name = tensor<string, []>("op_11553_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1065_cast = mul(x = var_11552_cast, y = var_11553_to_fp16)[name = tensor<string, []>("aw_chunk_1065_cast")];
            tensor<string, []> var_11556_equation_0 = const()[name = tensor<string, []>("op_11556_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11556_cast = einsum(equation = var_11556_equation_0, values = (var_11306_cast, var_11267_cast))[name = tensor<string, []>("op_11556_cast")];
            tensor<fp16, []> var_11557_to_fp16 = const()[name = tensor<string, []>("op_11557_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1067_cast = mul(x = var_11556_cast, y = var_11557_to_fp16)[name = tensor<string, []>("aw_chunk_1067_cast")];
            tensor<string, []> var_11560_equation_0 = const()[name = tensor<string, []>("op_11560_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11560_cast = einsum(equation = var_11560_equation_0, values = (var_11306_cast, var_11268_cast))[name = tensor<string, []>("op_11560_cast")];
            tensor<fp16, []> var_11561_to_fp16 = const()[name = tensor<string, []>("op_11561_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1069_cast = mul(x = var_11560_cast, y = var_11561_to_fp16)[name = tensor<string, []>("aw_chunk_1069_cast")];
            tensor<string, []> var_11564_equation_0 = const()[name = tensor<string, []>("op_11564_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11564_cast = einsum(equation = var_11564_equation_0, values = (var_11306_cast, var_11269_cast))[name = tensor<string, []>("op_11564_cast")];
            tensor<fp16, []> var_11565_to_fp16 = const()[name = tensor<string, []>("op_11565_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1071_cast = mul(x = var_11564_cast, y = var_11565_to_fp16)[name = tensor<string, []>("aw_chunk_1071_cast")];
            tensor<string, []> var_11568_equation_0 = const()[name = tensor<string, []>("op_11568_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11568_cast = einsum(equation = var_11568_equation_0, values = (var_11310_cast, var_11270_cast))[name = tensor<string, []>("op_11568_cast")];
            tensor<fp16, []> var_11569_to_fp16 = const()[name = tensor<string, []>("op_11569_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1073_cast = mul(x = var_11568_cast, y = var_11569_to_fp16)[name = tensor<string, []>("aw_chunk_1073_cast")];
            tensor<string, []> var_11572_equation_0 = const()[name = tensor<string, []>("op_11572_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11572_cast = einsum(equation = var_11572_equation_0, values = (var_11310_cast, var_11271_cast))[name = tensor<string, []>("op_11572_cast")];
            tensor<fp16, []> var_11573_to_fp16 = const()[name = tensor<string, []>("op_11573_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1075_cast = mul(x = var_11572_cast, y = var_11573_to_fp16)[name = tensor<string, []>("aw_chunk_1075_cast")];
            tensor<string, []> var_11576_equation_0 = const()[name = tensor<string, []>("op_11576_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11576_cast = einsum(equation = var_11576_equation_0, values = (var_11310_cast, var_11272_cast))[name = tensor<string, []>("op_11576_cast")];
            tensor<fp16, []> var_11577_to_fp16 = const()[name = tensor<string, []>("op_11577_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1077_cast = mul(x = var_11576_cast, y = var_11577_to_fp16)[name = tensor<string, []>("aw_chunk_1077_cast")];
            tensor<string, []> var_11580_equation_0 = const()[name = tensor<string, []>("op_11580_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11580_cast = einsum(equation = var_11580_equation_0, values = (var_11310_cast, var_11273_cast))[name = tensor<string, []>("op_11580_cast")];
            tensor<fp16, []> var_11581_to_fp16 = const()[name = tensor<string, []>("op_11581_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1079_cast = mul(x = var_11580_cast, y = var_11581_to_fp16)[name = tensor<string, []>("aw_chunk_1079_cast")];
            tensor<string, []> var_11584_equation_0 = const()[name = tensor<string, []>("op_11584_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11584_cast = einsum(equation = var_11584_equation_0, values = (var_11310_cast, var_11274_cast))[name = tensor<string, []>("op_11584_cast")];
            tensor<fp16, []> var_11585_to_fp16 = const()[name = tensor<string, []>("op_11585_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1081_cast = mul(x = var_11584_cast, y = var_11585_to_fp16)[name = tensor<string, []>("aw_chunk_1081_cast")];
            tensor<string, []> var_11588_equation_0 = const()[name = tensor<string, []>("op_11588_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11588_cast = einsum(equation = var_11588_equation_0, values = (var_11310_cast, var_11275_cast))[name = tensor<string, []>("op_11588_cast")];
            tensor<fp16, []> var_11589_to_fp16 = const()[name = tensor<string, []>("op_11589_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1083_cast = mul(x = var_11588_cast, y = var_11589_to_fp16)[name = tensor<string, []>("aw_chunk_1083_cast")];
            tensor<string, []> var_11592_equation_0 = const()[name = tensor<string, []>("op_11592_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11592_cast = einsum(equation = var_11592_equation_0, values = (var_11310_cast, var_11276_cast))[name = tensor<string, []>("op_11592_cast")];
            tensor<fp16, []> var_11593_to_fp16 = const()[name = tensor<string, []>("op_11593_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1085_cast = mul(x = var_11592_cast, y = var_11593_to_fp16)[name = tensor<string, []>("aw_chunk_1085_cast")];
            tensor<string, []> var_11596_equation_0 = const()[name = tensor<string, []>("op_11596_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11596_cast = einsum(equation = var_11596_equation_0, values = (var_11310_cast, var_11277_cast))[name = tensor<string, []>("op_11596_cast")];
            tensor<fp16, []> var_11597_to_fp16 = const()[name = tensor<string, []>("op_11597_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1087_cast = mul(x = var_11596_cast, y = var_11597_to_fp16)[name = tensor<string, []>("aw_chunk_1087_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11599_cast = softmax(axis = var_10375, x = aw_chunk_961_cast)[name = tensor<string, []>("op_11599_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11600_cast = softmax(axis = var_10375, x = aw_chunk_963_cast)[name = tensor<string, []>("op_11600_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11601_cast = softmax(axis = var_10375, x = aw_chunk_965_cast)[name = tensor<string, []>("op_11601_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11602_cast = softmax(axis = var_10375, x = aw_chunk_967_cast)[name = tensor<string, []>("op_11602_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11603_cast = softmax(axis = var_10375, x = aw_chunk_969_cast)[name = tensor<string, []>("op_11603_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11604_cast = softmax(axis = var_10375, x = aw_chunk_971_cast)[name = tensor<string, []>("op_11604_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11605_cast = softmax(axis = var_10375, x = aw_chunk_973_cast)[name = tensor<string, []>("op_11605_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11606_cast = softmax(axis = var_10375, x = aw_chunk_975_cast)[name = tensor<string, []>("op_11606_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11607_cast = softmax(axis = var_10375, x = aw_chunk_977_cast)[name = tensor<string, []>("op_11607_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11608_cast = softmax(axis = var_10375, x = aw_chunk_979_cast)[name = tensor<string, []>("op_11608_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11609_cast = softmax(axis = var_10375, x = aw_chunk_981_cast)[name = tensor<string, []>("op_11609_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11610_cast = softmax(axis = var_10375, x = aw_chunk_983_cast)[name = tensor<string, []>("op_11610_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11611_cast = softmax(axis = var_10375, x = aw_chunk_985_cast)[name = tensor<string, []>("op_11611_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11612_cast = softmax(axis = var_10375, x = aw_chunk_987_cast)[name = tensor<string, []>("op_11612_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11613_cast = softmax(axis = var_10375, x = aw_chunk_989_cast)[name = tensor<string, []>("op_11613_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11614_cast = softmax(axis = var_10375, x = aw_chunk_991_cast)[name = tensor<string, []>("op_11614_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11615_cast = softmax(axis = var_10375, x = aw_chunk_993_cast)[name = tensor<string, []>("op_11615_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11616_cast = softmax(axis = var_10375, x = aw_chunk_995_cast)[name = tensor<string, []>("op_11616_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11617_cast = softmax(axis = var_10375, x = aw_chunk_997_cast)[name = tensor<string, []>("op_11617_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11618_cast = softmax(axis = var_10375, x = aw_chunk_999_cast)[name = tensor<string, []>("op_11618_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11619_cast = softmax(axis = var_10375, x = aw_chunk_1001_cast)[name = tensor<string, []>("op_11619_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11620_cast = softmax(axis = var_10375, x = aw_chunk_1003_cast)[name = tensor<string, []>("op_11620_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11621_cast = softmax(axis = var_10375, x = aw_chunk_1005_cast)[name = tensor<string, []>("op_11621_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11622_cast = softmax(axis = var_10375, x = aw_chunk_1007_cast)[name = tensor<string, []>("op_11622_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11623_cast = softmax(axis = var_10375, x = aw_chunk_1009_cast)[name = tensor<string, []>("op_11623_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11624_cast = softmax(axis = var_10375, x = aw_chunk_1011_cast)[name = tensor<string, []>("op_11624_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11625_cast = softmax(axis = var_10375, x = aw_chunk_1013_cast)[name = tensor<string, []>("op_11625_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11626_cast = softmax(axis = var_10375, x = aw_chunk_1015_cast)[name = tensor<string, []>("op_11626_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11627_cast = softmax(axis = var_10375, x = aw_chunk_1017_cast)[name = tensor<string, []>("op_11627_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11628_cast = softmax(axis = var_10375, x = aw_chunk_1019_cast)[name = tensor<string, []>("op_11628_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11629_cast = softmax(axis = var_10375, x = aw_chunk_1021_cast)[name = tensor<string, []>("op_11629_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11630_cast = softmax(axis = var_10375, x = aw_chunk_1023_cast)[name = tensor<string, []>("op_11630_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11631_cast = softmax(axis = var_10375, x = aw_chunk_1025_cast)[name = tensor<string, []>("op_11631_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11632_cast = softmax(axis = var_10375, x = aw_chunk_1027_cast)[name = tensor<string, []>("op_11632_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11633_cast = softmax(axis = var_10375, x = aw_chunk_1029_cast)[name = tensor<string, []>("op_11633_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11634_cast = softmax(axis = var_10375, x = aw_chunk_1031_cast)[name = tensor<string, []>("op_11634_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11635_cast = softmax(axis = var_10375, x = aw_chunk_1033_cast)[name = tensor<string, []>("op_11635_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11636_cast = softmax(axis = var_10375, x = aw_chunk_1035_cast)[name = tensor<string, []>("op_11636_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11637_cast = softmax(axis = var_10375, x = aw_chunk_1037_cast)[name = tensor<string, []>("op_11637_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11638_cast = softmax(axis = var_10375, x = aw_chunk_1039_cast)[name = tensor<string, []>("op_11638_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11639_cast = softmax(axis = var_10375, x = aw_chunk_1041_cast)[name = tensor<string, []>("op_11639_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11640_cast = softmax(axis = var_10375, x = aw_chunk_1043_cast)[name = tensor<string, []>("op_11640_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11641_cast = softmax(axis = var_10375, x = aw_chunk_1045_cast)[name = tensor<string, []>("op_11641_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11642_cast = softmax(axis = var_10375, x = aw_chunk_1047_cast)[name = tensor<string, []>("op_11642_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11643_cast = softmax(axis = var_10375, x = aw_chunk_1049_cast)[name = tensor<string, []>("op_11643_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11644_cast = softmax(axis = var_10375, x = aw_chunk_1051_cast)[name = tensor<string, []>("op_11644_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11645_cast = softmax(axis = var_10375, x = aw_chunk_1053_cast)[name = tensor<string, []>("op_11645_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11646_cast = softmax(axis = var_10375, x = aw_chunk_1055_cast)[name = tensor<string, []>("op_11646_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11647_cast = softmax(axis = var_10375, x = aw_chunk_1057_cast)[name = tensor<string, []>("op_11647_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11648_cast = softmax(axis = var_10375, x = aw_chunk_1059_cast)[name = tensor<string, []>("op_11648_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11649_cast = softmax(axis = var_10375, x = aw_chunk_1061_cast)[name = tensor<string, []>("op_11649_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11650_cast = softmax(axis = var_10375, x = aw_chunk_1063_cast)[name = tensor<string, []>("op_11650_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11651_cast = softmax(axis = var_10375, x = aw_chunk_1065_cast)[name = tensor<string, []>("op_11651_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11652_cast = softmax(axis = var_10375, x = aw_chunk_1067_cast)[name = tensor<string, []>("op_11652_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11653_cast = softmax(axis = var_10375, x = aw_chunk_1069_cast)[name = tensor<string, []>("op_11653_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11654_cast = softmax(axis = var_10375, x = aw_chunk_1071_cast)[name = tensor<string, []>("op_11654_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11655_cast = softmax(axis = var_10375, x = aw_chunk_1073_cast)[name = tensor<string, []>("op_11655_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11656_cast = softmax(axis = var_10375, x = aw_chunk_1075_cast)[name = tensor<string, []>("op_11656_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11657_cast = softmax(axis = var_10375, x = aw_chunk_1077_cast)[name = tensor<string, []>("op_11657_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11658_cast = softmax(axis = var_10375, x = aw_chunk_1079_cast)[name = tensor<string, []>("op_11658_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11659_cast = softmax(axis = var_10375, x = aw_chunk_1081_cast)[name = tensor<string, []>("op_11659_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11660_cast = softmax(axis = var_10375, x = aw_chunk_1083_cast)[name = tensor<string, []>("op_11660_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11661_cast = softmax(axis = var_10375, x = aw_chunk_1085_cast)[name = tensor<string, []>("op_11661_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11662_cast = softmax(axis = var_10375, x = aw_chunk_1087_cast)[name = tensor<string, []>("op_11662_cast")];
            tensor<string, []> var_11664_equation_0 = const()[name = tensor<string, []>("op_11664_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11664_cast = einsum(equation = var_11664_equation_0, values = (var_11312_cast, var_11599_cast))[name = tensor<string, []>("op_11664_cast")];
            tensor<string, []> var_11666_equation_0 = const()[name = tensor<string, []>("op_11666_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11666_cast = einsum(equation = var_11666_equation_0, values = (var_11312_cast, var_11600_cast))[name = tensor<string, []>("op_11666_cast")];
            tensor<string, []> var_11668_equation_0 = const()[name = tensor<string, []>("op_11668_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11668_cast = einsum(equation = var_11668_equation_0, values = (var_11312_cast, var_11601_cast))[name = tensor<string, []>("op_11668_cast")];
            tensor<string, []> var_11670_equation_0 = const()[name = tensor<string, []>("op_11670_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11670_cast = einsum(equation = var_11670_equation_0, values = (var_11312_cast, var_11602_cast))[name = tensor<string, []>("op_11670_cast")];
            tensor<string, []> var_11672_equation_0 = const()[name = tensor<string, []>("op_11672_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11672_cast = einsum(equation = var_11672_equation_0, values = (var_11312_cast, var_11603_cast))[name = tensor<string, []>("op_11672_cast")];
            tensor<string, []> var_11674_equation_0 = const()[name = tensor<string, []>("op_11674_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11674_cast = einsum(equation = var_11674_equation_0, values = (var_11312_cast, var_11604_cast))[name = tensor<string, []>("op_11674_cast")];
            tensor<string, []> var_11676_equation_0 = const()[name = tensor<string, []>("op_11676_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11676_cast = einsum(equation = var_11676_equation_0, values = (var_11312_cast, var_11605_cast))[name = tensor<string, []>("op_11676_cast")];
            tensor<string, []> var_11678_equation_0 = const()[name = tensor<string, []>("op_11678_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11678_cast = einsum(equation = var_11678_equation_0, values = (var_11312_cast, var_11606_cast))[name = tensor<string, []>("op_11678_cast")];
            tensor<string, []> var_11680_equation_0 = const()[name = tensor<string, []>("op_11680_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11680_cast = einsum(equation = var_11680_equation_0, values = (var_11316_cast, var_11607_cast))[name = tensor<string, []>("op_11680_cast")];
            tensor<string, []> var_11682_equation_0 = const()[name = tensor<string, []>("op_11682_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11682_cast = einsum(equation = var_11682_equation_0, values = (var_11316_cast, var_11608_cast))[name = tensor<string, []>("op_11682_cast")];
            tensor<string, []> var_11684_equation_0 = const()[name = tensor<string, []>("op_11684_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11684_cast = einsum(equation = var_11684_equation_0, values = (var_11316_cast, var_11609_cast))[name = tensor<string, []>("op_11684_cast")];
            tensor<string, []> var_11686_equation_0 = const()[name = tensor<string, []>("op_11686_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11686_cast = einsum(equation = var_11686_equation_0, values = (var_11316_cast, var_11610_cast))[name = tensor<string, []>("op_11686_cast")];
            tensor<string, []> var_11688_equation_0 = const()[name = tensor<string, []>("op_11688_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11688_cast = einsum(equation = var_11688_equation_0, values = (var_11316_cast, var_11611_cast))[name = tensor<string, []>("op_11688_cast")];
            tensor<string, []> var_11690_equation_0 = const()[name = tensor<string, []>("op_11690_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11690_cast = einsum(equation = var_11690_equation_0, values = (var_11316_cast, var_11612_cast))[name = tensor<string, []>("op_11690_cast")];
            tensor<string, []> var_11692_equation_0 = const()[name = tensor<string, []>("op_11692_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11692_cast = einsum(equation = var_11692_equation_0, values = (var_11316_cast, var_11613_cast))[name = tensor<string, []>("op_11692_cast")];
            tensor<string, []> var_11694_equation_0 = const()[name = tensor<string, []>("op_11694_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11694_cast = einsum(equation = var_11694_equation_0, values = (var_11316_cast, var_11614_cast))[name = tensor<string, []>("op_11694_cast")];
            tensor<string, []> var_11696_equation_0 = const()[name = tensor<string, []>("op_11696_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11696_cast = einsum(equation = var_11696_equation_0, values = (var_11320_cast, var_11615_cast))[name = tensor<string, []>("op_11696_cast")];
            tensor<string, []> var_11698_equation_0 = const()[name = tensor<string, []>("op_11698_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11698_cast = einsum(equation = var_11698_equation_0, values = (var_11320_cast, var_11616_cast))[name = tensor<string, []>("op_11698_cast")];
            tensor<string, []> var_11700_equation_0 = const()[name = tensor<string, []>("op_11700_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11700_cast = einsum(equation = var_11700_equation_0, values = (var_11320_cast, var_11617_cast))[name = tensor<string, []>("op_11700_cast")];
            tensor<string, []> var_11702_equation_0 = const()[name = tensor<string, []>("op_11702_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11702_cast = einsum(equation = var_11702_equation_0, values = (var_11320_cast, var_11618_cast))[name = tensor<string, []>("op_11702_cast")];
            tensor<string, []> var_11704_equation_0 = const()[name = tensor<string, []>("op_11704_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11704_cast = einsum(equation = var_11704_equation_0, values = (var_11320_cast, var_11619_cast))[name = tensor<string, []>("op_11704_cast")];
            tensor<string, []> var_11706_equation_0 = const()[name = tensor<string, []>("op_11706_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11706_cast = einsum(equation = var_11706_equation_0, values = (var_11320_cast, var_11620_cast))[name = tensor<string, []>("op_11706_cast")];
            tensor<string, []> var_11708_equation_0 = const()[name = tensor<string, []>("op_11708_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11708_cast = einsum(equation = var_11708_equation_0, values = (var_11320_cast, var_11621_cast))[name = tensor<string, []>("op_11708_cast")];
            tensor<string, []> var_11710_equation_0 = const()[name = tensor<string, []>("op_11710_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11710_cast = einsum(equation = var_11710_equation_0, values = (var_11320_cast, var_11622_cast))[name = tensor<string, []>("op_11710_cast")];
            tensor<string, []> var_11712_equation_0 = const()[name = tensor<string, []>("op_11712_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11712_cast = einsum(equation = var_11712_equation_0, values = (var_11324_cast, var_11623_cast))[name = tensor<string, []>("op_11712_cast")];
            tensor<string, []> var_11714_equation_0 = const()[name = tensor<string, []>("op_11714_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11714_cast = einsum(equation = var_11714_equation_0, values = (var_11324_cast, var_11624_cast))[name = tensor<string, []>("op_11714_cast")];
            tensor<string, []> var_11716_equation_0 = const()[name = tensor<string, []>("op_11716_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11716_cast = einsum(equation = var_11716_equation_0, values = (var_11324_cast, var_11625_cast))[name = tensor<string, []>("op_11716_cast")];
            tensor<string, []> var_11718_equation_0 = const()[name = tensor<string, []>("op_11718_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11718_cast = einsum(equation = var_11718_equation_0, values = (var_11324_cast, var_11626_cast))[name = tensor<string, []>("op_11718_cast")];
            tensor<string, []> var_11720_equation_0 = const()[name = tensor<string, []>("op_11720_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11720_cast = einsum(equation = var_11720_equation_0, values = (var_11324_cast, var_11627_cast))[name = tensor<string, []>("op_11720_cast")];
            tensor<string, []> var_11722_equation_0 = const()[name = tensor<string, []>("op_11722_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11722_cast = einsum(equation = var_11722_equation_0, values = (var_11324_cast, var_11628_cast))[name = tensor<string, []>("op_11722_cast")];
            tensor<string, []> var_11724_equation_0 = const()[name = tensor<string, []>("op_11724_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11724_cast = einsum(equation = var_11724_equation_0, values = (var_11324_cast, var_11629_cast))[name = tensor<string, []>("op_11724_cast")];
            tensor<string, []> var_11726_equation_0 = const()[name = tensor<string, []>("op_11726_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11726_cast = einsum(equation = var_11726_equation_0, values = (var_11324_cast, var_11630_cast))[name = tensor<string, []>("op_11726_cast")];
            tensor<string, []> var_11728_equation_0 = const()[name = tensor<string, []>("op_11728_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11728_cast = einsum(equation = var_11728_equation_0, values = (var_11328_cast, var_11631_cast))[name = tensor<string, []>("op_11728_cast")];
            tensor<string, []> var_11730_equation_0 = const()[name = tensor<string, []>("op_11730_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11730_cast = einsum(equation = var_11730_equation_0, values = (var_11328_cast, var_11632_cast))[name = tensor<string, []>("op_11730_cast")];
            tensor<string, []> var_11732_equation_0 = const()[name = tensor<string, []>("op_11732_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11732_cast = einsum(equation = var_11732_equation_0, values = (var_11328_cast, var_11633_cast))[name = tensor<string, []>("op_11732_cast")];
            tensor<string, []> var_11734_equation_0 = const()[name = tensor<string, []>("op_11734_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11734_cast = einsum(equation = var_11734_equation_0, values = (var_11328_cast, var_11634_cast))[name = tensor<string, []>("op_11734_cast")];
            tensor<string, []> var_11736_equation_0 = const()[name = tensor<string, []>("op_11736_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11736_cast = einsum(equation = var_11736_equation_0, values = (var_11328_cast, var_11635_cast))[name = tensor<string, []>("op_11736_cast")];
            tensor<string, []> var_11738_equation_0 = const()[name = tensor<string, []>("op_11738_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11738_cast = einsum(equation = var_11738_equation_0, values = (var_11328_cast, var_11636_cast))[name = tensor<string, []>("op_11738_cast")];
            tensor<string, []> var_11740_equation_0 = const()[name = tensor<string, []>("op_11740_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11740_cast = einsum(equation = var_11740_equation_0, values = (var_11328_cast, var_11637_cast))[name = tensor<string, []>("op_11740_cast")];
            tensor<string, []> var_11742_equation_0 = const()[name = tensor<string, []>("op_11742_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11742_cast = einsum(equation = var_11742_equation_0, values = (var_11328_cast, var_11638_cast))[name = tensor<string, []>("op_11742_cast")];
            tensor<string, []> var_11744_equation_0 = const()[name = tensor<string, []>("op_11744_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11744_cast = einsum(equation = var_11744_equation_0, values = (var_11332_cast, var_11639_cast))[name = tensor<string, []>("op_11744_cast")];
            tensor<string, []> var_11746_equation_0 = const()[name = tensor<string, []>("op_11746_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11746_cast = einsum(equation = var_11746_equation_0, values = (var_11332_cast, var_11640_cast))[name = tensor<string, []>("op_11746_cast")];
            tensor<string, []> var_11748_equation_0 = const()[name = tensor<string, []>("op_11748_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11748_cast = einsum(equation = var_11748_equation_0, values = (var_11332_cast, var_11641_cast))[name = tensor<string, []>("op_11748_cast")];
            tensor<string, []> var_11750_equation_0 = const()[name = tensor<string, []>("op_11750_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11750_cast = einsum(equation = var_11750_equation_0, values = (var_11332_cast, var_11642_cast))[name = tensor<string, []>("op_11750_cast")];
            tensor<string, []> var_11752_equation_0 = const()[name = tensor<string, []>("op_11752_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11752_cast = einsum(equation = var_11752_equation_0, values = (var_11332_cast, var_11643_cast))[name = tensor<string, []>("op_11752_cast")];
            tensor<string, []> var_11754_equation_0 = const()[name = tensor<string, []>("op_11754_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11754_cast = einsum(equation = var_11754_equation_0, values = (var_11332_cast, var_11644_cast))[name = tensor<string, []>("op_11754_cast")];
            tensor<string, []> var_11756_equation_0 = const()[name = tensor<string, []>("op_11756_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11756_cast = einsum(equation = var_11756_equation_0, values = (var_11332_cast, var_11645_cast))[name = tensor<string, []>("op_11756_cast")];
            tensor<string, []> var_11758_equation_0 = const()[name = tensor<string, []>("op_11758_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11758_cast = einsum(equation = var_11758_equation_0, values = (var_11332_cast, var_11646_cast))[name = tensor<string, []>("op_11758_cast")];
            tensor<string, []> var_11760_equation_0 = const()[name = tensor<string, []>("op_11760_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11760_cast = einsum(equation = var_11760_equation_0, values = (var_11336_cast, var_11647_cast))[name = tensor<string, []>("op_11760_cast")];
            tensor<string, []> var_11762_equation_0 = const()[name = tensor<string, []>("op_11762_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11762_cast = einsum(equation = var_11762_equation_0, values = (var_11336_cast, var_11648_cast))[name = tensor<string, []>("op_11762_cast")];
            tensor<string, []> var_11764_equation_0 = const()[name = tensor<string, []>("op_11764_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11764_cast = einsum(equation = var_11764_equation_0, values = (var_11336_cast, var_11649_cast))[name = tensor<string, []>("op_11764_cast")];
            tensor<string, []> var_11766_equation_0 = const()[name = tensor<string, []>("op_11766_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11766_cast = einsum(equation = var_11766_equation_0, values = (var_11336_cast, var_11650_cast))[name = tensor<string, []>("op_11766_cast")];
            tensor<string, []> var_11768_equation_0 = const()[name = tensor<string, []>("op_11768_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11768_cast = einsum(equation = var_11768_equation_0, values = (var_11336_cast, var_11651_cast))[name = tensor<string, []>("op_11768_cast")];
            tensor<string, []> var_11770_equation_0 = const()[name = tensor<string, []>("op_11770_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11770_cast = einsum(equation = var_11770_equation_0, values = (var_11336_cast, var_11652_cast))[name = tensor<string, []>("op_11770_cast")];
            tensor<string, []> var_11772_equation_0 = const()[name = tensor<string, []>("op_11772_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11772_cast = einsum(equation = var_11772_equation_0, values = (var_11336_cast, var_11653_cast))[name = tensor<string, []>("op_11772_cast")];
            tensor<string, []> var_11774_equation_0 = const()[name = tensor<string, []>("op_11774_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11774_cast = einsum(equation = var_11774_equation_0, values = (var_11336_cast, var_11654_cast))[name = tensor<string, []>("op_11774_cast")];
            tensor<string, []> var_11776_equation_0 = const()[name = tensor<string, []>("op_11776_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11776_cast = einsum(equation = var_11776_equation_0, values = (var_11340_cast, var_11655_cast))[name = tensor<string, []>("op_11776_cast")];
            tensor<string, []> var_11778_equation_0 = const()[name = tensor<string, []>("op_11778_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11778_cast = einsum(equation = var_11778_equation_0, values = (var_11340_cast, var_11656_cast))[name = tensor<string, []>("op_11778_cast")];
            tensor<string, []> var_11780_equation_0 = const()[name = tensor<string, []>("op_11780_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11780_cast = einsum(equation = var_11780_equation_0, values = (var_11340_cast, var_11657_cast))[name = tensor<string, []>("op_11780_cast")];
            tensor<string, []> var_11782_equation_0 = const()[name = tensor<string, []>("op_11782_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11782_cast = einsum(equation = var_11782_equation_0, values = (var_11340_cast, var_11658_cast))[name = tensor<string, []>("op_11782_cast")];
            tensor<string, []> var_11784_equation_0 = const()[name = tensor<string, []>("op_11784_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11784_cast = einsum(equation = var_11784_equation_0, values = (var_11340_cast, var_11659_cast))[name = tensor<string, []>("op_11784_cast")];
            tensor<string, []> var_11786_equation_0 = const()[name = tensor<string, []>("op_11786_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11786_cast = einsum(equation = var_11786_equation_0, values = (var_11340_cast, var_11660_cast))[name = tensor<string, []>("op_11786_cast")];
            tensor<string, []> var_11788_equation_0 = const()[name = tensor<string, []>("op_11788_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11788_cast = einsum(equation = var_11788_equation_0, values = (var_11340_cast, var_11661_cast))[name = tensor<string, []>("op_11788_cast")];
            tensor<string, []> var_11790_equation_0 = const()[name = tensor<string, []>("op_11790_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_11790_cast = einsum(equation = var_11790_equation_0, values = (var_11340_cast, var_11662_cast))[name = tensor<string, []>("op_11790_cast")];
            tensor<bool, []> var_11792_interleave_0 = const()[name = tensor<string, []>("op_11792_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_11792_cast = concat(axis = var_10347, interleave = var_11792_interleave_0, values = (var_11664_cast, var_11666_cast, var_11668_cast, var_11670_cast, var_11672_cast, var_11674_cast, var_11676_cast, var_11678_cast))[name = tensor<string, []>("op_11792_cast")];
            tensor<bool, []> var_11794_interleave_0 = const()[name = tensor<string, []>("op_11794_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_11794_cast = concat(axis = var_10347, interleave = var_11794_interleave_0, values = (var_11680_cast, var_11682_cast, var_11684_cast, var_11686_cast, var_11688_cast, var_11690_cast, var_11692_cast, var_11694_cast))[name = tensor<string, []>("op_11794_cast")];
            tensor<bool, []> var_11796_interleave_0 = const()[name = tensor<string, []>("op_11796_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_11796_cast = concat(axis = var_10347, interleave = var_11796_interleave_0, values = (var_11696_cast, var_11698_cast, var_11700_cast, var_11702_cast, var_11704_cast, var_11706_cast, var_11708_cast, var_11710_cast))[name = tensor<string, []>("op_11796_cast")];
            tensor<bool, []> var_11798_interleave_0 = const()[name = tensor<string, []>("op_11798_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_11798_cast = concat(axis = var_10347, interleave = var_11798_interleave_0, values = (var_11712_cast, var_11714_cast, var_11716_cast, var_11718_cast, var_11720_cast, var_11722_cast, var_11724_cast, var_11726_cast))[name = tensor<string, []>("op_11798_cast")];
            tensor<bool, []> var_11800_interleave_0 = const()[name = tensor<string, []>("op_11800_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_11800_cast = concat(axis = var_10347, interleave = var_11800_interleave_0, values = (var_11728_cast, var_11730_cast, var_11732_cast, var_11734_cast, var_11736_cast, var_11738_cast, var_11740_cast, var_11742_cast))[name = tensor<string, []>("op_11800_cast")];
            tensor<bool, []> var_11802_interleave_0 = const()[name = tensor<string, []>("op_11802_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_11802_cast = concat(axis = var_10347, interleave = var_11802_interleave_0, values = (var_11744_cast, var_11746_cast, var_11748_cast, var_11750_cast, var_11752_cast, var_11754_cast, var_11756_cast, var_11758_cast))[name = tensor<string, []>("op_11802_cast")];
            tensor<bool, []> var_11804_interleave_0 = const()[name = tensor<string, []>("op_11804_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_11804_cast = concat(axis = var_10347, interleave = var_11804_interleave_0, values = (var_11760_cast, var_11762_cast, var_11764_cast, var_11766_cast, var_11768_cast, var_11770_cast, var_11772_cast, var_11774_cast))[name = tensor<string, []>("op_11804_cast")];
            tensor<bool, []> var_11806_interleave_0 = const()[name = tensor<string, []>("op_11806_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_11806_cast = concat(axis = var_10347, interleave = var_11806_interleave_0, values = (var_11776_cast, var_11778_cast, var_11780_cast, var_11782_cast, var_11784_cast, var_11786_cast, var_11788_cast, var_11790_cast))[name = tensor<string, []>("op_11806_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_10375, interleave = input_467_interleave_0, values = (var_11792_cast, var_11794_cast, var_11796_cast, var_11798_cast, var_11800_cast, var_11802_cast, var_11804_cast, var_11806_cast))[name = tensor<string, []>("input_467_cast")];
            tensor<int32, [2]> var_11812 = const()[name = tensor<string, []>("op_11812"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11814 = const()[name = tensor<string, []>("op_11814"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_11816_pad_type_0 = const()[name = tensor<string, []>("op_11816_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_11816_pad_0 = const()[name = tensor<string, []>("op_11816_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(635152640))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(635229504))), name = tensor<string, []>("up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            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, []>(635229696)))];
            tensor<fp16, [2, 320, 1, 4096]> var_11816_cast = conv(bias = up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_11814, groups = var_10375, pad = var_11816_pad_0, pad_type = var_11816_pad_type_0, strides = var_11812, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_467_cast)[name = tensor<string, []>("op_11816_cast")];
            tensor<fp16, [2, 320, 1, 4096]> inputs_83_cast = add(x = var_11816_cast, y = inputs_81_cast)[name = tensor<string, []>("inputs_83_cast")];
            tensor<int32, [1]> var_11820 = const()[name = tensor<string, []>("op_11820"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_83_cast = reduce_mean(axes = var_11820, keep_dims = var_10370, x = inputs_83_cast)[name = tensor<string, []>("channels_mean_83_cast")];
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_83_cast = sub(x = inputs_83_cast, y = channels_mean_83_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_83_cast")];
            tensor<int32, [1]> var_11824 = const()[name = tensor<string, []>("op_11824"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_11825_cast = reduce_mean(axes = var_11824, keep_dims = var_10370, x = zero_mean_sq_83_cast)[name = tensor<string, []>("op_11825_cast")];
            tensor<fp16, []> var_11826_to_fp16 = const()[name = tensor<string, []>("op_11826_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_11827_cast = add(x = var_11825_cast, y = var_11826_to_fp16)[name = tensor<string, []>("op_11827_cast")];
            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_11827_cast)[name = tensor<string, []>("denom_83_cast")];
            tensor<fp16, [2, 320, 1, 4096]> out_83_cast = mul(x = zero_mean_83_cast, y = denom_83_cast)[name = tensor<string, []>("out_83_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_11831_to_fp16 = const()[name = tensor<string, []>("op_11831_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(635230400)))];
            tensor<fp16, [2, 320, 1, 4096]> var_11832_cast = add(x = out_83_cast, y = var_11831_to_fp16)[name = tensor<string, []>("op_11832_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_11834_to_fp16 = const()[name = tensor<string, []>("op_11834_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(635231104)))];
            tensor<fp16, [2, 320, 1, 4096]> input_469_cast = mul(x = var_11832_cast, y = var_11834_to_fp16)[name = tensor<string, []>("input_469_cast")];
            tensor<int32, [2]> var_11842 = const()[name = tensor<string, []>("op_11842"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11844 = const()[name = tensor<string, []>("op_11844"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_11846_pad_type_0 = const()[name = tensor<string, []>("op_11846_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_11846_pad_0 = const()[name = tensor<string, []>("op_11846_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [614400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(635231808))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(635846272))), name = tensor<string, []>("up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([2560, 320, 1, 1])];
            tensor<fp16, [2560]> up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(635846464))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(635848448))), name = tensor<string, []>("up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([2560])];
            tensor<fp16, [2, 2560, 1, 4096]> var_11846_cast = conv(bias = up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_11844, groups = var_10375, pad = var_11846_pad_0, pad_type = var_11846_pad_type_0, strides = var_11842, weight = up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_469_cast)[name = tensor<string, []>("op_11846_cast")];
            tensor<int32, [2]> var_11847_split_sizes_0 = const()[name = tensor<string, []>("op_11847_split_sizes_0"), val = tensor<int32, [2]>([1280, 1280])];
            tensor<int32, []> var_11847_axis_0 = const()[name = tensor<string, []>("op_11847_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 1280, 1, 4096]> var_11847_cast_0, tensor<fp16, [2, 1280, 1, 4096]> var_11847_cast_1 = split(axis = var_11847_axis_0, split_sizes = var_11847_split_sizes_0, x = var_11846_cast)[name = tensor<string, []>("op_11847_cast")];
            tensor<string, []> var_11849_mode_0 = const()[name = tensor<string, []>("op_11849_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 1280, 1, 4096]> var_11849_cast = gelu(mode = var_11849_mode_0, x = var_11847_cast_1)[name = tensor<string, []>("op_11849_cast")];
            tensor<fp16, [2, 1280, 1, 4096]> input_471_cast = mul(x = var_11847_cast_0, y = var_11849_cast)[name = tensor<string, []>("input_471_cast")];
            tensor<int32, [2]> var_11853 = const()[name = tensor<string, []>("op_11853"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11855 = const()[name = tensor<string, []>("op_11855"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_11857_pad_type_0 = const()[name = tensor<string, []>("op_11857_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_11857_pad_0 = const()[name = tensor<string, []>("op_11857_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(635848640))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(636155904))), name = tensor<string, []>("up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 1280, 1, 1])];
            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, []>(636156096)))];
            tensor<fp16, [2, 320, 1, 4096]> var_11857_cast = conv(bias = up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_11855, groups = var_10375, pad = var_11857_pad_0, pad_type = var_11857_pad_type_0, strides = var_11853, weight = up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_471_cast)[name = tensor<string, []>("op_11857_cast")];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_293_cast = add(x = var_11857_cast, y = inputs_83_cast)[name = tensor<string, []>("hidden_states_293_cast")];
            tensor<int32, [4]> var_11859 = const()[name = tensor<string, []>("op_11859"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> input_473_cast = reshape(shape = var_11859, x = hidden_states_293_cast)[name = tensor<string, []>("input_473_cast")];
            tensor<int32, [2]> var_11863 = const()[name = tensor<string, []>("op_11863"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11865 = const()[name = tensor<string, []>("op_11865"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(636156800))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(636233664))), name = tensor<string, []>("up_blocks_3_attentions_0_proj_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            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, []>(636233856)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_295_cast = conv(bias = up_blocks_3_attentions_0_proj_out_bias_to_fp16, dilations = var_11865, groups = var_10375, pad = hidden_states_295_pad_0, pad_type = hidden_states_295_pad_type_0, strides = var_11863, weight = up_blocks_3_attentions_0_proj_out_weight_to_fp16_palettized, x = input_473_cast)[name = tensor<string, []>("hidden_states_295_cast")];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_297_cast = add(x = hidden_states_295_cast, y = hidden_states_283_cast)[name = tensor<string, []>("hidden_states_297_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_10375, interleave = input_475_interleave_0, values = (hidden_states_297_cast, input_35_cast))[name = tensor<string, []>("input_475_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)[name = tensor<string, []>("reshape_216_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)[name = tensor<string, []>("reduce_mean_162_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> sub_108_cast = sub(x = reshape_216_cast, y = reduce_mean_162_cast)[name = tensor<string, []>("sub_108_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> square_54_cast = square(x = sub_108_cast)[name = tensor<string, []>("square_54_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)[name = tensor<string, []>("reduce_mean_164_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)[name = tensor<string, []>("add_108_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_54_cast = sqrt(x = add_108_cast)[name = tensor<string, []>("sqrt_54_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> real_div_54_cast = real_div(x = sub_108_cast, y = sqrt_54_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_217_cast")];
            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, []>(636234560)))];
            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, []>(636235904)))];
            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_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_217_cast)[name = tensor<string, []>("add_109_cast")];
            tensor<fp16, [2, 640, 64, 64]> input_479_cast = silu(x = add_109_cast)[name = tensor<string, []>("input_479_cast")];
            tensor<int32, [2]> var_11883 = const()[name = tensor<string, []>("op_11883"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11885 = const()[name = tensor<string, []>("op_11885"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1382400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(636237248))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(637619712))), name = tensor<string, []>("up_blocks_3_resnets_1_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 640, 3, 3])];
            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, []>(637619904)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_299_cast = conv(bias = up_blocks_3_resnets_1_conv1_bias_to_fp16, dilations = var_11885, groups = var_10375, pad = hidden_states_299_pad_0, pad_type = hidden_states_299_pad_type_0, strides = var_11883, weight = up_blocks_3_resnets_1_conv1_weight_to_fp16_palettized, x = input_479_cast)[name = tensor<string, []>("hidden_states_299_cast")];
            tensor<int32, [2]> var_11891 = const()[name = tensor<string, []>("op_11891"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11893 = const()[name = tensor<string, []>("op_11893"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(637620608))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(637927872))), name = tensor<string, []>("up_blocks_3_resnets_1_time_emb_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 1280, 1, 1])];
            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, []>(637928064)))];
            tensor<fp16, [2, 320, 1, 1]> temb_41_cast = conv(bias = up_blocks_3_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_11893, groups = var_10375, pad = temb_41_pad_0, pad_type = temb_41_pad_type_0, strides = var_11891, weight = up_blocks_3_resnets_1_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor<string, []>("temb_41_cast")];
            tensor<fp16, [2, 320, 64, 64]> input_483_cast = add(x = hidden_states_299_cast, y = temb_41_cast)[name = tensor<string, []>("input_483_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)[name = tensor<string, []>("reshape_220_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)[name = tensor<string, []>("reduce_mean_165_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> sub_110_cast = sub(x = reshape_220_cast, y = reduce_mean_165_cast)[name = tensor<string, []>("sub_110_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> square_55_cast = square(x = sub_110_cast)[name = tensor<string, []>("square_55_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)[name = tensor<string, []>("reduce_mean_167_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)[name = tensor<string, []>("add_110_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_55_cast = sqrt(x = add_110_cast)[name = tensor<string, []>("sqrt_55_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> real_div_55_cast = real_div(x = sub_110_cast, y = sqrt_55_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_221_cast")];
            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, []>(637928768)))];
            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, []>(637929472)))];
            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_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_221_cast)[name = tensor<string, []>("add_111_cast")];
            tensor<fp16, [2, 320, 64, 64]> input_487_cast = silu(x = add_111_cast)[name = tensor<string, []>("input_487_cast")];
            tensor<int32, [2]> var_11903 = const()[name = tensor<string, []>("op_11903"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11905 = const()[name = tensor<string, []>("op_11905"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [691200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(637930176))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(638621440))), name = tensor<string, []>("up_blocks_3_resnets_1_conv2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 3, 3])];
            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, []>(638621632)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_301_cast = conv(bias = up_blocks_3_resnets_1_conv2_bias_to_fp16, dilations = var_11905, groups = var_10375, pad = hidden_states_301_pad_0, pad_type = hidden_states_301_pad_type_0, strides = var_11903, weight = up_blocks_3_resnets_1_conv2_weight_to_fp16_palettized, x = input_487_cast)[name = tensor<string, []>("hidden_states_301_cast")];
            tensor<int32, [2]> var_11910 = const()[name = tensor<string, []>("op_11910"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11912 = const()[name = tensor<string, []>("op_11912"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [153600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(638622336))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(638776000))), name = tensor<string, []>("up_blocks_3_resnets_1_conv_shortcut_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 640, 1, 1])];
            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, []>(638776192)))];
            tensor<fp16, [2, 320, 64, 64]> x_25_cast = conv(bias = up_blocks_3_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_11912, groups = var_10375, pad = x_25_pad_0, pad_type = x_25_pad_type_0, strides = var_11910, weight = up_blocks_3_resnets_1_conv_shortcut_weight_to_fp16_palettized, x = input_475_cast)[name = tensor<string, []>("x_25_cast")];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_303_cast = add(x = x_25_cast, y = hidden_states_301_cast)[name = tensor<string, []>("hidden_states_303_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)[name = tensor<string, []>("reshape_224_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)[name = tensor<string, []>("reduce_mean_168_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> sub_112_cast = sub(x = reshape_224_cast, y = reduce_mean_168_cast)[name = tensor<string, []>("sub_112_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> square_56_cast = square(x = sub_112_cast)[name = tensor<string, []>("square_56_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)[name = tensor<string, []>("reduce_mean_170_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)[name = tensor<string, []>("add_112_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_56_cast = sqrt(x = add_112_cast)[name = tensor<string, []>("sqrt_56_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> real_div_56_cast = real_div(x = sub_112_cast, y = sqrt_56_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_225_cast")];
            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, []>(638776896)))];
            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, []>(638777600)))];
            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_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_225_cast)[name = tensor<string, []>("add_113_cast")];
            tensor<int32, [2]> var_11932 = const()[name = tensor<string, []>("op_11932"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11934 = const()[name = tensor<string, []>("op_11934"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(638778304))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(638855168))), name = tensor<string, []>("up_blocks_3_attentions_1_proj_in_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            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, []>(638855360)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_305_cast = conv(bias = up_blocks_3_attentions_1_proj_in_bias_to_fp16, dilations = var_11934, groups = var_10375, pad = hidden_states_305_pad_0, pad_type = hidden_states_305_pad_type_0, strides = var_11932, weight = up_blocks_3_attentions_1_proj_in_weight_to_fp16_palettized, x = add_113_cast)[name = tensor<string, []>("hidden_states_305_cast")];
            tensor<int32, [4]> var_11939 = const()[name = tensor<string, []>("op_11939"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<fp16, [2, 320, 1, 4096]> inputs_85_cast = reshape(shape = var_11939, x = hidden_states_305_cast)[name = tensor<string, []>("inputs_85_cast")];
            tensor<int32, [1]> var_11949 = const()[name = tensor<string, []>("op_11949"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_85_cast = reduce_mean(axes = var_11949, keep_dims = var_10370, x = inputs_85_cast)[name = tensor<string, []>("channels_mean_85_cast")];
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_85_cast = sub(x = inputs_85_cast, y = channels_mean_85_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_85_cast")];
            tensor<int32, [1]> var_11953 = const()[name = tensor<string, []>("op_11953"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_11954_cast = reduce_mean(axes = var_11953, keep_dims = var_10370, x = zero_mean_sq_85_cast)[name = tensor<string, []>("op_11954_cast")];
            tensor<fp16, []> var_11955_to_fp16 = const()[name = tensor<string, []>("op_11955_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_11956_cast = add(x = var_11954_cast, y = var_11955_to_fp16)[name = tensor<string, []>("op_11956_cast")];
            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_11956_cast)[name = tensor<string, []>("denom_85_cast")];
            tensor<fp16, [2, 320, 1, 4096]> out_85_cast = mul(x = zero_mean_85_cast, y = denom_85_cast)[name = tensor<string, []>("out_85_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_11960_to_fp16 = const()[name = tensor<string, []>("op_11960_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(638856064)))];
            tensor<fp16, [2, 320, 1, 4096]> var_11961_cast = add(x = out_85_cast, y = var_11960_to_fp16)[name = tensor<string, []>("op_11961_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_11963_to_fp16 = const()[name = tensor<string, []>("op_11963_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(638856768)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_307_cast = mul(x = var_11961_cast, y = var_11963_to_fp16)[name = tensor<string, []>("hidden_states_307_cast")];
            tensor<int32, [2]> var_11970 = const()[name = tensor<string, []>("op_11970"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11972 = const()[name = tensor<string, []>("op_11972"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(638857472))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(638934336))), name = tensor<string, []>("up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            tensor<fp16, [2, 320, 1, 4096]> q_57_cast = conv(dilations = var_11972, groups = var_10375, pad = q_57_pad_0, pad_type = q_57_pad_type_0, strides = var_11970, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_307_cast)[name = tensor<string, []>("q_57_cast")];
            tensor<int32, [2]> var_11976 = const()[name = tensor<string, []>("op_11976"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11978 = const()[name = tensor<string, []>("op_11978"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(638934528))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(639011392))), name = tensor<string, []>("up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            tensor<fp16, [2, 320, 1, 4096]> k_113_cast = conv(dilations = var_11978, groups = var_10375, pad = k_113_pad_0, pad_type = k_113_pad_type_0, strides = var_11976, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_307_cast)[name = tensor<string, []>("k_113_cast")];
            tensor<int32, [2]> var_11982 = const()[name = tensor<string, []>("op_11982"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11984 = const()[name = tensor<string, []>("op_11984"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(639011584))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(639088448))), name = tensor<string, []>("up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            tensor<fp16, [2, 320, 1, 4096]> v_57_cast = conv(dilations = var_11984, groups = var_10375, pad = v_57_pad_0, pad_type = v_57_pad_type_0, strides = var_11982, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_307_cast)[name = tensor<string, []>("v_57_cast")];
            tensor<int32, [4]> var_11988_begin_0 = const()[name = tensor<string, []>("op_11988_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11988_end_0 = const()[name = tensor<string, []>("op_11988_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_11988_end_mask_0 = const()[name = tensor<string, []>("op_11988_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_11988_cast = slice_by_index(begin = var_11988_begin_0, end = var_11988_end_0, end_mask = var_11988_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_11988_cast")];
            tensor<int32, [4]> var_11992_begin_0 = const()[name = tensor<string, []>("op_11992_begin_0"), val = tensor<int32, [4]>([0, 40, 0, 0])];
            tensor<int32, [4]> var_11992_end_0 = const()[name = tensor<string, []>("op_11992_end_0"), val = tensor<int32, [4]>([2, 80, 1, 4096])];
            tensor<bool, [4]> var_11992_end_mask_0 = const()[name = tensor<string, []>("op_11992_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_11992_cast = slice_by_index(begin = var_11992_begin_0, end = var_11992_end_0, end_mask = var_11992_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_11992_cast")];
            tensor<int32, [4]> var_11996_begin_0 = const()[name = tensor<string, []>("op_11996_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_11996_end_0 = const()[name = tensor<string, []>("op_11996_end_0"), val = tensor<int32, [4]>([2, 120, 1, 4096])];
            tensor<bool, [4]> var_11996_end_mask_0 = const()[name = tensor<string, []>("op_11996_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_11996_cast = slice_by_index(begin = var_11996_begin_0, end = var_11996_end_0, end_mask = var_11996_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_11996_cast")];
            tensor<int32, [4]> var_12000_begin_0 = const()[name = tensor<string, []>("op_12000_begin_0"), val = tensor<int32, [4]>([0, 120, 0, 0])];
            tensor<int32, [4]> var_12000_end_0 = const()[name = tensor<string, []>("op_12000_end_0"), val = tensor<int32, [4]>([2, 160, 1, 4096])];
            tensor<bool, [4]> var_12000_end_mask_0 = const()[name = tensor<string, []>("op_12000_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_12000_cast = slice_by_index(begin = var_12000_begin_0, end = var_12000_end_0, end_mask = var_12000_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_12000_cast")];
            tensor<int32, [4]> var_12004_begin_0 = const()[name = tensor<string, []>("op_12004_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_12004_end_0 = const()[name = tensor<string, []>("op_12004_end_0"), val = tensor<int32, [4]>([2, 200, 1, 4096])];
            tensor<bool, [4]> var_12004_end_mask_0 = const()[name = tensor<string, []>("op_12004_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_12004_cast = slice_by_index(begin = var_12004_begin_0, end = var_12004_end_0, end_mask = var_12004_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_12004_cast")];
            tensor<int32, [4]> var_12008_begin_0 = const()[name = tensor<string, []>("op_12008_begin_0"), val = tensor<int32, [4]>([0, 200, 0, 0])];
            tensor<int32, [4]> var_12008_end_0 = const()[name = tensor<string, []>("op_12008_end_0"), val = tensor<int32, [4]>([2, 240, 1, 4096])];
            tensor<bool, [4]> var_12008_end_mask_0 = const()[name = tensor<string, []>("op_12008_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_12008_cast = slice_by_index(begin = var_12008_begin_0, end = var_12008_end_0, end_mask = var_12008_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_12008_cast")];
            tensor<int32, [4]> var_12012_begin_0 = const()[name = tensor<string, []>("op_12012_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_12012_end_0 = const()[name = tensor<string, []>("op_12012_end_0"), val = tensor<int32, [4]>([2, 280, 1, 4096])];
            tensor<bool, [4]> var_12012_end_mask_0 = const()[name = tensor<string, []>("op_12012_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_12012_cast = slice_by_index(begin = var_12012_begin_0, end = var_12012_end_0, end_mask = var_12012_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_12012_cast")];
            tensor<int32, [4]> var_12016_begin_0 = const()[name = tensor<string, []>("op_12016_begin_0"), val = tensor<int32, [4]>([0, 280, 0, 0])];
            tensor<int32, [4]> var_12016_end_0 = const()[name = tensor<string, []>("op_12016_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_12016_end_mask_0 = const()[name = tensor<string, []>("op_12016_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_12016_cast = slice_by_index(begin = var_12016_begin_0, end = var_12016_end_0, end_mask = var_12016_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_12016_cast")];
            tensor<int32, [4]> var_12019_begin_0 = const()[name = tensor<string, []>("op_12019_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12019_end_0 = const()[name = tensor<string, []>("op_12019_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_12019_end_mask_0 = const()[name = tensor<string, []>("op_12019_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12019_cast = slice_by_index(begin = var_12019_begin_0, end = var_12019_end_0, end_mask = var_12019_end_mask_0, x = var_11988_cast)[name = tensor<string, []>("op_12019_cast")];
            tensor<int32, [4]> var_12020_begin_0 = const()[name = tensor<string, []>("op_12020_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12020_end_0 = const()[name = tensor<string, []>("op_12020_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_12020_end_mask_0 = const()[name = tensor<string, []>("op_12020_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12020_cast = slice_by_index(begin = var_12020_begin_0, end = var_12020_end_0, end_mask = var_12020_end_mask_0, x = var_11988_cast)[name = tensor<string, []>("op_12020_cast")];
            tensor<int32, [4]> var_12021_begin_0 = const()[name = tensor<string, []>("op_12021_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_12021_end_0 = const()[name = tensor<string, []>("op_12021_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_12021_end_mask_0 = const()[name = tensor<string, []>("op_12021_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12021_cast = slice_by_index(begin = var_12021_begin_0, end = var_12021_end_0, end_mask = var_12021_end_mask_0, x = var_11988_cast)[name = tensor<string, []>("op_12021_cast")];
            tensor<int32, [4]> var_12022_begin_0 = const()[name = tensor<string, []>("op_12022_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_12022_end_0 = const()[name = tensor<string, []>("op_12022_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_12022_end_mask_0 = const()[name = tensor<string, []>("op_12022_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12022_cast = slice_by_index(begin = var_12022_begin_0, end = var_12022_end_0, end_mask = var_12022_end_mask_0, x = var_11988_cast)[name = tensor<string, []>("op_12022_cast")];
            tensor<int32, [4]> var_12023_begin_0 = const()[name = tensor<string, []>("op_12023_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_12023_end_0 = const()[name = tensor<string, []>("op_12023_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_12023_end_mask_0 = const()[name = tensor<string, []>("op_12023_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12023_cast = slice_by_index(begin = var_12023_begin_0, end = var_12023_end_0, end_mask = var_12023_end_mask_0, x = var_11988_cast)[name = tensor<string, []>("op_12023_cast")];
            tensor<int32, [4]> var_12024_begin_0 = const()[name = tensor<string, []>("op_12024_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_12024_end_0 = const()[name = tensor<string, []>("op_12024_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_12024_end_mask_0 = const()[name = tensor<string, []>("op_12024_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12024_cast = slice_by_index(begin = var_12024_begin_0, end = var_12024_end_0, end_mask = var_12024_end_mask_0, x = var_11988_cast)[name = tensor<string, []>("op_12024_cast")];
            tensor<int32, [4]> var_12025_begin_0 = const()[name = tensor<string, []>("op_12025_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_12025_end_0 = const()[name = tensor<string, []>("op_12025_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_12025_end_mask_0 = const()[name = tensor<string, []>("op_12025_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12025_cast = slice_by_index(begin = var_12025_begin_0, end = var_12025_end_0, end_mask = var_12025_end_mask_0, x = var_11988_cast)[name = tensor<string, []>("op_12025_cast")];
            tensor<int32, [4]> var_12026_begin_0 = const()[name = tensor<string, []>("op_12026_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_12026_end_0 = const()[name = tensor<string, []>("op_12026_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_12026_end_mask_0 = const()[name = tensor<string, []>("op_12026_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12026_cast = slice_by_index(begin = var_12026_begin_0, end = var_12026_end_0, end_mask = var_12026_end_mask_0, x = var_11988_cast)[name = tensor<string, []>("op_12026_cast")];
            tensor<int32, [4]> var_12027_begin_0 = const()[name = tensor<string, []>("op_12027_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12027_end_0 = const()[name = tensor<string, []>("op_12027_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_12027_end_mask_0 = const()[name = tensor<string, []>("op_12027_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12027_cast = slice_by_index(begin = var_12027_begin_0, end = var_12027_end_0, end_mask = var_12027_end_mask_0, x = var_11992_cast)[name = tensor<string, []>("op_12027_cast")];
            tensor<int32, [4]> var_12028_begin_0 = const()[name = tensor<string, []>("op_12028_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12028_end_0 = const()[name = tensor<string, []>("op_12028_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_12028_end_mask_0 = const()[name = tensor<string, []>("op_12028_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12028_cast = slice_by_index(begin = var_12028_begin_0, end = var_12028_end_0, end_mask = var_12028_end_mask_0, x = var_11992_cast)[name = tensor<string, []>("op_12028_cast")];
            tensor<int32, [4]> var_12029_begin_0 = const()[name = tensor<string, []>("op_12029_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_12029_end_0 = const()[name = tensor<string, []>("op_12029_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_12029_end_mask_0 = const()[name = tensor<string, []>("op_12029_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12029_cast = slice_by_index(begin = var_12029_begin_0, end = var_12029_end_0, end_mask = var_12029_end_mask_0, x = var_11992_cast)[name = tensor<string, []>("op_12029_cast")];
            tensor<int32, [4]> var_12030_begin_0 = const()[name = tensor<string, []>("op_12030_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_12030_end_0 = const()[name = tensor<string, []>("op_12030_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_12030_end_mask_0 = const()[name = tensor<string, []>("op_12030_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12030_cast = slice_by_index(begin = var_12030_begin_0, end = var_12030_end_0, end_mask = var_12030_end_mask_0, x = var_11992_cast)[name = tensor<string, []>("op_12030_cast")];
            tensor<int32, [4]> var_12031_begin_0 = const()[name = tensor<string, []>("op_12031_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_12031_end_0 = const()[name = tensor<string, []>("op_12031_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_12031_end_mask_0 = const()[name = tensor<string, []>("op_12031_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12031_cast = slice_by_index(begin = var_12031_begin_0, end = var_12031_end_0, end_mask = var_12031_end_mask_0, x = var_11992_cast)[name = tensor<string, []>("op_12031_cast")];
            tensor<int32, [4]> var_12032_begin_0 = const()[name = tensor<string, []>("op_12032_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_12032_end_0 = const()[name = tensor<string, []>("op_12032_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_12032_end_mask_0 = const()[name = tensor<string, []>("op_12032_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12032_cast = slice_by_index(begin = var_12032_begin_0, end = var_12032_end_0, end_mask = var_12032_end_mask_0, x = var_11992_cast)[name = tensor<string, []>("op_12032_cast")];
            tensor<int32, [4]> var_12033_begin_0 = const()[name = tensor<string, []>("op_12033_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_12033_end_0 = const()[name = tensor<string, []>("op_12033_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_12033_end_mask_0 = const()[name = tensor<string, []>("op_12033_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12033_cast = slice_by_index(begin = var_12033_begin_0, end = var_12033_end_0, end_mask = var_12033_end_mask_0, x = var_11992_cast)[name = tensor<string, []>("op_12033_cast")];
            tensor<int32, [4]> var_12034_begin_0 = const()[name = tensor<string, []>("op_12034_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_12034_end_0 = const()[name = tensor<string, []>("op_12034_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_12034_end_mask_0 = const()[name = tensor<string, []>("op_12034_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12034_cast = slice_by_index(begin = var_12034_begin_0, end = var_12034_end_0, end_mask = var_12034_end_mask_0, x = var_11992_cast)[name = tensor<string, []>("op_12034_cast")];
            tensor<int32, [4]> var_12035_begin_0 = const()[name = tensor<string, []>("op_12035_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12035_end_0 = const()[name = tensor<string, []>("op_12035_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_12035_end_mask_0 = const()[name = tensor<string, []>("op_12035_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12035_cast = slice_by_index(begin = var_12035_begin_0, end = var_12035_end_0, end_mask = var_12035_end_mask_0, x = var_11996_cast)[name = tensor<string, []>("op_12035_cast")];
            tensor<int32, [4]> var_12036_begin_0 = const()[name = tensor<string, []>("op_12036_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12036_end_0 = const()[name = tensor<string, []>("op_12036_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_12036_end_mask_0 = const()[name = tensor<string, []>("op_12036_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12036_cast = slice_by_index(begin = var_12036_begin_0, end = var_12036_end_0, end_mask = var_12036_end_mask_0, x = var_11996_cast)[name = tensor<string, []>("op_12036_cast")];
            tensor<int32, [4]> var_12037_begin_0 = const()[name = tensor<string, []>("op_12037_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_12037_end_0 = const()[name = tensor<string, []>("op_12037_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_12037_end_mask_0 = const()[name = tensor<string, []>("op_12037_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12037_cast = slice_by_index(begin = var_12037_begin_0, end = var_12037_end_0, end_mask = var_12037_end_mask_0, x = var_11996_cast)[name = tensor<string, []>("op_12037_cast")];
            tensor<int32, [4]> var_12038_begin_0 = const()[name = tensor<string, []>("op_12038_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_12038_end_0 = const()[name = tensor<string, []>("op_12038_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_12038_end_mask_0 = const()[name = tensor<string, []>("op_12038_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12038_cast = slice_by_index(begin = var_12038_begin_0, end = var_12038_end_0, end_mask = var_12038_end_mask_0, x = var_11996_cast)[name = tensor<string, []>("op_12038_cast")];
            tensor<int32, [4]> var_12039_begin_0 = const()[name = tensor<string, []>("op_12039_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_12039_end_0 = const()[name = tensor<string, []>("op_12039_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_12039_end_mask_0 = const()[name = tensor<string, []>("op_12039_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12039_cast = slice_by_index(begin = var_12039_begin_0, end = var_12039_end_0, end_mask = var_12039_end_mask_0, x = var_11996_cast)[name = tensor<string, []>("op_12039_cast")];
            tensor<int32, [4]> var_12040_begin_0 = const()[name = tensor<string, []>("op_12040_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_12040_end_0 = const()[name = tensor<string, []>("op_12040_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_12040_end_mask_0 = const()[name = tensor<string, []>("op_12040_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12040_cast = slice_by_index(begin = var_12040_begin_0, end = var_12040_end_0, end_mask = var_12040_end_mask_0, x = var_11996_cast)[name = tensor<string, []>("op_12040_cast")];
            tensor<int32, [4]> var_12041_begin_0 = const()[name = tensor<string, []>("op_12041_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_12041_end_0 = const()[name = tensor<string, []>("op_12041_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_12041_end_mask_0 = const()[name = tensor<string, []>("op_12041_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12041_cast = slice_by_index(begin = var_12041_begin_0, end = var_12041_end_0, end_mask = var_12041_end_mask_0, x = var_11996_cast)[name = tensor<string, []>("op_12041_cast")];
            tensor<int32, [4]> var_12042_begin_0 = const()[name = tensor<string, []>("op_12042_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_12042_end_0 = const()[name = tensor<string, []>("op_12042_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_12042_end_mask_0 = const()[name = tensor<string, []>("op_12042_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12042_cast = slice_by_index(begin = var_12042_begin_0, end = var_12042_end_0, end_mask = var_12042_end_mask_0, x = var_11996_cast)[name = tensor<string, []>("op_12042_cast")];
            tensor<int32, [4]> var_12043_begin_0 = const()[name = tensor<string, []>("op_12043_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12043_end_0 = const()[name = tensor<string, []>("op_12043_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_12043_end_mask_0 = const()[name = tensor<string, []>("op_12043_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12043_cast = slice_by_index(begin = var_12043_begin_0, end = var_12043_end_0, end_mask = var_12043_end_mask_0, x = var_12000_cast)[name = tensor<string, []>("op_12043_cast")];
            tensor<int32, [4]> var_12044_begin_0 = const()[name = tensor<string, []>("op_12044_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12044_end_0 = const()[name = tensor<string, []>("op_12044_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_12044_end_mask_0 = const()[name = tensor<string, []>("op_12044_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12044_cast = slice_by_index(begin = var_12044_begin_0, end = var_12044_end_0, end_mask = var_12044_end_mask_0, x = var_12000_cast)[name = tensor<string, []>("op_12044_cast")];
            tensor<int32, [4]> var_12045_begin_0 = const()[name = tensor<string, []>("op_12045_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_12045_end_0 = const()[name = tensor<string, []>("op_12045_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_12045_end_mask_0 = const()[name = tensor<string, []>("op_12045_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12045_cast = slice_by_index(begin = var_12045_begin_0, end = var_12045_end_0, end_mask = var_12045_end_mask_0, x = var_12000_cast)[name = tensor<string, []>("op_12045_cast")];
            tensor<int32, [4]> var_12046_begin_0 = const()[name = tensor<string, []>("op_12046_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_12046_end_0 = const()[name = tensor<string, []>("op_12046_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_12046_end_mask_0 = const()[name = tensor<string, []>("op_12046_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12046_cast = slice_by_index(begin = var_12046_begin_0, end = var_12046_end_0, end_mask = var_12046_end_mask_0, x = var_12000_cast)[name = tensor<string, []>("op_12046_cast")];
            tensor<int32, [4]> var_12047_begin_0 = const()[name = tensor<string, []>("op_12047_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_12047_end_0 = const()[name = tensor<string, []>("op_12047_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_12047_end_mask_0 = const()[name = tensor<string, []>("op_12047_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12047_cast = slice_by_index(begin = var_12047_begin_0, end = var_12047_end_0, end_mask = var_12047_end_mask_0, x = var_12000_cast)[name = tensor<string, []>("op_12047_cast")];
            tensor<int32, [4]> var_12048_begin_0 = const()[name = tensor<string, []>("op_12048_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_12048_end_0 = const()[name = tensor<string, []>("op_12048_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_12048_end_mask_0 = const()[name = tensor<string, []>("op_12048_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12048_cast = slice_by_index(begin = var_12048_begin_0, end = var_12048_end_0, end_mask = var_12048_end_mask_0, x = var_12000_cast)[name = tensor<string, []>("op_12048_cast")];
            tensor<int32, [4]> var_12049_begin_0 = const()[name = tensor<string, []>("op_12049_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_12049_end_0 = const()[name = tensor<string, []>("op_12049_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_12049_end_mask_0 = const()[name = tensor<string, []>("op_12049_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12049_cast = slice_by_index(begin = var_12049_begin_0, end = var_12049_end_0, end_mask = var_12049_end_mask_0, x = var_12000_cast)[name = tensor<string, []>("op_12049_cast")];
            tensor<int32, [4]> var_12050_begin_0 = const()[name = tensor<string, []>("op_12050_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_12050_end_0 = const()[name = tensor<string, []>("op_12050_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_12050_end_mask_0 = const()[name = tensor<string, []>("op_12050_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12050_cast = slice_by_index(begin = var_12050_begin_0, end = var_12050_end_0, end_mask = var_12050_end_mask_0, x = var_12000_cast)[name = tensor<string, []>("op_12050_cast")];
            tensor<int32, [4]> var_12051_begin_0 = const()[name = tensor<string, []>("op_12051_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12051_end_0 = const()[name = tensor<string, []>("op_12051_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_12051_end_mask_0 = const()[name = tensor<string, []>("op_12051_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12051_cast = slice_by_index(begin = var_12051_begin_0, end = var_12051_end_0, end_mask = var_12051_end_mask_0, x = var_12004_cast)[name = tensor<string, []>("op_12051_cast")];
            tensor<int32, [4]> var_12052_begin_0 = const()[name = tensor<string, []>("op_12052_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12052_end_0 = const()[name = tensor<string, []>("op_12052_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_12052_end_mask_0 = const()[name = tensor<string, []>("op_12052_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12052_cast = slice_by_index(begin = var_12052_begin_0, end = var_12052_end_0, end_mask = var_12052_end_mask_0, x = var_12004_cast)[name = tensor<string, []>("op_12052_cast")];
            tensor<int32, [4]> var_12053_begin_0 = const()[name = tensor<string, []>("op_12053_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_12053_end_0 = const()[name = tensor<string, []>("op_12053_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_12053_end_mask_0 = const()[name = tensor<string, []>("op_12053_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12053_cast = slice_by_index(begin = var_12053_begin_0, end = var_12053_end_0, end_mask = var_12053_end_mask_0, x = var_12004_cast)[name = tensor<string, []>("op_12053_cast")];
            tensor<int32, [4]> var_12054_begin_0 = const()[name = tensor<string, []>("op_12054_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_12054_end_0 = const()[name = tensor<string, []>("op_12054_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_12054_end_mask_0 = const()[name = tensor<string, []>("op_12054_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12054_cast = slice_by_index(begin = var_12054_begin_0, end = var_12054_end_0, end_mask = var_12054_end_mask_0, x = var_12004_cast)[name = tensor<string, []>("op_12054_cast")];
            tensor<int32, [4]> var_12055_begin_0 = const()[name = tensor<string, []>("op_12055_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_12055_end_0 = const()[name = tensor<string, []>("op_12055_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_12055_end_mask_0 = const()[name = tensor<string, []>("op_12055_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12055_cast = slice_by_index(begin = var_12055_begin_0, end = var_12055_end_0, end_mask = var_12055_end_mask_0, x = var_12004_cast)[name = tensor<string, []>("op_12055_cast")];
            tensor<int32, [4]> var_12056_begin_0 = const()[name = tensor<string, []>("op_12056_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_12056_end_0 = const()[name = tensor<string, []>("op_12056_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_12056_end_mask_0 = const()[name = tensor<string, []>("op_12056_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12056_cast = slice_by_index(begin = var_12056_begin_0, end = var_12056_end_0, end_mask = var_12056_end_mask_0, x = var_12004_cast)[name = tensor<string, []>("op_12056_cast")];
            tensor<int32, [4]> var_12057_begin_0 = const()[name = tensor<string, []>("op_12057_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_12057_end_0 = const()[name = tensor<string, []>("op_12057_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_12057_end_mask_0 = const()[name = tensor<string, []>("op_12057_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12057_cast = slice_by_index(begin = var_12057_begin_0, end = var_12057_end_0, end_mask = var_12057_end_mask_0, x = var_12004_cast)[name = tensor<string, []>("op_12057_cast")];
            tensor<int32, [4]> var_12058_begin_0 = const()[name = tensor<string, []>("op_12058_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_12058_end_0 = const()[name = tensor<string, []>("op_12058_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_12058_end_mask_0 = const()[name = tensor<string, []>("op_12058_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12058_cast = slice_by_index(begin = var_12058_begin_0, end = var_12058_end_0, end_mask = var_12058_end_mask_0, x = var_12004_cast)[name = tensor<string, []>("op_12058_cast")];
            tensor<int32, [4]> var_12059_begin_0 = const()[name = tensor<string, []>("op_12059_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12059_end_0 = const()[name = tensor<string, []>("op_12059_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_12059_end_mask_0 = const()[name = tensor<string, []>("op_12059_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12059_cast = slice_by_index(begin = var_12059_begin_0, end = var_12059_end_0, end_mask = var_12059_end_mask_0, x = var_12008_cast)[name = tensor<string, []>("op_12059_cast")];
            tensor<int32, [4]> var_12060_begin_0 = const()[name = tensor<string, []>("op_12060_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12060_end_0 = const()[name = tensor<string, []>("op_12060_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_12060_end_mask_0 = const()[name = tensor<string, []>("op_12060_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12060_cast = slice_by_index(begin = var_12060_begin_0, end = var_12060_end_0, end_mask = var_12060_end_mask_0, x = var_12008_cast)[name = tensor<string, []>("op_12060_cast")];
            tensor<int32, [4]> var_12061_begin_0 = const()[name = tensor<string, []>("op_12061_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_12061_end_0 = const()[name = tensor<string, []>("op_12061_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_12061_end_mask_0 = const()[name = tensor<string, []>("op_12061_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12061_cast = slice_by_index(begin = var_12061_begin_0, end = var_12061_end_0, end_mask = var_12061_end_mask_0, x = var_12008_cast)[name = tensor<string, []>("op_12061_cast")];
            tensor<int32, [4]> var_12062_begin_0 = const()[name = tensor<string, []>("op_12062_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_12062_end_0 = const()[name = tensor<string, []>("op_12062_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_12062_end_mask_0 = const()[name = tensor<string, []>("op_12062_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12062_cast = slice_by_index(begin = var_12062_begin_0, end = var_12062_end_0, end_mask = var_12062_end_mask_0, x = var_12008_cast)[name = tensor<string, []>("op_12062_cast")];
            tensor<int32, [4]> var_12063_begin_0 = const()[name = tensor<string, []>("op_12063_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_12063_end_0 = const()[name = tensor<string, []>("op_12063_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_12063_end_mask_0 = const()[name = tensor<string, []>("op_12063_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12063_cast = slice_by_index(begin = var_12063_begin_0, end = var_12063_end_0, end_mask = var_12063_end_mask_0, x = var_12008_cast)[name = tensor<string, []>("op_12063_cast")];
            tensor<int32, [4]> var_12064_begin_0 = const()[name = tensor<string, []>("op_12064_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_12064_end_0 = const()[name = tensor<string, []>("op_12064_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_12064_end_mask_0 = const()[name = tensor<string, []>("op_12064_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12064_cast = slice_by_index(begin = var_12064_begin_0, end = var_12064_end_0, end_mask = var_12064_end_mask_0, x = var_12008_cast)[name = tensor<string, []>("op_12064_cast")];
            tensor<int32, [4]> var_12065_begin_0 = const()[name = tensor<string, []>("op_12065_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_12065_end_0 = const()[name = tensor<string, []>("op_12065_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_12065_end_mask_0 = const()[name = tensor<string, []>("op_12065_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12065_cast = slice_by_index(begin = var_12065_begin_0, end = var_12065_end_0, end_mask = var_12065_end_mask_0, x = var_12008_cast)[name = tensor<string, []>("op_12065_cast")];
            tensor<int32, [4]> var_12066_begin_0 = const()[name = tensor<string, []>("op_12066_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_12066_end_0 = const()[name = tensor<string, []>("op_12066_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_12066_end_mask_0 = const()[name = tensor<string, []>("op_12066_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12066_cast = slice_by_index(begin = var_12066_begin_0, end = var_12066_end_0, end_mask = var_12066_end_mask_0, x = var_12008_cast)[name = tensor<string, []>("op_12066_cast")];
            tensor<int32, [4]> var_12067_begin_0 = const()[name = tensor<string, []>("op_12067_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12067_end_0 = const()[name = tensor<string, []>("op_12067_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_12067_end_mask_0 = const()[name = tensor<string, []>("op_12067_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12067_cast = slice_by_index(begin = var_12067_begin_0, end = var_12067_end_0, end_mask = var_12067_end_mask_0, x = var_12012_cast)[name = tensor<string, []>("op_12067_cast")];
            tensor<int32, [4]> var_12068_begin_0 = const()[name = tensor<string, []>("op_12068_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12068_end_0 = const()[name = tensor<string, []>("op_12068_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_12068_end_mask_0 = const()[name = tensor<string, []>("op_12068_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12068_cast = slice_by_index(begin = var_12068_begin_0, end = var_12068_end_0, end_mask = var_12068_end_mask_0, x = var_12012_cast)[name = tensor<string, []>("op_12068_cast")];
            tensor<int32, [4]> var_12069_begin_0 = const()[name = tensor<string, []>("op_12069_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_12069_end_0 = const()[name = tensor<string, []>("op_12069_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_12069_end_mask_0 = const()[name = tensor<string, []>("op_12069_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12069_cast = slice_by_index(begin = var_12069_begin_0, end = var_12069_end_0, end_mask = var_12069_end_mask_0, x = var_12012_cast)[name = tensor<string, []>("op_12069_cast")];
            tensor<int32, [4]> var_12070_begin_0 = const()[name = tensor<string, []>("op_12070_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_12070_end_0 = const()[name = tensor<string, []>("op_12070_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_12070_end_mask_0 = const()[name = tensor<string, []>("op_12070_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12070_cast = slice_by_index(begin = var_12070_begin_0, end = var_12070_end_0, end_mask = var_12070_end_mask_0, x = var_12012_cast)[name = tensor<string, []>("op_12070_cast")];
            tensor<int32, [4]> var_12071_begin_0 = const()[name = tensor<string, []>("op_12071_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_12071_end_0 = const()[name = tensor<string, []>("op_12071_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_12071_end_mask_0 = const()[name = tensor<string, []>("op_12071_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12071_cast = slice_by_index(begin = var_12071_begin_0, end = var_12071_end_0, end_mask = var_12071_end_mask_0, x = var_12012_cast)[name = tensor<string, []>("op_12071_cast")];
            tensor<int32, [4]> var_12072_begin_0 = const()[name = tensor<string, []>("op_12072_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_12072_end_0 = const()[name = tensor<string, []>("op_12072_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_12072_end_mask_0 = const()[name = tensor<string, []>("op_12072_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12072_cast = slice_by_index(begin = var_12072_begin_0, end = var_12072_end_0, end_mask = var_12072_end_mask_0, x = var_12012_cast)[name = tensor<string, []>("op_12072_cast")];
            tensor<int32, [4]> var_12073_begin_0 = const()[name = tensor<string, []>("op_12073_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_12073_end_0 = const()[name = tensor<string, []>("op_12073_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_12073_end_mask_0 = const()[name = tensor<string, []>("op_12073_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12073_cast = slice_by_index(begin = var_12073_begin_0, end = var_12073_end_0, end_mask = var_12073_end_mask_0, x = var_12012_cast)[name = tensor<string, []>("op_12073_cast")];
            tensor<int32, [4]> var_12074_begin_0 = const()[name = tensor<string, []>("op_12074_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_12074_end_0 = const()[name = tensor<string, []>("op_12074_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_12074_end_mask_0 = const()[name = tensor<string, []>("op_12074_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12074_cast = slice_by_index(begin = var_12074_begin_0, end = var_12074_end_0, end_mask = var_12074_end_mask_0, x = var_12012_cast)[name = tensor<string, []>("op_12074_cast")];
            tensor<int32, [4]> var_12075_begin_0 = const()[name = tensor<string, []>("op_12075_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12075_end_0 = const()[name = tensor<string, []>("op_12075_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_12075_end_mask_0 = const()[name = tensor<string, []>("op_12075_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12075_cast = slice_by_index(begin = var_12075_begin_0, end = var_12075_end_0, end_mask = var_12075_end_mask_0, x = var_12016_cast)[name = tensor<string, []>("op_12075_cast")];
            tensor<int32, [4]> var_12076_begin_0 = const()[name = tensor<string, []>("op_12076_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12076_end_0 = const()[name = tensor<string, []>("op_12076_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_12076_end_mask_0 = const()[name = tensor<string, []>("op_12076_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12076_cast = slice_by_index(begin = var_12076_begin_0, end = var_12076_end_0, end_mask = var_12076_end_mask_0, x = var_12016_cast)[name = tensor<string, []>("op_12076_cast")];
            tensor<int32, [4]> var_12077_begin_0 = const()[name = tensor<string, []>("op_12077_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_12077_end_0 = const()[name = tensor<string, []>("op_12077_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_12077_end_mask_0 = const()[name = tensor<string, []>("op_12077_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12077_cast = slice_by_index(begin = var_12077_begin_0, end = var_12077_end_0, end_mask = var_12077_end_mask_0, x = var_12016_cast)[name = tensor<string, []>("op_12077_cast")];
            tensor<int32, [4]> var_12078_begin_0 = const()[name = tensor<string, []>("op_12078_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_12078_end_0 = const()[name = tensor<string, []>("op_12078_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_12078_end_mask_0 = const()[name = tensor<string, []>("op_12078_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12078_cast = slice_by_index(begin = var_12078_begin_0, end = var_12078_end_0, end_mask = var_12078_end_mask_0, x = var_12016_cast)[name = tensor<string, []>("op_12078_cast")];
            tensor<int32, [4]> var_12079_begin_0 = const()[name = tensor<string, []>("op_12079_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_12079_end_0 = const()[name = tensor<string, []>("op_12079_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_12079_end_mask_0 = const()[name = tensor<string, []>("op_12079_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12079_cast = slice_by_index(begin = var_12079_begin_0, end = var_12079_end_0, end_mask = var_12079_end_mask_0, x = var_12016_cast)[name = tensor<string, []>("op_12079_cast")];
            tensor<int32, [4]> var_12080_begin_0 = const()[name = tensor<string, []>("op_12080_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_12080_end_0 = const()[name = tensor<string, []>("op_12080_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_12080_end_mask_0 = const()[name = tensor<string, []>("op_12080_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12080_cast = slice_by_index(begin = var_12080_begin_0, end = var_12080_end_0, end_mask = var_12080_end_mask_0, x = var_12016_cast)[name = tensor<string, []>("op_12080_cast")];
            tensor<int32, [4]> var_12081_begin_0 = const()[name = tensor<string, []>("op_12081_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_12081_end_0 = const()[name = tensor<string, []>("op_12081_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_12081_end_mask_0 = const()[name = tensor<string, []>("op_12081_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12081_cast = slice_by_index(begin = var_12081_begin_0, end = var_12081_end_0, end_mask = var_12081_end_mask_0, x = var_12016_cast)[name = tensor<string, []>("op_12081_cast")];
            tensor<int32, [4]> var_12082_begin_0 = const()[name = tensor<string, []>("op_12082_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_12082_end_0 = const()[name = tensor<string, []>("op_12082_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_12082_end_mask_0 = const()[name = tensor<string, []>("op_12082_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12082_cast = slice_by_index(begin = var_12082_begin_0, end = var_12082_end_0, end_mask = var_12082_end_mask_0, x = var_12016_cast)[name = tensor<string, []>("op_12082_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_12087_begin_0 = const()[name = tensor<string, []>("op_12087_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12087_end_0 = const()[name = tensor<string, []>("op_12087_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 40])];
            tensor<bool, [4]> var_12087_end_mask_0 = const()[name = tensor<string, []>("op_12087_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)[name = tensor<string, []>("transpose_3")];
            tensor<fp16, [2, 4096, 1, 40]> var_12087_cast = slice_by_index(begin = var_12087_begin_0, end = var_12087_end_0, end_mask = var_12087_end_mask_0, x = transpose_3)[name = tensor<string, []>("op_12087_cast")];
            tensor<int32, [4]> var_12091_begin_0 = const()[name = tensor<string, []>("op_12091_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 40])];
            tensor<int32, [4]> var_12091_end_0 = const()[name = tensor<string, []>("op_12091_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 80])];
            tensor<bool, [4]> var_12091_end_mask_0 = const()[name = tensor<string, []>("op_12091_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_12091_cast = slice_by_index(begin = var_12091_begin_0, end = var_12091_end_0, end_mask = var_12091_end_mask_0, x = transpose_3)[name = tensor<string, []>("op_12091_cast")];
            tensor<int32, [4]> var_12095_begin_0 = const()[name = tensor<string, []>("op_12095_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 80])];
            tensor<int32, [4]> var_12095_end_0 = const()[name = tensor<string, []>("op_12095_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 120])];
            tensor<bool, [4]> var_12095_end_mask_0 = const()[name = tensor<string, []>("op_12095_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_12095_cast = slice_by_index(begin = var_12095_begin_0, end = var_12095_end_0, end_mask = var_12095_end_mask_0, x = transpose_3)[name = tensor<string, []>("op_12095_cast")];
            tensor<int32, [4]> var_12099_begin_0 = const()[name = tensor<string, []>("op_12099_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 120])];
            tensor<int32, [4]> var_12099_end_0 = const()[name = tensor<string, []>("op_12099_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 160])];
            tensor<bool, [4]> var_12099_end_mask_0 = const()[name = tensor<string, []>("op_12099_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_12099_cast = slice_by_index(begin = var_12099_begin_0, end = var_12099_end_0, end_mask = var_12099_end_mask_0, x = transpose_3)[name = tensor<string, []>("op_12099_cast")];
            tensor<int32, [4]> var_12103_begin_0 = const()[name = tensor<string, []>("op_12103_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_12103_end_0 = const()[name = tensor<string, []>("op_12103_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 200])];
            tensor<bool, [4]> var_12103_end_mask_0 = const()[name = tensor<string, []>("op_12103_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_12103_cast = slice_by_index(begin = var_12103_begin_0, end = var_12103_end_0, end_mask = var_12103_end_mask_0, x = transpose_3)[name = tensor<string, []>("op_12103_cast")];
            tensor<int32, [4]> var_12107_begin_0 = const()[name = tensor<string, []>("op_12107_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 200])];
            tensor<int32, [4]> var_12107_end_0 = const()[name = tensor<string, []>("op_12107_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 240])];
            tensor<bool, [4]> var_12107_end_mask_0 = const()[name = tensor<string, []>("op_12107_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_12107_cast = slice_by_index(begin = var_12107_begin_0, end = var_12107_end_0, end_mask = var_12107_end_mask_0, x = transpose_3)[name = tensor<string, []>("op_12107_cast")];
            tensor<int32, [4]> var_12111_begin_0 = const()[name = tensor<string, []>("op_12111_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 240])];
            tensor<int32, [4]> var_12111_end_0 = const()[name = tensor<string, []>("op_12111_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 280])];
            tensor<bool, [4]> var_12111_end_mask_0 = const()[name = tensor<string, []>("op_12111_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_12111_cast = slice_by_index(begin = var_12111_begin_0, end = var_12111_end_0, end_mask = var_12111_end_mask_0, x = transpose_3)[name = tensor<string, []>("op_12111_cast")];
            tensor<int32, [4]> var_12115_begin_0 = const()[name = tensor<string, []>("op_12115_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 280])];
            tensor<int32, [4]> var_12115_end_0 = const()[name = tensor<string, []>("op_12115_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_12115_end_mask_0 = const()[name = tensor<string, []>("op_12115_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_12115_cast = slice_by_index(begin = var_12115_begin_0, end = var_12115_end_0, end_mask = var_12115_end_mask_0, x = transpose_3)[name = tensor<string, []>("op_12115_cast")];
            tensor<int32, [4]> var_12117_begin_0 = const()[name = tensor<string, []>("op_12117_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12117_end_0 = const()[name = tensor<string, []>("op_12117_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_12117_end_mask_0 = const()[name = tensor<string, []>("op_12117_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_12117_cast = slice_by_index(begin = var_12117_begin_0, end = var_12117_end_0, end_mask = var_12117_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_12117_cast")];
            tensor<int32, [4]> var_12121_begin_0 = const()[name = tensor<string, []>("op_12121_begin_0"), val = tensor<int32, [4]>([0, 40, 0, 0])];
            tensor<int32, [4]> var_12121_end_0 = const()[name = tensor<string, []>("op_12121_end_0"), val = tensor<int32, [4]>([2, 80, 1, 4096])];
            tensor<bool, [4]> var_12121_end_mask_0 = const()[name = tensor<string, []>("op_12121_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_12121_cast = slice_by_index(begin = var_12121_begin_0, end = var_12121_end_0, end_mask = var_12121_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_12121_cast")];
            tensor<int32, [4]> var_12125_begin_0 = const()[name = tensor<string, []>("op_12125_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_12125_end_0 = const()[name = tensor<string, []>("op_12125_end_0"), val = tensor<int32, [4]>([2, 120, 1, 4096])];
            tensor<bool, [4]> var_12125_end_mask_0 = const()[name = tensor<string, []>("op_12125_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_12125_cast = slice_by_index(begin = var_12125_begin_0, end = var_12125_end_0, end_mask = var_12125_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_12125_cast")];
            tensor<int32, [4]> var_12129_begin_0 = const()[name = tensor<string, []>("op_12129_begin_0"), val = tensor<int32, [4]>([0, 120, 0, 0])];
            tensor<int32, [4]> var_12129_end_0 = const()[name = tensor<string, []>("op_12129_end_0"), val = tensor<int32, [4]>([2, 160, 1, 4096])];
            tensor<bool, [4]> var_12129_end_mask_0 = const()[name = tensor<string, []>("op_12129_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_12129_cast = slice_by_index(begin = var_12129_begin_0, end = var_12129_end_0, end_mask = var_12129_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_12129_cast")];
            tensor<int32, [4]> var_12133_begin_0 = const()[name = tensor<string, []>("op_12133_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_12133_end_0 = const()[name = tensor<string, []>("op_12133_end_0"), val = tensor<int32, [4]>([2, 200, 1, 4096])];
            tensor<bool, [4]> var_12133_end_mask_0 = const()[name = tensor<string, []>("op_12133_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_12133_cast = slice_by_index(begin = var_12133_begin_0, end = var_12133_end_0, end_mask = var_12133_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_12133_cast")];
            tensor<int32, [4]> var_12137_begin_0 = const()[name = tensor<string, []>("op_12137_begin_0"), val = tensor<int32, [4]>([0, 200, 0, 0])];
            tensor<int32, [4]> var_12137_end_0 = const()[name = tensor<string, []>("op_12137_end_0"), val = tensor<int32, [4]>([2, 240, 1, 4096])];
            tensor<bool, [4]> var_12137_end_mask_0 = const()[name = tensor<string, []>("op_12137_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_12137_cast = slice_by_index(begin = var_12137_begin_0, end = var_12137_end_0, end_mask = var_12137_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_12137_cast")];
            tensor<int32, [4]> var_12141_begin_0 = const()[name = tensor<string, []>("op_12141_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_12141_end_0 = const()[name = tensor<string, []>("op_12141_end_0"), val = tensor<int32, [4]>([2, 280, 1, 4096])];
            tensor<bool, [4]> var_12141_end_mask_0 = const()[name = tensor<string, []>("op_12141_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_12141_cast = slice_by_index(begin = var_12141_begin_0, end = var_12141_end_0, end_mask = var_12141_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_12141_cast")];
            tensor<int32, [4]> var_12145_begin_0 = const()[name = tensor<string, []>("op_12145_begin_0"), val = tensor<int32, [4]>([0, 280, 0, 0])];
            tensor<int32, [4]> var_12145_end_0 = const()[name = tensor<string, []>("op_12145_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_12145_end_mask_0 = const()[name = tensor<string, []>("op_12145_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_12145_cast = slice_by_index(begin = var_12145_begin_0, end = var_12145_end_0, end_mask = var_12145_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_12145_cast")];
            tensor<string, []> var_12149_equation_0 = const()[name = tensor<string, []>("op_12149_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12149_cast = einsum(equation = var_12149_equation_0, values = (var_12087_cast, var_12019_cast))[name = tensor<string, []>("op_12149_cast")];
            tensor<fp16, []> var_12150_to_fp16 = const()[name = tensor<string, []>("op_12150_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1089_cast = mul(x = var_12149_cast, y = var_12150_to_fp16)[name = tensor<string, []>("aw_chunk_1089_cast")];
            tensor<string, []> var_12153_equation_0 = const()[name = tensor<string, []>("op_12153_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12153_cast = einsum(equation = var_12153_equation_0, values = (var_12087_cast, var_12020_cast))[name = tensor<string, []>("op_12153_cast")];
            tensor<fp16, []> var_12154_to_fp16 = const()[name = tensor<string, []>("op_12154_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1091_cast = mul(x = var_12153_cast, y = var_12154_to_fp16)[name = tensor<string, []>("aw_chunk_1091_cast")];
            tensor<string, []> var_12157_equation_0 = const()[name = tensor<string, []>("op_12157_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12157_cast = einsum(equation = var_12157_equation_0, values = (var_12087_cast, var_12021_cast))[name = tensor<string, []>("op_12157_cast")];
            tensor<fp16, []> var_12158_to_fp16 = const()[name = tensor<string, []>("op_12158_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1093_cast = mul(x = var_12157_cast, y = var_12158_to_fp16)[name = tensor<string, []>("aw_chunk_1093_cast")];
            tensor<string, []> var_12161_equation_0 = const()[name = tensor<string, []>("op_12161_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12161_cast = einsum(equation = var_12161_equation_0, values = (var_12087_cast, var_12022_cast))[name = tensor<string, []>("op_12161_cast")];
            tensor<fp16, []> var_12162_to_fp16 = const()[name = tensor<string, []>("op_12162_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1095_cast = mul(x = var_12161_cast, y = var_12162_to_fp16)[name = tensor<string, []>("aw_chunk_1095_cast")];
            tensor<string, []> var_12165_equation_0 = const()[name = tensor<string, []>("op_12165_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12165_cast = einsum(equation = var_12165_equation_0, values = (var_12087_cast, var_12023_cast))[name = tensor<string, []>("op_12165_cast")];
            tensor<fp16, []> var_12166_to_fp16 = const()[name = tensor<string, []>("op_12166_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1097_cast = mul(x = var_12165_cast, y = var_12166_to_fp16)[name = tensor<string, []>("aw_chunk_1097_cast")];
            tensor<string, []> var_12169_equation_0 = const()[name = tensor<string, []>("op_12169_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12169_cast = einsum(equation = var_12169_equation_0, values = (var_12087_cast, var_12024_cast))[name = tensor<string, []>("op_12169_cast")];
            tensor<fp16, []> var_12170_to_fp16 = const()[name = tensor<string, []>("op_12170_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1099_cast = mul(x = var_12169_cast, y = var_12170_to_fp16)[name = tensor<string, []>("aw_chunk_1099_cast")];
            tensor<string, []> var_12173_equation_0 = const()[name = tensor<string, []>("op_12173_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12173_cast = einsum(equation = var_12173_equation_0, values = (var_12087_cast, var_12025_cast))[name = tensor<string, []>("op_12173_cast")];
            tensor<fp16, []> var_12174_to_fp16 = const()[name = tensor<string, []>("op_12174_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1101_cast = mul(x = var_12173_cast, y = var_12174_to_fp16)[name = tensor<string, []>("aw_chunk_1101_cast")];
            tensor<string, []> var_12177_equation_0 = const()[name = tensor<string, []>("op_12177_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12177_cast = einsum(equation = var_12177_equation_0, values = (var_12087_cast, var_12026_cast))[name = tensor<string, []>("op_12177_cast")];
            tensor<fp16, []> var_12178_to_fp16 = const()[name = tensor<string, []>("op_12178_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1103_cast = mul(x = var_12177_cast, y = var_12178_to_fp16)[name = tensor<string, []>("aw_chunk_1103_cast")];
            tensor<string, []> var_12181_equation_0 = const()[name = tensor<string, []>("op_12181_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12181_cast = einsum(equation = var_12181_equation_0, values = (var_12091_cast, var_12027_cast))[name = tensor<string, []>("op_12181_cast")];
            tensor<fp16, []> var_12182_to_fp16 = const()[name = tensor<string, []>("op_12182_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1105_cast = mul(x = var_12181_cast, y = var_12182_to_fp16)[name = tensor<string, []>("aw_chunk_1105_cast")];
            tensor<string, []> var_12185_equation_0 = const()[name = tensor<string, []>("op_12185_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12185_cast = einsum(equation = var_12185_equation_0, values = (var_12091_cast, var_12028_cast))[name = tensor<string, []>("op_12185_cast")];
            tensor<fp16, []> var_12186_to_fp16 = const()[name = tensor<string, []>("op_12186_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1107_cast = mul(x = var_12185_cast, y = var_12186_to_fp16)[name = tensor<string, []>("aw_chunk_1107_cast")];
            tensor<string, []> var_12189_equation_0 = const()[name = tensor<string, []>("op_12189_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12189_cast = einsum(equation = var_12189_equation_0, values = (var_12091_cast, var_12029_cast))[name = tensor<string, []>("op_12189_cast")];
            tensor<fp16, []> var_12190_to_fp16 = const()[name = tensor<string, []>("op_12190_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1109_cast = mul(x = var_12189_cast, y = var_12190_to_fp16)[name = tensor<string, []>("aw_chunk_1109_cast")];
            tensor<string, []> var_12193_equation_0 = const()[name = tensor<string, []>("op_12193_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12193_cast = einsum(equation = var_12193_equation_0, values = (var_12091_cast, var_12030_cast))[name = tensor<string, []>("op_12193_cast")];
            tensor<fp16, []> var_12194_to_fp16 = const()[name = tensor<string, []>("op_12194_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1111_cast = mul(x = var_12193_cast, y = var_12194_to_fp16)[name = tensor<string, []>("aw_chunk_1111_cast")];
            tensor<string, []> var_12197_equation_0 = const()[name = tensor<string, []>("op_12197_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12197_cast = einsum(equation = var_12197_equation_0, values = (var_12091_cast, var_12031_cast))[name = tensor<string, []>("op_12197_cast")];
            tensor<fp16, []> var_12198_to_fp16 = const()[name = tensor<string, []>("op_12198_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1113_cast = mul(x = var_12197_cast, y = var_12198_to_fp16)[name = tensor<string, []>("aw_chunk_1113_cast")];
            tensor<string, []> var_12201_equation_0 = const()[name = tensor<string, []>("op_12201_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12201_cast = einsum(equation = var_12201_equation_0, values = (var_12091_cast, var_12032_cast))[name = tensor<string, []>("op_12201_cast")];
            tensor<fp16, []> var_12202_to_fp16 = const()[name = tensor<string, []>("op_12202_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1115_cast = mul(x = var_12201_cast, y = var_12202_to_fp16)[name = tensor<string, []>("aw_chunk_1115_cast")];
            tensor<string, []> var_12205_equation_0 = const()[name = tensor<string, []>("op_12205_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12205_cast = einsum(equation = var_12205_equation_0, values = (var_12091_cast, var_12033_cast))[name = tensor<string, []>("op_12205_cast")];
            tensor<fp16, []> var_12206_to_fp16 = const()[name = tensor<string, []>("op_12206_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1117_cast = mul(x = var_12205_cast, y = var_12206_to_fp16)[name = tensor<string, []>("aw_chunk_1117_cast")];
            tensor<string, []> var_12209_equation_0 = const()[name = tensor<string, []>("op_12209_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12209_cast = einsum(equation = var_12209_equation_0, values = (var_12091_cast, var_12034_cast))[name = tensor<string, []>("op_12209_cast")];
            tensor<fp16, []> var_12210_to_fp16 = const()[name = tensor<string, []>("op_12210_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1119_cast = mul(x = var_12209_cast, y = var_12210_to_fp16)[name = tensor<string, []>("aw_chunk_1119_cast")];
            tensor<string, []> var_12213_equation_0 = const()[name = tensor<string, []>("op_12213_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12213_cast = einsum(equation = var_12213_equation_0, values = (var_12095_cast, var_12035_cast))[name = tensor<string, []>("op_12213_cast")];
            tensor<fp16, []> var_12214_to_fp16 = const()[name = tensor<string, []>("op_12214_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1121_cast = mul(x = var_12213_cast, y = var_12214_to_fp16)[name = tensor<string, []>("aw_chunk_1121_cast")];
            tensor<string, []> var_12217_equation_0 = const()[name = tensor<string, []>("op_12217_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12217_cast = einsum(equation = var_12217_equation_0, values = (var_12095_cast, var_12036_cast))[name = tensor<string, []>("op_12217_cast")];
            tensor<fp16, []> var_12218_to_fp16 = const()[name = tensor<string, []>("op_12218_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1123_cast = mul(x = var_12217_cast, y = var_12218_to_fp16)[name = tensor<string, []>("aw_chunk_1123_cast")];
            tensor<string, []> var_12221_equation_0 = const()[name = tensor<string, []>("op_12221_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12221_cast = einsum(equation = var_12221_equation_0, values = (var_12095_cast, var_12037_cast))[name = tensor<string, []>("op_12221_cast")];
            tensor<fp16, []> var_12222_to_fp16 = const()[name = tensor<string, []>("op_12222_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1125_cast = mul(x = var_12221_cast, y = var_12222_to_fp16)[name = tensor<string, []>("aw_chunk_1125_cast")];
            tensor<string, []> var_12225_equation_0 = const()[name = tensor<string, []>("op_12225_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12225_cast = einsum(equation = var_12225_equation_0, values = (var_12095_cast, var_12038_cast))[name = tensor<string, []>("op_12225_cast")];
            tensor<fp16, []> var_12226_to_fp16 = const()[name = tensor<string, []>("op_12226_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1127_cast = mul(x = var_12225_cast, y = var_12226_to_fp16)[name = tensor<string, []>("aw_chunk_1127_cast")];
            tensor<string, []> var_12229_equation_0 = const()[name = tensor<string, []>("op_12229_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12229_cast = einsum(equation = var_12229_equation_0, values = (var_12095_cast, var_12039_cast))[name = tensor<string, []>("op_12229_cast")];
            tensor<fp16, []> var_12230_to_fp16 = const()[name = tensor<string, []>("op_12230_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1129_cast = mul(x = var_12229_cast, y = var_12230_to_fp16)[name = tensor<string, []>("aw_chunk_1129_cast")];
            tensor<string, []> var_12233_equation_0 = const()[name = tensor<string, []>("op_12233_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12233_cast = einsum(equation = var_12233_equation_0, values = (var_12095_cast, var_12040_cast))[name = tensor<string, []>("op_12233_cast")];
            tensor<fp16, []> var_12234_to_fp16 = const()[name = tensor<string, []>("op_12234_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1131_cast = mul(x = var_12233_cast, y = var_12234_to_fp16)[name = tensor<string, []>("aw_chunk_1131_cast")];
            tensor<string, []> var_12237_equation_0 = const()[name = tensor<string, []>("op_12237_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12237_cast = einsum(equation = var_12237_equation_0, values = (var_12095_cast, var_12041_cast))[name = tensor<string, []>("op_12237_cast")];
            tensor<fp16, []> var_12238_to_fp16 = const()[name = tensor<string, []>("op_12238_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1133_cast = mul(x = var_12237_cast, y = var_12238_to_fp16)[name = tensor<string, []>("aw_chunk_1133_cast")];
            tensor<string, []> var_12241_equation_0 = const()[name = tensor<string, []>("op_12241_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12241_cast = einsum(equation = var_12241_equation_0, values = (var_12095_cast, var_12042_cast))[name = tensor<string, []>("op_12241_cast")];
            tensor<fp16, []> var_12242_to_fp16 = const()[name = tensor<string, []>("op_12242_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1135_cast = mul(x = var_12241_cast, y = var_12242_to_fp16)[name = tensor<string, []>("aw_chunk_1135_cast")];
            tensor<string, []> var_12245_equation_0 = const()[name = tensor<string, []>("op_12245_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12245_cast = einsum(equation = var_12245_equation_0, values = (var_12099_cast, var_12043_cast))[name = tensor<string, []>("op_12245_cast")];
            tensor<fp16, []> var_12246_to_fp16 = const()[name = tensor<string, []>("op_12246_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1137_cast = mul(x = var_12245_cast, y = var_12246_to_fp16)[name = tensor<string, []>("aw_chunk_1137_cast")];
            tensor<string, []> var_12249_equation_0 = const()[name = tensor<string, []>("op_12249_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12249_cast = einsum(equation = var_12249_equation_0, values = (var_12099_cast, var_12044_cast))[name = tensor<string, []>("op_12249_cast")];
            tensor<fp16, []> var_12250_to_fp16 = const()[name = tensor<string, []>("op_12250_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1139_cast = mul(x = var_12249_cast, y = var_12250_to_fp16)[name = tensor<string, []>("aw_chunk_1139_cast")];
            tensor<string, []> var_12253_equation_0 = const()[name = tensor<string, []>("op_12253_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12253_cast = einsum(equation = var_12253_equation_0, values = (var_12099_cast, var_12045_cast))[name = tensor<string, []>("op_12253_cast")];
            tensor<fp16, []> var_12254_to_fp16 = const()[name = tensor<string, []>("op_12254_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1141_cast = mul(x = var_12253_cast, y = var_12254_to_fp16)[name = tensor<string, []>("aw_chunk_1141_cast")];
            tensor<string, []> var_12257_equation_0 = const()[name = tensor<string, []>("op_12257_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12257_cast = einsum(equation = var_12257_equation_0, values = (var_12099_cast, var_12046_cast))[name = tensor<string, []>("op_12257_cast")];
            tensor<fp16, []> var_12258_to_fp16 = const()[name = tensor<string, []>("op_12258_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1143_cast = mul(x = var_12257_cast, y = var_12258_to_fp16)[name = tensor<string, []>("aw_chunk_1143_cast")];
            tensor<string, []> var_12261_equation_0 = const()[name = tensor<string, []>("op_12261_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12261_cast = einsum(equation = var_12261_equation_0, values = (var_12099_cast, var_12047_cast))[name = tensor<string, []>("op_12261_cast")];
            tensor<fp16, []> var_12262_to_fp16 = const()[name = tensor<string, []>("op_12262_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1145_cast = mul(x = var_12261_cast, y = var_12262_to_fp16)[name = tensor<string, []>("aw_chunk_1145_cast")];
            tensor<string, []> var_12265_equation_0 = const()[name = tensor<string, []>("op_12265_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12265_cast = einsum(equation = var_12265_equation_0, values = (var_12099_cast, var_12048_cast))[name = tensor<string, []>("op_12265_cast")];
            tensor<fp16, []> var_12266_to_fp16 = const()[name = tensor<string, []>("op_12266_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1147_cast = mul(x = var_12265_cast, y = var_12266_to_fp16)[name = tensor<string, []>("aw_chunk_1147_cast")];
            tensor<string, []> var_12269_equation_0 = const()[name = tensor<string, []>("op_12269_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12269_cast = einsum(equation = var_12269_equation_0, values = (var_12099_cast, var_12049_cast))[name = tensor<string, []>("op_12269_cast")];
            tensor<fp16, []> var_12270_to_fp16 = const()[name = tensor<string, []>("op_12270_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1149_cast = mul(x = var_12269_cast, y = var_12270_to_fp16)[name = tensor<string, []>("aw_chunk_1149_cast")];
            tensor<string, []> var_12273_equation_0 = const()[name = tensor<string, []>("op_12273_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12273_cast = einsum(equation = var_12273_equation_0, values = (var_12099_cast, var_12050_cast))[name = tensor<string, []>("op_12273_cast")];
            tensor<fp16, []> var_12274_to_fp16 = const()[name = tensor<string, []>("op_12274_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1151_cast = mul(x = var_12273_cast, y = var_12274_to_fp16)[name = tensor<string, []>("aw_chunk_1151_cast")];
            tensor<string, []> var_12277_equation_0 = const()[name = tensor<string, []>("op_12277_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12277_cast = einsum(equation = var_12277_equation_0, values = (var_12103_cast, var_12051_cast))[name = tensor<string, []>("op_12277_cast")];
            tensor<fp16, []> var_12278_to_fp16 = const()[name = tensor<string, []>("op_12278_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1153_cast = mul(x = var_12277_cast, y = var_12278_to_fp16)[name = tensor<string, []>("aw_chunk_1153_cast")];
            tensor<string, []> var_12281_equation_0 = const()[name = tensor<string, []>("op_12281_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12281_cast = einsum(equation = var_12281_equation_0, values = (var_12103_cast, var_12052_cast))[name = tensor<string, []>("op_12281_cast")];
            tensor<fp16, []> var_12282_to_fp16 = const()[name = tensor<string, []>("op_12282_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1155_cast = mul(x = var_12281_cast, y = var_12282_to_fp16)[name = tensor<string, []>("aw_chunk_1155_cast")];
            tensor<string, []> var_12285_equation_0 = const()[name = tensor<string, []>("op_12285_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12285_cast = einsum(equation = var_12285_equation_0, values = (var_12103_cast, var_12053_cast))[name = tensor<string, []>("op_12285_cast")];
            tensor<fp16, []> var_12286_to_fp16 = const()[name = tensor<string, []>("op_12286_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1157_cast = mul(x = var_12285_cast, y = var_12286_to_fp16)[name = tensor<string, []>("aw_chunk_1157_cast")];
            tensor<string, []> var_12289_equation_0 = const()[name = tensor<string, []>("op_12289_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12289_cast = einsum(equation = var_12289_equation_0, values = (var_12103_cast, var_12054_cast))[name = tensor<string, []>("op_12289_cast")];
            tensor<fp16, []> var_12290_to_fp16 = const()[name = tensor<string, []>("op_12290_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1159_cast = mul(x = var_12289_cast, y = var_12290_to_fp16)[name = tensor<string, []>("aw_chunk_1159_cast")];
            tensor<string, []> var_12293_equation_0 = const()[name = tensor<string, []>("op_12293_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12293_cast = einsum(equation = var_12293_equation_0, values = (var_12103_cast, var_12055_cast))[name = tensor<string, []>("op_12293_cast")];
            tensor<fp16, []> var_12294_to_fp16 = const()[name = tensor<string, []>("op_12294_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1161_cast = mul(x = var_12293_cast, y = var_12294_to_fp16)[name = tensor<string, []>("aw_chunk_1161_cast")];
            tensor<string, []> var_12297_equation_0 = const()[name = tensor<string, []>("op_12297_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12297_cast = einsum(equation = var_12297_equation_0, values = (var_12103_cast, var_12056_cast))[name = tensor<string, []>("op_12297_cast")];
            tensor<fp16, []> var_12298_to_fp16 = const()[name = tensor<string, []>("op_12298_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1163_cast = mul(x = var_12297_cast, y = var_12298_to_fp16)[name = tensor<string, []>("aw_chunk_1163_cast")];
            tensor<string, []> var_12301_equation_0 = const()[name = tensor<string, []>("op_12301_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12301_cast = einsum(equation = var_12301_equation_0, values = (var_12103_cast, var_12057_cast))[name = tensor<string, []>("op_12301_cast")];
            tensor<fp16, []> var_12302_to_fp16 = const()[name = tensor<string, []>("op_12302_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1165_cast = mul(x = var_12301_cast, y = var_12302_to_fp16)[name = tensor<string, []>("aw_chunk_1165_cast")];
            tensor<string, []> var_12305_equation_0 = const()[name = tensor<string, []>("op_12305_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12305_cast = einsum(equation = var_12305_equation_0, values = (var_12103_cast, var_12058_cast))[name = tensor<string, []>("op_12305_cast")];
            tensor<fp16, []> var_12306_to_fp16 = const()[name = tensor<string, []>("op_12306_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1167_cast = mul(x = var_12305_cast, y = var_12306_to_fp16)[name = tensor<string, []>("aw_chunk_1167_cast")];
            tensor<string, []> var_12309_equation_0 = const()[name = tensor<string, []>("op_12309_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12309_cast = einsum(equation = var_12309_equation_0, values = (var_12107_cast, var_12059_cast))[name = tensor<string, []>("op_12309_cast")];
            tensor<fp16, []> var_12310_to_fp16 = const()[name = tensor<string, []>("op_12310_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1169_cast = mul(x = var_12309_cast, y = var_12310_to_fp16)[name = tensor<string, []>("aw_chunk_1169_cast")];
            tensor<string, []> var_12313_equation_0 = const()[name = tensor<string, []>("op_12313_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12313_cast = einsum(equation = var_12313_equation_0, values = (var_12107_cast, var_12060_cast))[name = tensor<string, []>("op_12313_cast")];
            tensor<fp16, []> var_12314_to_fp16 = const()[name = tensor<string, []>("op_12314_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1171_cast = mul(x = var_12313_cast, y = var_12314_to_fp16)[name = tensor<string, []>("aw_chunk_1171_cast")];
            tensor<string, []> var_12317_equation_0 = const()[name = tensor<string, []>("op_12317_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12317_cast = einsum(equation = var_12317_equation_0, values = (var_12107_cast, var_12061_cast))[name = tensor<string, []>("op_12317_cast")];
            tensor<fp16, []> var_12318_to_fp16 = const()[name = tensor<string, []>("op_12318_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1173_cast = mul(x = var_12317_cast, y = var_12318_to_fp16)[name = tensor<string, []>("aw_chunk_1173_cast")];
            tensor<string, []> var_12321_equation_0 = const()[name = tensor<string, []>("op_12321_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12321_cast = einsum(equation = var_12321_equation_0, values = (var_12107_cast, var_12062_cast))[name = tensor<string, []>("op_12321_cast")];
            tensor<fp16, []> var_12322_to_fp16 = const()[name = tensor<string, []>("op_12322_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1175_cast = mul(x = var_12321_cast, y = var_12322_to_fp16)[name = tensor<string, []>("aw_chunk_1175_cast")];
            tensor<string, []> var_12325_equation_0 = const()[name = tensor<string, []>("op_12325_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12325_cast = einsum(equation = var_12325_equation_0, values = (var_12107_cast, var_12063_cast))[name = tensor<string, []>("op_12325_cast")];
            tensor<fp16, []> var_12326_to_fp16 = const()[name = tensor<string, []>("op_12326_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1177_cast = mul(x = var_12325_cast, y = var_12326_to_fp16)[name = tensor<string, []>("aw_chunk_1177_cast")];
            tensor<string, []> var_12329_equation_0 = const()[name = tensor<string, []>("op_12329_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12329_cast = einsum(equation = var_12329_equation_0, values = (var_12107_cast, var_12064_cast))[name = tensor<string, []>("op_12329_cast")];
            tensor<fp16, []> var_12330_to_fp16 = const()[name = tensor<string, []>("op_12330_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1179_cast = mul(x = var_12329_cast, y = var_12330_to_fp16)[name = tensor<string, []>("aw_chunk_1179_cast")];
            tensor<string, []> var_12333_equation_0 = const()[name = tensor<string, []>("op_12333_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12333_cast = einsum(equation = var_12333_equation_0, values = (var_12107_cast, var_12065_cast))[name = tensor<string, []>("op_12333_cast")];
            tensor<fp16, []> var_12334_to_fp16 = const()[name = tensor<string, []>("op_12334_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1181_cast = mul(x = var_12333_cast, y = var_12334_to_fp16)[name = tensor<string, []>("aw_chunk_1181_cast")];
            tensor<string, []> var_12337_equation_0 = const()[name = tensor<string, []>("op_12337_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12337_cast = einsum(equation = var_12337_equation_0, values = (var_12107_cast, var_12066_cast))[name = tensor<string, []>("op_12337_cast")];
            tensor<fp16, []> var_12338_to_fp16 = const()[name = tensor<string, []>("op_12338_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1183_cast = mul(x = var_12337_cast, y = var_12338_to_fp16)[name = tensor<string, []>("aw_chunk_1183_cast")];
            tensor<string, []> var_12341_equation_0 = const()[name = tensor<string, []>("op_12341_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12341_cast = einsum(equation = var_12341_equation_0, values = (var_12111_cast, var_12067_cast))[name = tensor<string, []>("op_12341_cast")];
            tensor<fp16, []> var_12342_to_fp16 = const()[name = tensor<string, []>("op_12342_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1185_cast = mul(x = var_12341_cast, y = var_12342_to_fp16)[name = tensor<string, []>("aw_chunk_1185_cast")];
            tensor<string, []> var_12345_equation_0 = const()[name = tensor<string, []>("op_12345_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12345_cast = einsum(equation = var_12345_equation_0, values = (var_12111_cast, var_12068_cast))[name = tensor<string, []>("op_12345_cast")];
            tensor<fp16, []> var_12346_to_fp16 = const()[name = tensor<string, []>("op_12346_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1187_cast = mul(x = var_12345_cast, y = var_12346_to_fp16)[name = tensor<string, []>("aw_chunk_1187_cast")];
            tensor<string, []> var_12349_equation_0 = const()[name = tensor<string, []>("op_12349_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12349_cast = einsum(equation = var_12349_equation_0, values = (var_12111_cast, var_12069_cast))[name = tensor<string, []>("op_12349_cast")];
            tensor<fp16, []> var_12350_to_fp16 = const()[name = tensor<string, []>("op_12350_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1189_cast = mul(x = var_12349_cast, y = var_12350_to_fp16)[name = tensor<string, []>("aw_chunk_1189_cast")];
            tensor<string, []> var_12353_equation_0 = const()[name = tensor<string, []>("op_12353_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12353_cast = einsum(equation = var_12353_equation_0, values = (var_12111_cast, var_12070_cast))[name = tensor<string, []>("op_12353_cast")];
            tensor<fp16, []> var_12354_to_fp16 = const()[name = tensor<string, []>("op_12354_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1191_cast = mul(x = var_12353_cast, y = var_12354_to_fp16)[name = tensor<string, []>("aw_chunk_1191_cast")];
            tensor<string, []> var_12357_equation_0 = const()[name = tensor<string, []>("op_12357_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12357_cast = einsum(equation = var_12357_equation_0, values = (var_12111_cast, var_12071_cast))[name = tensor<string, []>("op_12357_cast")];
            tensor<fp16, []> var_12358_to_fp16 = const()[name = tensor<string, []>("op_12358_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1193_cast = mul(x = var_12357_cast, y = var_12358_to_fp16)[name = tensor<string, []>("aw_chunk_1193_cast")];
            tensor<string, []> var_12361_equation_0 = const()[name = tensor<string, []>("op_12361_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12361_cast = einsum(equation = var_12361_equation_0, values = (var_12111_cast, var_12072_cast))[name = tensor<string, []>("op_12361_cast")];
            tensor<fp16, []> var_12362_to_fp16 = const()[name = tensor<string, []>("op_12362_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1195_cast = mul(x = var_12361_cast, y = var_12362_to_fp16)[name = tensor<string, []>("aw_chunk_1195_cast")];
            tensor<string, []> var_12365_equation_0 = const()[name = tensor<string, []>("op_12365_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12365_cast = einsum(equation = var_12365_equation_0, values = (var_12111_cast, var_12073_cast))[name = tensor<string, []>("op_12365_cast")];
            tensor<fp16, []> var_12366_to_fp16 = const()[name = tensor<string, []>("op_12366_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1197_cast = mul(x = var_12365_cast, y = var_12366_to_fp16)[name = tensor<string, []>("aw_chunk_1197_cast")];
            tensor<string, []> var_12369_equation_0 = const()[name = tensor<string, []>("op_12369_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12369_cast = einsum(equation = var_12369_equation_0, values = (var_12111_cast, var_12074_cast))[name = tensor<string, []>("op_12369_cast")];
            tensor<fp16, []> var_12370_to_fp16 = const()[name = tensor<string, []>("op_12370_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1199_cast = mul(x = var_12369_cast, y = var_12370_to_fp16)[name = tensor<string, []>("aw_chunk_1199_cast")];
            tensor<string, []> var_12373_equation_0 = const()[name = tensor<string, []>("op_12373_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12373_cast = einsum(equation = var_12373_equation_0, values = (var_12115_cast, var_12075_cast))[name = tensor<string, []>("op_12373_cast")];
            tensor<fp16, []> var_12374_to_fp16 = const()[name = tensor<string, []>("op_12374_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1201_cast = mul(x = var_12373_cast, y = var_12374_to_fp16)[name = tensor<string, []>("aw_chunk_1201_cast")];
            tensor<string, []> var_12377_equation_0 = const()[name = tensor<string, []>("op_12377_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12377_cast = einsum(equation = var_12377_equation_0, values = (var_12115_cast, var_12076_cast))[name = tensor<string, []>("op_12377_cast")];
            tensor<fp16, []> var_12378_to_fp16 = const()[name = tensor<string, []>("op_12378_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1203_cast = mul(x = var_12377_cast, y = var_12378_to_fp16)[name = tensor<string, []>("aw_chunk_1203_cast")];
            tensor<string, []> var_12381_equation_0 = const()[name = tensor<string, []>("op_12381_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12381_cast = einsum(equation = var_12381_equation_0, values = (var_12115_cast, var_12077_cast))[name = tensor<string, []>("op_12381_cast")];
            tensor<fp16, []> var_12382_to_fp16 = const()[name = tensor<string, []>("op_12382_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1205_cast = mul(x = var_12381_cast, y = var_12382_to_fp16)[name = tensor<string, []>("aw_chunk_1205_cast")];
            tensor<string, []> var_12385_equation_0 = const()[name = tensor<string, []>("op_12385_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12385_cast = einsum(equation = var_12385_equation_0, values = (var_12115_cast, var_12078_cast))[name = tensor<string, []>("op_12385_cast")];
            tensor<fp16, []> var_12386_to_fp16 = const()[name = tensor<string, []>("op_12386_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1207_cast = mul(x = var_12385_cast, y = var_12386_to_fp16)[name = tensor<string, []>("aw_chunk_1207_cast")];
            tensor<string, []> var_12389_equation_0 = const()[name = tensor<string, []>("op_12389_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12389_cast = einsum(equation = var_12389_equation_0, values = (var_12115_cast, var_12079_cast))[name = tensor<string, []>("op_12389_cast")];
            tensor<fp16, []> var_12390_to_fp16 = const()[name = tensor<string, []>("op_12390_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1209_cast = mul(x = var_12389_cast, y = var_12390_to_fp16)[name = tensor<string, []>("aw_chunk_1209_cast")];
            tensor<string, []> var_12393_equation_0 = const()[name = tensor<string, []>("op_12393_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12393_cast = einsum(equation = var_12393_equation_0, values = (var_12115_cast, var_12080_cast))[name = tensor<string, []>("op_12393_cast")];
            tensor<fp16, []> var_12394_to_fp16 = const()[name = tensor<string, []>("op_12394_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1211_cast = mul(x = var_12393_cast, y = var_12394_to_fp16)[name = tensor<string, []>("aw_chunk_1211_cast")];
            tensor<string, []> var_12397_equation_0 = const()[name = tensor<string, []>("op_12397_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12397_cast = einsum(equation = var_12397_equation_0, values = (var_12115_cast, var_12081_cast))[name = tensor<string, []>("op_12397_cast")];
            tensor<fp16, []> var_12398_to_fp16 = const()[name = tensor<string, []>("op_12398_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1213_cast = mul(x = var_12397_cast, y = var_12398_to_fp16)[name = tensor<string, []>("aw_chunk_1213_cast")];
            tensor<string, []> var_12401_equation_0 = const()[name = tensor<string, []>("op_12401_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_12401_cast = einsum(equation = var_12401_equation_0, values = (var_12115_cast, var_12082_cast))[name = tensor<string, []>("op_12401_cast")];
            tensor<fp16, []> var_12402_to_fp16 = const()[name = tensor<string, []>("op_12402_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1215_cast = mul(x = var_12401_cast, y = var_12402_to_fp16)[name = tensor<string, []>("aw_chunk_1215_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12404_cast = softmax(axis = var_10375, x = aw_chunk_1089_cast)[name = tensor<string, []>("op_12404_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12405_cast = softmax(axis = var_10375, x = aw_chunk_1091_cast)[name = tensor<string, []>("op_12405_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12406_cast = softmax(axis = var_10375, x = aw_chunk_1093_cast)[name = tensor<string, []>("op_12406_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12407_cast = softmax(axis = var_10375, x = aw_chunk_1095_cast)[name = tensor<string, []>("op_12407_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12408_cast = softmax(axis = var_10375, x = aw_chunk_1097_cast)[name = tensor<string, []>("op_12408_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12409_cast = softmax(axis = var_10375, x = aw_chunk_1099_cast)[name = tensor<string, []>("op_12409_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12410_cast = softmax(axis = var_10375, x = aw_chunk_1101_cast)[name = tensor<string, []>("op_12410_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12411_cast = softmax(axis = var_10375, x = aw_chunk_1103_cast)[name = tensor<string, []>("op_12411_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12412_cast = softmax(axis = var_10375, x = aw_chunk_1105_cast)[name = tensor<string, []>("op_12412_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12413_cast = softmax(axis = var_10375, x = aw_chunk_1107_cast)[name = tensor<string, []>("op_12413_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12414_cast = softmax(axis = var_10375, x = aw_chunk_1109_cast)[name = tensor<string, []>("op_12414_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12415_cast = softmax(axis = var_10375, x = aw_chunk_1111_cast)[name = tensor<string, []>("op_12415_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12416_cast = softmax(axis = var_10375, x = aw_chunk_1113_cast)[name = tensor<string, []>("op_12416_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12417_cast = softmax(axis = var_10375, x = aw_chunk_1115_cast)[name = tensor<string, []>("op_12417_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12418_cast = softmax(axis = var_10375, x = aw_chunk_1117_cast)[name = tensor<string, []>("op_12418_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12419_cast = softmax(axis = var_10375, x = aw_chunk_1119_cast)[name = tensor<string, []>("op_12419_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12420_cast = softmax(axis = var_10375, x = aw_chunk_1121_cast)[name = tensor<string, []>("op_12420_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12421_cast = softmax(axis = var_10375, x = aw_chunk_1123_cast)[name = tensor<string, []>("op_12421_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12422_cast = softmax(axis = var_10375, x = aw_chunk_1125_cast)[name = tensor<string, []>("op_12422_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12423_cast = softmax(axis = var_10375, x = aw_chunk_1127_cast)[name = tensor<string, []>("op_12423_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12424_cast = softmax(axis = var_10375, x = aw_chunk_1129_cast)[name = tensor<string, []>("op_12424_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12425_cast = softmax(axis = var_10375, x = aw_chunk_1131_cast)[name = tensor<string, []>("op_12425_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12426_cast = softmax(axis = var_10375, x = aw_chunk_1133_cast)[name = tensor<string, []>("op_12426_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12427_cast = softmax(axis = var_10375, x = aw_chunk_1135_cast)[name = tensor<string, []>("op_12427_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12428_cast = softmax(axis = var_10375, x = aw_chunk_1137_cast)[name = tensor<string, []>("op_12428_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12429_cast = softmax(axis = var_10375, x = aw_chunk_1139_cast)[name = tensor<string, []>("op_12429_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12430_cast = softmax(axis = var_10375, x = aw_chunk_1141_cast)[name = tensor<string, []>("op_12430_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12431_cast = softmax(axis = var_10375, x = aw_chunk_1143_cast)[name = tensor<string, []>("op_12431_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12432_cast = softmax(axis = var_10375, x = aw_chunk_1145_cast)[name = tensor<string, []>("op_12432_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12433_cast = softmax(axis = var_10375, x = aw_chunk_1147_cast)[name = tensor<string, []>("op_12433_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12434_cast = softmax(axis = var_10375, x = aw_chunk_1149_cast)[name = tensor<string, []>("op_12434_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12435_cast = softmax(axis = var_10375, x = aw_chunk_1151_cast)[name = tensor<string, []>("op_12435_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12436_cast = softmax(axis = var_10375, x = aw_chunk_1153_cast)[name = tensor<string, []>("op_12436_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12437_cast = softmax(axis = var_10375, x = aw_chunk_1155_cast)[name = tensor<string, []>("op_12437_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12438_cast = softmax(axis = var_10375, x = aw_chunk_1157_cast)[name = tensor<string, []>("op_12438_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12439_cast = softmax(axis = var_10375, x = aw_chunk_1159_cast)[name = tensor<string, []>("op_12439_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12440_cast = softmax(axis = var_10375, x = aw_chunk_1161_cast)[name = tensor<string, []>("op_12440_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12441_cast = softmax(axis = var_10375, x = aw_chunk_1163_cast)[name = tensor<string, []>("op_12441_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12442_cast = softmax(axis = var_10375, x = aw_chunk_1165_cast)[name = tensor<string, []>("op_12442_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12443_cast = softmax(axis = var_10375, x = aw_chunk_1167_cast)[name = tensor<string, []>("op_12443_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12444_cast = softmax(axis = var_10375, x = aw_chunk_1169_cast)[name = tensor<string, []>("op_12444_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12445_cast = softmax(axis = var_10375, x = aw_chunk_1171_cast)[name = tensor<string, []>("op_12445_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12446_cast = softmax(axis = var_10375, x = aw_chunk_1173_cast)[name = tensor<string, []>("op_12446_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12447_cast = softmax(axis = var_10375, x = aw_chunk_1175_cast)[name = tensor<string, []>("op_12447_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12448_cast = softmax(axis = var_10375, x = aw_chunk_1177_cast)[name = tensor<string, []>("op_12448_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12449_cast = softmax(axis = var_10375, x = aw_chunk_1179_cast)[name = tensor<string, []>("op_12449_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12450_cast = softmax(axis = var_10375, x = aw_chunk_1181_cast)[name = tensor<string, []>("op_12450_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12451_cast = softmax(axis = var_10375, x = aw_chunk_1183_cast)[name = tensor<string, []>("op_12451_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12452_cast = softmax(axis = var_10375, x = aw_chunk_1185_cast)[name = tensor<string, []>("op_12452_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12453_cast = softmax(axis = var_10375, x = aw_chunk_1187_cast)[name = tensor<string, []>("op_12453_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12454_cast = softmax(axis = var_10375, x = aw_chunk_1189_cast)[name = tensor<string, []>("op_12454_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12455_cast = softmax(axis = var_10375, x = aw_chunk_1191_cast)[name = tensor<string, []>("op_12455_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12456_cast = softmax(axis = var_10375, x = aw_chunk_1193_cast)[name = tensor<string, []>("op_12456_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12457_cast = softmax(axis = var_10375, x = aw_chunk_1195_cast)[name = tensor<string, []>("op_12457_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12458_cast = softmax(axis = var_10375, x = aw_chunk_1197_cast)[name = tensor<string, []>("op_12458_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12459_cast = softmax(axis = var_10375, x = aw_chunk_1199_cast)[name = tensor<string, []>("op_12459_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12460_cast = softmax(axis = var_10375, x = aw_chunk_1201_cast)[name = tensor<string, []>("op_12460_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12461_cast = softmax(axis = var_10375, x = aw_chunk_1203_cast)[name = tensor<string, []>("op_12461_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12462_cast = softmax(axis = var_10375, x = aw_chunk_1205_cast)[name = tensor<string, []>("op_12462_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12463_cast = softmax(axis = var_10375, x = aw_chunk_1207_cast)[name = tensor<string, []>("op_12463_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12464_cast = softmax(axis = var_10375, x = aw_chunk_1209_cast)[name = tensor<string, []>("op_12464_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12465_cast = softmax(axis = var_10375, x = aw_chunk_1211_cast)[name = tensor<string, []>("op_12465_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12466_cast = softmax(axis = var_10375, x = aw_chunk_1213_cast)[name = tensor<string, []>("op_12466_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_12467_cast = softmax(axis = var_10375, x = aw_chunk_1215_cast)[name = tensor<string, []>("op_12467_cast")];
            tensor<string, []> var_12469_equation_0 = const()[name = tensor<string, []>("op_12469_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12469_cast = einsum(equation = var_12469_equation_0, values = (var_12117_cast, var_12404_cast))[name = tensor<string, []>("op_12469_cast")];
            tensor<string, []> var_12471_equation_0 = const()[name = tensor<string, []>("op_12471_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12471_cast = einsum(equation = var_12471_equation_0, values = (var_12117_cast, var_12405_cast))[name = tensor<string, []>("op_12471_cast")];
            tensor<string, []> var_12473_equation_0 = const()[name = tensor<string, []>("op_12473_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12473_cast = einsum(equation = var_12473_equation_0, values = (var_12117_cast, var_12406_cast))[name = tensor<string, []>("op_12473_cast")];
            tensor<string, []> var_12475_equation_0 = const()[name = tensor<string, []>("op_12475_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12475_cast = einsum(equation = var_12475_equation_0, values = (var_12117_cast, var_12407_cast))[name = tensor<string, []>("op_12475_cast")];
            tensor<string, []> var_12477_equation_0 = const()[name = tensor<string, []>("op_12477_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12477_cast = einsum(equation = var_12477_equation_0, values = (var_12117_cast, var_12408_cast))[name = tensor<string, []>("op_12477_cast")];
            tensor<string, []> var_12479_equation_0 = const()[name = tensor<string, []>("op_12479_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12479_cast = einsum(equation = var_12479_equation_0, values = (var_12117_cast, var_12409_cast))[name = tensor<string, []>("op_12479_cast")];
            tensor<string, []> var_12481_equation_0 = const()[name = tensor<string, []>("op_12481_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12481_cast = einsum(equation = var_12481_equation_0, values = (var_12117_cast, var_12410_cast))[name = tensor<string, []>("op_12481_cast")];
            tensor<string, []> var_12483_equation_0 = const()[name = tensor<string, []>("op_12483_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12483_cast = einsum(equation = var_12483_equation_0, values = (var_12117_cast, var_12411_cast))[name = tensor<string, []>("op_12483_cast")];
            tensor<string, []> var_12485_equation_0 = const()[name = tensor<string, []>("op_12485_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12485_cast = einsum(equation = var_12485_equation_0, values = (var_12121_cast, var_12412_cast))[name = tensor<string, []>("op_12485_cast")];
            tensor<string, []> var_12487_equation_0 = const()[name = tensor<string, []>("op_12487_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12487_cast = einsum(equation = var_12487_equation_0, values = (var_12121_cast, var_12413_cast))[name = tensor<string, []>("op_12487_cast")];
            tensor<string, []> var_12489_equation_0 = const()[name = tensor<string, []>("op_12489_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12489_cast = einsum(equation = var_12489_equation_0, values = (var_12121_cast, var_12414_cast))[name = tensor<string, []>("op_12489_cast")];
            tensor<string, []> var_12491_equation_0 = const()[name = tensor<string, []>("op_12491_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12491_cast = einsum(equation = var_12491_equation_0, values = (var_12121_cast, var_12415_cast))[name = tensor<string, []>("op_12491_cast")];
            tensor<string, []> var_12493_equation_0 = const()[name = tensor<string, []>("op_12493_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12493_cast = einsum(equation = var_12493_equation_0, values = (var_12121_cast, var_12416_cast))[name = tensor<string, []>("op_12493_cast")];
            tensor<string, []> var_12495_equation_0 = const()[name = tensor<string, []>("op_12495_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12495_cast = einsum(equation = var_12495_equation_0, values = (var_12121_cast, var_12417_cast))[name = tensor<string, []>("op_12495_cast")];
            tensor<string, []> var_12497_equation_0 = const()[name = tensor<string, []>("op_12497_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12497_cast = einsum(equation = var_12497_equation_0, values = (var_12121_cast, var_12418_cast))[name = tensor<string, []>("op_12497_cast")];
            tensor<string, []> var_12499_equation_0 = const()[name = tensor<string, []>("op_12499_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12499_cast = einsum(equation = var_12499_equation_0, values = (var_12121_cast, var_12419_cast))[name = tensor<string, []>("op_12499_cast")];
            tensor<string, []> var_12501_equation_0 = const()[name = tensor<string, []>("op_12501_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12501_cast = einsum(equation = var_12501_equation_0, values = (var_12125_cast, var_12420_cast))[name = tensor<string, []>("op_12501_cast")];
            tensor<string, []> var_12503_equation_0 = const()[name = tensor<string, []>("op_12503_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12503_cast = einsum(equation = var_12503_equation_0, values = (var_12125_cast, var_12421_cast))[name = tensor<string, []>("op_12503_cast")];
            tensor<string, []> var_12505_equation_0 = const()[name = tensor<string, []>("op_12505_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12505_cast = einsum(equation = var_12505_equation_0, values = (var_12125_cast, var_12422_cast))[name = tensor<string, []>("op_12505_cast")];
            tensor<string, []> var_12507_equation_0 = const()[name = tensor<string, []>("op_12507_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12507_cast = einsum(equation = var_12507_equation_0, values = (var_12125_cast, var_12423_cast))[name = tensor<string, []>("op_12507_cast")];
            tensor<string, []> var_12509_equation_0 = const()[name = tensor<string, []>("op_12509_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12509_cast = einsum(equation = var_12509_equation_0, values = (var_12125_cast, var_12424_cast))[name = tensor<string, []>("op_12509_cast")];
            tensor<string, []> var_12511_equation_0 = const()[name = tensor<string, []>("op_12511_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12511_cast = einsum(equation = var_12511_equation_0, values = (var_12125_cast, var_12425_cast))[name = tensor<string, []>("op_12511_cast")];
            tensor<string, []> var_12513_equation_0 = const()[name = tensor<string, []>("op_12513_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12513_cast = einsum(equation = var_12513_equation_0, values = (var_12125_cast, var_12426_cast))[name = tensor<string, []>("op_12513_cast")];
            tensor<string, []> var_12515_equation_0 = const()[name = tensor<string, []>("op_12515_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12515_cast = einsum(equation = var_12515_equation_0, values = (var_12125_cast, var_12427_cast))[name = tensor<string, []>("op_12515_cast")];
            tensor<string, []> var_12517_equation_0 = const()[name = tensor<string, []>("op_12517_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12517_cast = einsum(equation = var_12517_equation_0, values = (var_12129_cast, var_12428_cast))[name = tensor<string, []>("op_12517_cast")];
            tensor<string, []> var_12519_equation_0 = const()[name = tensor<string, []>("op_12519_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12519_cast = einsum(equation = var_12519_equation_0, values = (var_12129_cast, var_12429_cast))[name = tensor<string, []>("op_12519_cast")];
            tensor<string, []> var_12521_equation_0 = const()[name = tensor<string, []>("op_12521_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12521_cast = einsum(equation = var_12521_equation_0, values = (var_12129_cast, var_12430_cast))[name = tensor<string, []>("op_12521_cast")];
            tensor<string, []> var_12523_equation_0 = const()[name = tensor<string, []>("op_12523_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12523_cast = einsum(equation = var_12523_equation_0, values = (var_12129_cast, var_12431_cast))[name = tensor<string, []>("op_12523_cast")];
            tensor<string, []> var_12525_equation_0 = const()[name = tensor<string, []>("op_12525_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12525_cast = einsum(equation = var_12525_equation_0, values = (var_12129_cast, var_12432_cast))[name = tensor<string, []>("op_12525_cast")];
            tensor<string, []> var_12527_equation_0 = const()[name = tensor<string, []>("op_12527_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12527_cast = einsum(equation = var_12527_equation_0, values = (var_12129_cast, var_12433_cast))[name = tensor<string, []>("op_12527_cast")];
            tensor<string, []> var_12529_equation_0 = const()[name = tensor<string, []>("op_12529_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12529_cast = einsum(equation = var_12529_equation_0, values = (var_12129_cast, var_12434_cast))[name = tensor<string, []>("op_12529_cast")];
            tensor<string, []> var_12531_equation_0 = const()[name = tensor<string, []>("op_12531_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12531_cast = einsum(equation = var_12531_equation_0, values = (var_12129_cast, var_12435_cast))[name = tensor<string, []>("op_12531_cast")];
            tensor<string, []> var_12533_equation_0 = const()[name = tensor<string, []>("op_12533_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12533_cast = einsum(equation = var_12533_equation_0, values = (var_12133_cast, var_12436_cast))[name = tensor<string, []>("op_12533_cast")];
            tensor<string, []> var_12535_equation_0 = const()[name = tensor<string, []>("op_12535_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12535_cast = einsum(equation = var_12535_equation_0, values = (var_12133_cast, var_12437_cast))[name = tensor<string, []>("op_12535_cast")];
            tensor<string, []> var_12537_equation_0 = const()[name = tensor<string, []>("op_12537_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12537_cast = einsum(equation = var_12537_equation_0, values = (var_12133_cast, var_12438_cast))[name = tensor<string, []>("op_12537_cast")];
            tensor<string, []> var_12539_equation_0 = const()[name = tensor<string, []>("op_12539_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12539_cast = einsum(equation = var_12539_equation_0, values = (var_12133_cast, var_12439_cast))[name = tensor<string, []>("op_12539_cast")];
            tensor<string, []> var_12541_equation_0 = const()[name = tensor<string, []>("op_12541_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12541_cast = einsum(equation = var_12541_equation_0, values = (var_12133_cast, var_12440_cast))[name = tensor<string, []>("op_12541_cast")];
            tensor<string, []> var_12543_equation_0 = const()[name = tensor<string, []>("op_12543_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12543_cast = einsum(equation = var_12543_equation_0, values = (var_12133_cast, var_12441_cast))[name = tensor<string, []>("op_12543_cast")];
            tensor<string, []> var_12545_equation_0 = const()[name = tensor<string, []>("op_12545_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12545_cast = einsum(equation = var_12545_equation_0, values = (var_12133_cast, var_12442_cast))[name = tensor<string, []>("op_12545_cast")];
            tensor<string, []> var_12547_equation_0 = const()[name = tensor<string, []>("op_12547_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12547_cast = einsum(equation = var_12547_equation_0, values = (var_12133_cast, var_12443_cast))[name = tensor<string, []>("op_12547_cast")];
            tensor<string, []> var_12549_equation_0 = const()[name = tensor<string, []>("op_12549_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12549_cast = einsum(equation = var_12549_equation_0, values = (var_12137_cast, var_12444_cast))[name = tensor<string, []>("op_12549_cast")];
            tensor<string, []> var_12551_equation_0 = const()[name = tensor<string, []>("op_12551_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12551_cast = einsum(equation = var_12551_equation_0, values = (var_12137_cast, var_12445_cast))[name = tensor<string, []>("op_12551_cast")];
            tensor<string, []> var_12553_equation_0 = const()[name = tensor<string, []>("op_12553_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12553_cast = einsum(equation = var_12553_equation_0, values = (var_12137_cast, var_12446_cast))[name = tensor<string, []>("op_12553_cast")];
            tensor<string, []> var_12555_equation_0 = const()[name = tensor<string, []>("op_12555_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12555_cast = einsum(equation = var_12555_equation_0, values = (var_12137_cast, var_12447_cast))[name = tensor<string, []>("op_12555_cast")];
            tensor<string, []> var_12557_equation_0 = const()[name = tensor<string, []>("op_12557_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12557_cast = einsum(equation = var_12557_equation_0, values = (var_12137_cast, var_12448_cast))[name = tensor<string, []>("op_12557_cast")];
            tensor<string, []> var_12559_equation_0 = const()[name = tensor<string, []>("op_12559_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12559_cast = einsum(equation = var_12559_equation_0, values = (var_12137_cast, var_12449_cast))[name = tensor<string, []>("op_12559_cast")];
            tensor<string, []> var_12561_equation_0 = const()[name = tensor<string, []>("op_12561_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12561_cast = einsum(equation = var_12561_equation_0, values = (var_12137_cast, var_12450_cast))[name = tensor<string, []>("op_12561_cast")];
            tensor<string, []> var_12563_equation_0 = const()[name = tensor<string, []>("op_12563_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12563_cast = einsum(equation = var_12563_equation_0, values = (var_12137_cast, var_12451_cast))[name = tensor<string, []>("op_12563_cast")];
            tensor<string, []> var_12565_equation_0 = const()[name = tensor<string, []>("op_12565_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12565_cast = einsum(equation = var_12565_equation_0, values = (var_12141_cast, var_12452_cast))[name = tensor<string, []>("op_12565_cast")];
            tensor<string, []> var_12567_equation_0 = const()[name = tensor<string, []>("op_12567_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12567_cast = einsum(equation = var_12567_equation_0, values = (var_12141_cast, var_12453_cast))[name = tensor<string, []>("op_12567_cast")];
            tensor<string, []> var_12569_equation_0 = const()[name = tensor<string, []>("op_12569_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12569_cast = einsum(equation = var_12569_equation_0, values = (var_12141_cast, var_12454_cast))[name = tensor<string, []>("op_12569_cast")];
            tensor<string, []> var_12571_equation_0 = const()[name = tensor<string, []>("op_12571_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12571_cast = einsum(equation = var_12571_equation_0, values = (var_12141_cast, var_12455_cast))[name = tensor<string, []>("op_12571_cast")];
            tensor<string, []> var_12573_equation_0 = const()[name = tensor<string, []>("op_12573_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12573_cast = einsum(equation = var_12573_equation_0, values = (var_12141_cast, var_12456_cast))[name = tensor<string, []>("op_12573_cast")];
            tensor<string, []> var_12575_equation_0 = const()[name = tensor<string, []>("op_12575_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12575_cast = einsum(equation = var_12575_equation_0, values = (var_12141_cast, var_12457_cast))[name = tensor<string, []>("op_12575_cast")];
            tensor<string, []> var_12577_equation_0 = const()[name = tensor<string, []>("op_12577_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12577_cast = einsum(equation = var_12577_equation_0, values = (var_12141_cast, var_12458_cast))[name = tensor<string, []>("op_12577_cast")];
            tensor<string, []> var_12579_equation_0 = const()[name = tensor<string, []>("op_12579_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12579_cast = einsum(equation = var_12579_equation_0, values = (var_12141_cast, var_12459_cast))[name = tensor<string, []>("op_12579_cast")];
            tensor<string, []> var_12581_equation_0 = const()[name = tensor<string, []>("op_12581_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12581_cast = einsum(equation = var_12581_equation_0, values = (var_12145_cast, var_12460_cast))[name = tensor<string, []>("op_12581_cast")];
            tensor<string, []> var_12583_equation_0 = const()[name = tensor<string, []>("op_12583_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12583_cast = einsum(equation = var_12583_equation_0, values = (var_12145_cast, var_12461_cast))[name = tensor<string, []>("op_12583_cast")];
            tensor<string, []> var_12585_equation_0 = const()[name = tensor<string, []>("op_12585_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12585_cast = einsum(equation = var_12585_equation_0, values = (var_12145_cast, var_12462_cast))[name = tensor<string, []>("op_12585_cast")];
            tensor<string, []> var_12587_equation_0 = const()[name = tensor<string, []>("op_12587_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12587_cast = einsum(equation = var_12587_equation_0, values = (var_12145_cast, var_12463_cast))[name = tensor<string, []>("op_12587_cast")];
            tensor<string, []> var_12589_equation_0 = const()[name = tensor<string, []>("op_12589_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12589_cast = einsum(equation = var_12589_equation_0, values = (var_12145_cast, var_12464_cast))[name = tensor<string, []>("op_12589_cast")];
            tensor<string, []> var_12591_equation_0 = const()[name = tensor<string, []>("op_12591_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12591_cast = einsum(equation = var_12591_equation_0, values = (var_12145_cast, var_12465_cast))[name = tensor<string, []>("op_12591_cast")];
            tensor<string, []> var_12593_equation_0 = const()[name = tensor<string, []>("op_12593_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12593_cast = einsum(equation = var_12593_equation_0, values = (var_12145_cast, var_12466_cast))[name = tensor<string, []>("op_12593_cast")];
            tensor<string, []> var_12595_equation_0 = const()[name = tensor<string, []>("op_12595_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_12595_cast = einsum(equation = var_12595_equation_0, values = (var_12145_cast, var_12467_cast))[name = tensor<string, []>("op_12595_cast")];
            tensor<bool, []> var_12597_interleave_0 = const()[name = tensor<string, []>("op_12597_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_12597_cast = concat(axis = var_10347, interleave = var_12597_interleave_0, values = (var_12469_cast, var_12471_cast, var_12473_cast, var_12475_cast, var_12477_cast, var_12479_cast, var_12481_cast, var_12483_cast))[name = tensor<string, []>("op_12597_cast")];
            tensor<bool, []> var_12599_interleave_0 = const()[name = tensor<string, []>("op_12599_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_12599_cast = concat(axis = var_10347, interleave = var_12599_interleave_0, values = (var_12485_cast, var_12487_cast, var_12489_cast, var_12491_cast, var_12493_cast, var_12495_cast, var_12497_cast, var_12499_cast))[name = tensor<string, []>("op_12599_cast")];
            tensor<bool, []> var_12601_interleave_0 = const()[name = tensor<string, []>("op_12601_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_12601_cast = concat(axis = var_10347, interleave = var_12601_interleave_0, values = (var_12501_cast, var_12503_cast, var_12505_cast, var_12507_cast, var_12509_cast, var_12511_cast, var_12513_cast, var_12515_cast))[name = tensor<string, []>("op_12601_cast")];
            tensor<bool, []> var_12603_interleave_0 = const()[name = tensor<string, []>("op_12603_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_12603_cast = concat(axis = var_10347, interleave = var_12603_interleave_0, values = (var_12517_cast, var_12519_cast, var_12521_cast, var_12523_cast, var_12525_cast, var_12527_cast, var_12529_cast, var_12531_cast))[name = tensor<string, []>("op_12603_cast")];
            tensor<bool, []> var_12605_interleave_0 = const()[name = tensor<string, []>("op_12605_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_12605_cast = concat(axis = var_10347, interleave = var_12605_interleave_0, values = (var_12533_cast, var_12535_cast, var_12537_cast, var_12539_cast, var_12541_cast, var_12543_cast, var_12545_cast, var_12547_cast))[name = tensor<string, []>("op_12605_cast")];
            tensor<bool, []> var_12607_interleave_0 = const()[name = tensor<string, []>("op_12607_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_12607_cast = concat(axis = var_10347, interleave = var_12607_interleave_0, values = (var_12549_cast, var_12551_cast, var_12553_cast, var_12555_cast, var_12557_cast, var_12559_cast, var_12561_cast, var_12563_cast))[name = tensor<string, []>("op_12607_cast")];
            tensor<bool, []> var_12609_interleave_0 = const()[name = tensor<string, []>("op_12609_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_12609_cast = concat(axis = var_10347, interleave = var_12609_interleave_0, values = (var_12565_cast, var_12567_cast, var_12569_cast, var_12571_cast, var_12573_cast, var_12575_cast, var_12577_cast, var_12579_cast))[name = tensor<string, []>("op_12609_cast")];
            tensor<bool, []> var_12611_interleave_0 = const()[name = tensor<string, []>("op_12611_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_12611_cast = concat(axis = var_10347, interleave = var_12611_interleave_0, values = (var_12581_cast, var_12583_cast, var_12585_cast, var_12587_cast, var_12589_cast, var_12591_cast, var_12593_cast, var_12595_cast))[name = tensor<string, []>("op_12611_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_10375, interleave = input_491_interleave_0, values = (var_12597_cast, var_12599_cast, var_12601_cast, var_12603_cast, var_12605_cast, var_12607_cast, var_12609_cast, var_12611_cast))[name = tensor<string, []>("input_491_cast")];
            tensor<int32, [2]> var_12617 = const()[name = tensor<string, []>("op_12617"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12619 = const()[name = tensor<string, []>("op_12619"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_12621_pad_type_0 = const()[name = tensor<string, []>("op_12621_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_12621_pad_0 = const()[name = tensor<string, []>("op_12621_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(639088640))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(639165504))), name = tensor<string, []>("up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            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, []>(639165696)))];
            tensor<fp16, [2, 320, 1, 4096]> var_12621_cast = conv(bias = up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_12619, groups = var_10375, pad = var_12621_pad_0, pad_type = var_12621_pad_type_0, strides = var_12617, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_491_cast)[name = tensor<string, []>("op_12621_cast")];
            tensor<fp16, [2, 320, 1, 4096]> inputs_87_cast = add(x = var_12621_cast, y = inputs_85_cast)[name = tensor<string, []>("inputs_87_cast")];
            tensor<int32, [1]> var_12625 = const()[name = tensor<string, []>("op_12625"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_87_cast = reduce_mean(axes = var_12625, keep_dims = var_10370, x = inputs_87_cast)[name = tensor<string, []>("channels_mean_87_cast")];
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_87_cast = sub(x = inputs_87_cast, y = channels_mean_87_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_87_cast")];
            tensor<int32, [1]> var_12629 = const()[name = tensor<string, []>("op_12629"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_12630_cast = reduce_mean(axes = var_12629, keep_dims = var_10370, x = zero_mean_sq_87_cast)[name = tensor<string, []>("op_12630_cast")];
            tensor<fp16, []> var_12631_to_fp16 = const()[name = tensor<string, []>("op_12631_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_12632_cast = add(x = var_12630_cast, y = var_12631_to_fp16)[name = tensor<string, []>("op_12632_cast")];
            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_12632_cast)[name = tensor<string, []>("denom_87_cast")];
            tensor<fp16, [2, 320, 1, 4096]> out_87_cast = mul(x = zero_mean_87_cast, y = denom_87_cast)[name = tensor<string, []>("out_87_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_12636_to_fp16 = const()[name = tensor<string, []>("op_12636_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(639166400)))];
            tensor<fp16, [2, 320, 1, 4096]> var_12637_cast = add(x = out_87_cast, y = var_12636_to_fp16)[name = tensor<string, []>("op_12637_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_12639_to_fp16 = const()[name = tensor<string, []>("op_12639_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(639167104)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_309_cast = mul(x = var_12637_cast, y = var_12639_to_fp16)[name = tensor<string, []>("hidden_states_309_cast")];
            tensor<int32, [2]> var_12646 = const()[name = tensor<string, []>("op_12646"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12648 = const()[name = tensor<string, []>("op_12648"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(639167808))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(639244672))), name = tensor<string, []>("up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            tensor<fp16, [2, 320, 1, 4096]> q_59_cast = conv(dilations = var_12648, groups = var_10375, pad = q_59_pad_0, pad_type = q_59_pad_type_0, strides = var_12646, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_309_cast)[name = tensor<string, []>("q_59_cast")];
            tensor<int32, [2]> var_12652 = const()[name = tensor<string, []>("op_12652"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12654 = const()[name = tensor<string, []>("op_12654"), 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, 768, 1, 1]> up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [184320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(639244864))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(639429248))), name = tensor<string, []>("up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 768, 1, 1])];
            tensor<fp16, [2, 320, 1, 77]> k_117_cast = conv(dilations = var_12654, groups = var_10375, pad = k_117_pad_0, pad_type = k_117_pad_type_0, strides = var_12652, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_117_cast")];
            tensor<int32, [2]> var_12658 = const()[name = tensor<string, []>("op_12658"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12660 = const()[name = tensor<string, []>("op_12660"), 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, 768, 1, 1]> up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [184320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(639429440))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(639613824))), name = tensor<string, []>("up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 768, 1, 1])];
            tensor<fp16, [2, 320, 1, 77]> v_59_cast = conv(dilations = var_12660, groups = var_10375, pad = v_59_pad_0, pad_type = v_59_pad_type_0, strides = var_12658, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_59_cast")];
            tensor<int32, [4]> var_12664_begin_0 = const()[name = tensor<string, []>("op_12664_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12664_end_0 = const()[name = tensor<string, []>("op_12664_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_12664_end_mask_0 = const()[name = tensor<string, []>("op_12664_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_12664_cast = slice_by_index(begin = var_12664_begin_0, end = var_12664_end_0, end_mask = var_12664_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_12664_cast")];
            tensor<int32, [4]> var_12668_begin_0 = const()[name = tensor<string, []>("op_12668_begin_0"), val = tensor<int32, [4]>([0, 40, 0, 0])];
            tensor<int32, [4]> var_12668_end_0 = const()[name = tensor<string, []>("op_12668_end_0"), val = tensor<int32, [4]>([2, 80, 1, 4096])];
            tensor<bool, [4]> var_12668_end_mask_0 = const()[name = tensor<string, []>("op_12668_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_12668_cast = slice_by_index(begin = var_12668_begin_0, end = var_12668_end_0, end_mask = var_12668_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_12668_cast")];
            tensor<int32, [4]> var_12672_begin_0 = const()[name = tensor<string, []>("op_12672_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_12672_end_0 = const()[name = tensor<string, []>("op_12672_end_0"), val = tensor<int32, [4]>([2, 120, 1, 4096])];
            tensor<bool, [4]> var_12672_end_mask_0 = const()[name = tensor<string, []>("op_12672_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_12672_cast = slice_by_index(begin = var_12672_begin_0, end = var_12672_end_0, end_mask = var_12672_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_12672_cast")];
            tensor<int32, [4]> var_12676_begin_0 = const()[name = tensor<string, []>("op_12676_begin_0"), val = tensor<int32, [4]>([0, 120, 0, 0])];
            tensor<int32, [4]> var_12676_end_0 = const()[name = tensor<string, []>("op_12676_end_0"), val = tensor<int32, [4]>([2, 160, 1, 4096])];
            tensor<bool, [4]> var_12676_end_mask_0 = const()[name = tensor<string, []>("op_12676_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_12676_cast = slice_by_index(begin = var_12676_begin_0, end = var_12676_end_0, end_mask = var_12676_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_12676_cast")];
            tensor<int32, [4]> var_12680_begin_0 = const()[name = tensor<string, []>("op_12680_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_12680_end_0 = const()[name = tensor<string, []>("op_12680_end_0"), val = tensor<int32, [4]>([2, 200, 1, 4096])];
            tensor<bool, [4]> var_12680_end_mask_0 = const()[name = tensor<string, []>("op_12680_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_12680_cast = slice_by_index(begin = var_12680_begin_0, end = var_12680_end_0, end_mask = var_12680_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_12680_cast")];
            tensor<int32, [4]> var_12684_begin_0 = const()[name = tensor<string, []>("op_12684_begin_0"), val = tensor<int32, [4]>([0, 200, 0, 0])];
            tensor<int32, [4]> var_12684_end_0 = const()[name = tensor<string, []>("op_12684_end_0"), val = tensor<int32, [4]>([2, 240, 1, 4096])];
            tensor<bool, [4]> var_12684_end_mask_0 = const()[name = tensor<string, []>("op_12684_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_12684_cast = slice_by_index(begin = var_12684_begin_0, end = var_12684_end_0, end_mask = var_12684_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_12684_cast")];
            tensor<int32, [4]> var_12688_begin_0 = const()[name = tensor<string, []>("op_12688_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_12688_end_0 = const()[name = tensor<string, []>("op_12688_end_0"), val = tensor<int32, [4]>([2, 280, 1, 4096])];
            tensor<bool, [4]> var_12688_end_mask_0 = const()[name = tensor<string, []>("op_12688_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_12688_cast = slice_by_index(begin = var_12688_begin_0, end = var_12688_end_0, end_mask = var_12688_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_12688_cast")];
            tensor<int32, [4]> var_12692_begin_0 = const()[name = tensor<string, []>("op_12692_begin_0"), val = tensor<int32, [4]>([0, 280, 0, 0])];
            tensor<int32, [4]> var_12692_end_0 = const()[name = tensor<string, []>("op_12692_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_12692_end_mask_0 = const()[name = tensor<string, []>("op_12692_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_12692_cast = slice_by_index(begin = var_12692_begin_0, end = var_12692_end_0, end_mask = var_12692_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_12692_cast")];
            tensor<int32, [4]> var_12695_begin_0 = const()[name = tensor<string, []>("op_12695_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12695_end_0 = const()[name = tensor<string, []>("op_12695_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_12695_end_mask_0 = const()[name = tensor<string, []>("op_12695_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12695_cast = slice_by_index(begin = var_12695_begin_0, end = var_12695_end_0, end_mask = var_12695_end_mask_0, x = var_12664_cast)[name = tensor<string, []>("op_12695_cast")];
            tensor<int32, [4]> var_12696_begin_0 = const()[name = tensor<string, []>("op_12696_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12696_end_0 = const()[name = tensor<string, []>("op_12696_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_12696_end_mask_0 = const()[name = tensor<string, []>("op_12696_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12696_cast = slice_by_index(begin = var_12696_begin_0, end = var_12696_end_0, end_mask = var_12696_end_mask_0, x = var_12664_cast)[name = tensor<string, []>("op_12696_cast")];
            tensor<int32, [4]> var_12697_begin_0 = const()[name = tensor<string, []>("op_12697_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_12697_end_0 = const()[name = tensor<string, []>("op_12697_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_12697_end_mask_0 = const()[name = tensor<string, []>("op_12697_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12697_cast = slice_by_index(begin = var_12697_begin_0, end = var_12697_end_0, end_mask = var_12697_end_mask_0, x = var_12664_cast)[name = tensor<string, []>("op_12697_cast")];
            tensor<int32, [4]> var_12698_begin_0 = const()[name = tensor<string, []>("op_12698_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_12698_end_0 = const()[name = tensor<string, []>("op_12698_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_12698_end_mask_0 = const()[name = tensor<string, []>("op_12698_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12698_cast = slice_by_index(begin = var_12698_begin_0, end = var_12698_end_0, end_mask = var_12698_end_mask_0, x = var_12664_cast)[name = tensor<string, []>("op_12698_cast")];
            tensor<int32, [4]> var_12699_begin_0 = const()[name = tensor<string, []>("op_12699_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_12699_end_0 = const()[name = tensor<string, []>("op_12699_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_12699_end_mask_0 = const()[name = tensor<string, []>("op_12699_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12699_cast = slice_by_index(begin = var_12699_begin_0, end = var_12699_end_0, end_mask = var_12699_end_mask_0, x = var_12664_cast)[name = tensor<string, []>("op_12699_cast")];
            tensor<int32, [4]> var_12700_begin_0 = const()[name = tensor<string, []>("op_12700_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_12700_end_0 = const()[name = tensor<string, []>("op_12700_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_12700_end_mask_0 = const()[name = tensor<string, []>("op_12700_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12700_cast = slice_by_index(begin = var_12700_begin_0, end = var_12700_end_0, end_mask = var_12700_end_mask_0, x = var_12664_cast)[name = tensor<string, []>("op_12700_cast")];
            tensor<int32, [4]> var_12701_begin_0 = const()[name = tensor<string, []>("op_12701_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_12701_end_0 = const()[name = tensor<string, []>("op_12701_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_12701_end_mask_0 = const()[name = tensor<string, []>("op_12701_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12701_cast = slice_by_index(begin = var_12701_begin_0, end = var_12701_end_0, end_mask = var_12701_end_mask_0, x = var_12664_cast)[name = tensor<string, []>("op_12701_cast")];
            tensor<int32, [4]> var_12702_begin_0 = const()[name = tensor<string, []>("op_12702_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_12702_end_0 = const()[name = tensor<string, []>("op_12702_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_12702_end_mask_0 = const()[name = tensor<string, []>("op_12702_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12702_cast = slice_by_index(begin = var_12702_begin_0, end = var_12702_end_0, end_mask = var_12702_end_mask_0, x = var_12664_cast)[name = tensor<string, []>("op_12702_cast")];
            tensor<int32, [4]> var_12703_begin_0 = const()[name = tensor<string, []>("op_12703_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12703_end_0 = const()[name = tensor<string, []>("op_12703_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_12703_end_mask_0 = const()[name = tensor<string, []>("op_12703_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12703_cast = slice_by_index(begin = var_12703_begin_0, end = var_12703_end_0, end_mask = var_12703_end_mask_0, x = var_12668_cast)[name = tensor<string, []>("op_12703_cast")];
            tensor<int32, [4]> var_12704_begin_0 = const()[name = tensor<string, []>("op_12704_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12704_end_0 = const()[name = tensor<string, []>("op_12704_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_12704_end_mask_0 = const()[name = tensor<string, []>("op_12704_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12704_cast = slice_by_index(begin = var_12704_begin_0, end = var_12704_end_0, end_mask = var_12704_end_mask_0, x = var_12668_cast)[name = tensor<string, []>("op_12704_cast")];
            tensor<int32, [4]> var_12705_begin_0 = const()[name = tensor<string, []>("op_12705_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_12705_end_0 = const()[name = tensor<string, []>("op_12705_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_12705_end_mask_0 = const()[name = tensor<string, []>("op_12705_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12705_cast = slice_by_index(begin = var_12705_begin_0, end = var_12705_end_0, end_mask = var_12705_end_mask_0, x = var_12668_cast)[name = tensor<string, []>("op_12705_cast")];
            tensor<int32, [4]> var_12706_begin_0 = const()[name = tensor<string, []>("op_12706_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_12706_end_0 = const()[name = tensor<string, []>("op_12706_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_12706_end_mask_0 = const()[name = tensor<string, []>("op_12706_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12706_cast = slice_by_index(begin = var_12706_begin_0, end = var_12706_end_0, end_mask = var_12706_end_mask_0, x = var_12668_cast)[name = tensor<string, []>("op_12706_cast")];
            tensor<int32, [4]> var_12707_begin_0 = const()[name = tensor<string, []>("op_12707_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_12707_end_0 = const()[name = tensor<string, []>("op_12707_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_12707_end_mask_0 = const()[name = tensor<string, []>("op_12707_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12707_cast = slice_by_index(begin = var_12707_begin_0, end = var_12707_end_0, end_mask = var_12707_end_mask_0, x = var_12668_cast)[name = tensor<string, []>("op_12707_cast")];
            tensor<int32, [4]> var_12708_begin_0 = const()[name = tensor<string, []>("op_12708_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_12708_end_0 = const()[name = tensor<string, []>("op_12708_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_12708_end_mask_0 = const()[name = tensor<string, []>("op_12708_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12708_cast = slice_by_index(begin = var_12708_begin_0, end = var_12708_end_0, end_mask = var_12708_end_mask_0, x = var_12668_cast)[name = tensor<string, []>("op_12708_cast")];
            tensor<int32, [4]> var_12709_begin_0 = const()[name = tensor<string, []>("op_12709_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_12709_end_0 = const()[name = tensor<string, []>("op_12709_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_12709_end_mask_0 = const()[name = tensor<string, []>("op_12709_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12709_cast = slice_by_index(begin = var_12709_begin_0, end = var_12709_end_0, end_mask = var_12709_end_mask_0, x = var_12668_cast)[name = tensor<string, []>("op_12709_cast")];
            tensor<int32, [4]> var_12710_begin_0 = const()[name = tensor<string, []>("op_12710_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_12710_end_0 = const()[name = tensor<string, []>("op_12710_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_12710_end_mask_0 = const()[name = tensor<string, []>("op_12710_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12710_cast = slice_by_index(begin = var_12710_begin_0, end = var_12710_end_0, end_mask = var_12710_end_mask_0, x = var_12668_cast)[name = tensor<string, []>("op_12710_cast")];
            tensor<int32, [4]> var_12711_begin_0 = const()[name = tensor<string, []>("op_12711_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12711_end_0 = const()[name = tensor<string, []>("op_12711_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_12711_end_mask_0 = const()[name = tensor<string, []>("op_12711_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12711_cast = slice_by_index(begin = var_12711_begin_0, end = var_12711_end_0, end_mask = var_12711_end_mask_0, x = var_12672_cast)[name = tensor<string, []>("op_12711_cast")];
            tensor<int32, [4]> var_12712_begin_0 = const()[name = tensor<string, []>("op_12712_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12712_end_0 = const()[name = tensor<string, []>("op_12712_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_12712_end_mask_0 = const()[name = tensor<string, []>("op_12712_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12712_cast = slice_by_index(begin = var_12712_begin_0, end = var_12712_end_0, end_mask = var_12712_end_mask_0, x = var_12672_cast)[name = tensor<string, []>("op_12712_cast")];
            tensor<int32, [4]> var_12713_begin_0 = const()[name = tensor<string, []>("op_12713_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_12713_end_0 = const()[name = tensor<string, []>("op_12713_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_12713_end_mask_0 = const()[name = tensor<string, []>("op_12713_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12713_cast = slice_by_index(begin = var_12713_begin_0, end = var_12713_end_0, end_mask = var_12713_end_mask_0, x = var_12672_cast)[name = tensor<string, []>("op_12713_cast")];
            tensor<int32, [4]> var_12714_begin_0 = const()[name = tensor<string, []>("op_12714_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_12714_end_0 = const()[name = tensor<string, []>("op_12714_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_12714_end_mask_0 = const()[name = tensor<string, []>("op_12714_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12714_cast = slice_by_index(begin = var_12714_begin_0, end = var_12714_end_0, end_mask = var_12714_end_mask_0, x = var_12672_cast)[name = tensor<string, []>("op_12714_cast")];
            tensor<int32, [4]> var_12715_begin_0 = const()[name = tensor<string, []>("op_12715_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_12715_end_0 = const()[name = tensor<string, []>("op_12715_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_12715_end_mask_0 = const()[name = tensor<string, []>("op_12715_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12715_cast = slice_by_index(begin = var_12715_begin_0, end = var_12715_end_0, end_mask = var_12715_end_mask_0, x = var_12672_cast)[name = tensor<string, []>("op_12715_cast")];
            tensor<int32, [4]> var_12716_begin_0 = const()[name = tensor<string, []>("op_12716_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_12716_end_0 = const()[name = tensor<string, []>("op_12716_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_12716_end_mask_0 = const()[name = tensor<string, []>("op_12716_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12716_cast = slice_by_index(begin = var_12716_begin_0, end = var_12716_end_0, end_mask = var_12716_end_mask_0, x = var_12672_cast)[name = tensor<string, []>("op_12716_cast")];
            tensor<int32, [4]> var_12717_begin_0 = const()[name = tensor<string, []>("op_12717_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_12717_end_0 = const()[name = tensor<string, []>("op_12717_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_12717_end_mask_0 = const()[name = tensor<string, []>("op_12717_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12717_cast = slice_by_index(begin = var_12717_begin_0, end = var_12717_end_0, end_mask = var_12717_end_mask_0, x = var_12672_cast)[name = tensor<string, []>("op_12717_cast")];
            tensor<int32, [4]> var_12718_begin_0 = const()[name = tensor<string, []>("op_12718_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_12718_end_0 = const()[name = tensor<string, []>("op_12718_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_12718_end_mask_0 = const()[name = tensor<string, []>("op_12718_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12718_cast = slice_by_index(begin = var_12718_begin_0, end = var_12718_end_0, end_mask = var_12718_end_mask_0, x = var_12672_cast)[name = tensor<string, []>("op_12718_cast")];
            tensor<int32, [4]> var_12719_begin_0 = const()[name = tensor<string, []>("op_12719_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12719_end_0 = const()[name = tensor<string, []>("op_12719_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_12719_end_mask_0 = const()[name = tensor<string, []>("op_12719_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12719_cast = slice_by_index(begin = var_12719_begin_0, end = var_12719_end_0, end_mask = var_12719_end_mask_0, x = var_12676_cast)[name = tensor<string, []>("op_12719_cast")];
            tensor<int32, [4]> var_12720_begin_0 = const()[name = tensor<string, []>("op_12720_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12720_end_0 = const()[name = tensor<string, []>("op_12720_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_12720_end_mask_0 = const()[name = tensor<string, []>("op_12720_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12720_cast = slice_by_index(begin = var_12720_begin_0, end = var_12720_end_0, end_mask = var_12720_end_mask_0, x = var_12676_cast)[name = tensor<string, []>("op_12720_cast")];
            tensor<int32, [4]> var_12721_begin_0 = const()[name = tensor<string, []>("op_12721_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_12721_end_0 = const()[name = tensor<string, []>("op_12721_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_12721_end_mask_0 = const()[name = tensor<string, []>("op_12721_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12721_cast = slice_by_index(begin = var_12721_begin_0, end = var_12721_end_0, end_mask = var_12721_end_mask_0, x = var_12676_cast)[name = tensor<string, []>("op_12721_cast")];
            tensor<int32, [4]> var_12722_begin_0 = const()[name = tensor<string, []>("op_12722_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_12722_end_0 = const()[name = tensor<string, []>("op_12722_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_12722_end_mask_0 = const()[name = tensor<string, []>("op_12722_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12722_cast = slice_by_index(begin = var_12722_begin_0, end = var_12722_end_0, end_mask = var_12722_end_mask_0, x = var_12676_cast)[name = tensor<string, []>("op_12722_cast")];
            tensor<int32, [4]> var_12723_begin_0 = const()[name = tensor<string, []>("op_12723_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_12723_end_0 = const()[name = tensor<string, []>("op_12723_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_12723_end_mask_0 = const()[name = tensor<string, []>("op_12723_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12723_cast = slice_by_index(begin = var_12723_begin_0, end = var_12723_end_0, end_mask = var_12723_end_mask_0, x = var_12676_cast)[name = tensor<string, []>("op_12723_cast")];
            tensor<int32, [4]> var_12724_begin_0 = const()[name = tensor<string, []>("op_12724_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_12724_end_0 = const()[name = tensor<string, []>("op_12724_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_12724_end_mask_0 = const()[name = tensor<string, []>("op_12724_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12724_cast = slice_by_index(begin = var_12724_begin_0, end = var_12724_end_0, end_mask = var_12724_end_mask_0, x = var_12676_cast)[name = tensor<string, []>("op_12724_cast")];
            tensor<int32, [4]> var_12725_begin_0 = const()[name = tensor<string, []>("op_12725_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_12725_end_0 = const()[name = tensor<string, []>("op_12725_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_12725_end_mask_0 = const()[name = tensor<string, []>("op_12725_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12725_cast = slice_by_index(begin = var_12725_begin_0, end = var_12725_end_0, end_mask = var_12725_end_mask_0, x = var_12676_cast)[name = tensor<string, []>("op_12725_cast")];
            tensor<int32, [4]> var_12726_begin_0 = const()[name = tensor<string, []>("op_12726_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_12726_end_0 = const()[name = tensor<string, []>("op_12726_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_12726_end_mask_0 = const()[name = tensor<string, []>("op_12726_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12726_cast = slice_by_index(begin = var_12726_begin_0, end = var_12726_end_0, end_mask = var_12726_end_mask_0, x = var_12676_cast)[name = tensor<string, []>("op_12726_cast")];
            tensor<int32, [4]> var_12727_begin_0 = const()[name = tensor<string, []>("op_12727_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12727_end_0 = const()[name = tensor<string, []>("op_12727_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_12727_end_mask_0 = const()[name = tensor<string, []>("op_12727_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12727_cast = slice_by_index(begin = var_12727_begin_0, end = var_12727_end_0, end_mask = var_12727_end_mask_0, x = var_12680_cast)[name = tensor<string, []>("op_12727_cast")];
            tensor<int32, [4]> var_12728_begin_0 = const()[name = tensor<string, []>("op_12728_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12728_end_0 = const()[name = tensor<string, []>("op_12728_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_12728_end_mask_0 = const()[name = tensor<string, []>("op_12728_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12728_cast = slice_by_index(begin = var_12728_begin_0, end = var_12728_end_0, end_mask = var_12728_end_mask_0, x = var_12680_cast)[name = tensor<string, []>("op_12728_cast")];
            tensor<int32, [4]> var_12729_begin_0 = const()[name = tensor<string, []>("op_12729_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_12729_end_0 = const()[name = tensor<string, []>("op_12729_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_12729_end_mask_0 = const()[name = tensor<string, []>("op_12729_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12729_cast = slice_by_index(begin = var_12729_begin_0, end = var_12729_end_0, end_mask = var_12729_end_mask_0, x = var_12680_cast)[name = tensor<string, []>("op_12729_cast")];
            tensor<int32, [4]> var_12730_begin_0 = const()[name = tensor<string, []>("op_12730_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_12730_end_0 = const()[name = tensor<string, []>("op_12730_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_12730_end_mask_0 = const()[name = tensor<string, []>("op_12730_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12730_cast = slice_by_index(begin = var_12730_begin_0, end = var_12730_end_0, end_mask = var_12730_end_mask_0, x = var_12680_cast)[name = tensor<string, []>("op_12730_cast")];
            tensor<int32, [4]> var_12731_begin_0 = const()[name = tensor<string, []>("op_12731_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_12731_end_0 = const()[name = tensor<string, []>("op_12731_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_12731_end_mask_0 = const()[name = tensor<string, []>("op_12731_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12731_cast = slice_by_index(begin = var_12731_begin_0, end = var_12731_end_0, end_mask = var_12731_end_mask_0, x = var_12680_cast)[name = tensor<string, []>("op_12731_cast")];
            tensor<int32, [4]> var_12732_begin_0 = const()[name = tensor<string, []>("op_12732_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_12732_end_0 = const()[name = tensor<string, []>("op_12732_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_12732_end_mask_0 = const()[name = tensor<string, []>("op_12732_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12732_cast = slice_by_index(begin = var_12732_begin_0, end = var_12732_end_0, end_mask = var_12732_end_mask_0, x = var_12680_cast)[name = tensor<string, []>("op_12732_cast")];
            tensor<int32, [4]> var_12733_begin_0 = const()[name = tensor<string, []>("op_12733_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_12733_end_0 = const()[name = tensor<string, []>("op_12733_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_12733_end_mask_0 = const()[name = tensor<string, []>("op_12733_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12733_cast = slice_by_index(begin = var_12733_begin_0, end = var_12733_end_0, end_mask = var_12733_end_mask_0, x = var_12680_cast)[name = tensor<string, []>("op_12733_cast")];
            tensor<int32, [4]> var_12734_begin_0 = const()[name = tensor<string, []>("op_12734_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_12734_end_0 = const()[name = tensor<string, []>("op_12734_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_12734_end_mask_0 = const()[name = tensor<string, []>("op_12734_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12734_cast = slice_by_index(begin = var_12734_begin_0, end = var_12734_end_0, end_mask = var_12734_end_mask_0, x = var_12680_cast)[name = tensor<string, []>("op_12734_cast")];
            tensor<int32, [4]> var_12735_begin_0 = const()[name = tensor<string, []>("op_12735_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12735_end_0 = const()[name = tensor<string, []>("op_12735_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_12735_end_mask_0 = const()[name = tensor<string, []>("op_12735_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12735_cast = slice_by_index(begin = var_12735_begin_0, end = var_12735_end_0, end_mask = var_12735_end_mask_0, x = var_12684_cast)[name = tensor<string, []>("op_12735_cast")];
            tensor<int32, [4]> var_12736_begin_0 = const()[name = tensor<string, []>("op_12736_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12736_end_0 = const()[name = tensor<string, []>("op_12736_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_12736_end_mask_0 = const()[name = tensor<string, []>("op_12736_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12736_cast = slice_by_index(begin = var_12736_begin_0, end = var_12736_end_0, end_mask = var_12736_end_mask_0, x = var_12684_cast)[name = tensor<string, []>("op_12736_cast")];
            tensor<int32, [4]> var_12737_begin_0 = const()[name = tensor<string, []>("op_12737_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_12737_end_0 = const()[name = tensor<string, []>("op_12737_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_12737_end_mask_0 = const()[name = tensor<string, []>("op_12737_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12737_cast = slice_by_index(begin = var_12737_begin_0, end = var_12737_end_0, end_mask = var_12737_end_mask_0, x = var_12684_cast)[name = tensor<string, []>("op_12737_cast")];
            tensor<int32, [4]> var_12738_begin_0 = const()[name = tensor<string, []>("op_12738_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_12738_end_0 = const()[name = tensor<string, []>("op_12738_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_12738_end_mask_0 = const()[name = tensor<string, []>("op_12738_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12738_cast = slice_by_index(begin = var_12738_begin_0, end = var_12738_end_0, end_mask = var_12738_end_mask_0, x = var_12684_cast)[name = tensor<string, []>("op_12738_cast")];
            tensor<int32, [4]> var_12739_begin_0 = const()[name = tensor<string, []>("op_12739_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_12739_end_0 = const()[name = tensor<string, []>("op_12739_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_12739_end_mask_0 = const()[name = tensor<string, []>("op_12739_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12739_cast = slice_by_index(begin = var_12739_begin_0, end = var_12739_end_0, end_mask = var_12739_end_mask_0, x = var_12684_cast)[name = tensor<string, []>("op_12739_cast")];
            tensor<int32, [4]> var_12740_begin_0 = const()[name = tensor<string, []>("op_12740_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_12740_end_0 = const()[name = tensor<string, []>("op_12740_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_12740_end_mask_0 = const()[name = tensor<string, []>("op_12740_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12740_cast = slice_by_index(begin = var_12740_begin_0, end = var_12740_end_0, end_mask = var_12740_end_mask_0, x = var_12684_cast)[name = tensor<string, []>("op_12740_cast")];
            tensor<int32, [4]> var_12741_begin_0 = const()[name = tensor<string, []>("op_12741_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_12741_end_0 = const()[name = tensor<string, []>("op_12741_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_12741_end_mask_0 = const()[name = tensor<string, []>("op_12741_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12741_cast = slice_by_index(begin = var_12741_begin_0, end = var_12741_end_0, end_mask = var_12741_end_mask_0, x = var_12684_cast)[name = tensor<string, []>("op_12741_cast")];
            tensor<int32, [4]> var_12742_begin_0 = const()[name = tensor<string, []>("op_12742_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_12742_end_0 = const()[name = tensor<string, []>("op_12742_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_12742_end_mask_0 = const()[name = tensor<string, []>("op_12742_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12742_cast = slice_by_index(begin = var_12742_begin_0, end = var_12742_end_0, end_mask = var_12742_end_mask_0, x = var_12684_cast)[name = tensor<string, []>("op_12742_cast")];
            tensor<int32, [4]> var_12743_begin_0 = const()[name = tensor<string, []>("op_12743_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12743_end_0 = const()[name = tensor<string, []>("op_12743_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_12743_end_mask_0 = const()[name = tensor<string, []>("op_12743_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12743_cast = slice_by_index(begin = var_12743_begin_0, end = var_12743_end_0, end_mask = var_12743_end_mask_0, x = var_12688_cast)[name = tensor<string, []>("op_12743_cast")];
            tensor<int32, [4]> var_12744_begin_0 = const()[name = tensor<string, []>("op_12744_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12744_end_0 = const()[name = tensor<string, []>("op_12744_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_12744_end_mask_0 = const()[name = tensor<string, []>("op_12744_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12744_cast = slice_by_index(begin = var_12744_begin_0, end = var_12744_end_0, end_mask = var_12744_end_mask_0, x = var_12688_cast)[name = tensor<string, []>("op_12744_cast")];
            tensor<int32, [4]> var_12745_begin_0 = const()[name = tensor<string, []>("op_12745_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_12745_end_0 = const()[name = tensor<string, []>("op_12745_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_12745_end_mask_0 = const()[name = tensor<string, []>("op_12745_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12745_cast = slice_by_index(begin = var_12745_begin_0, end = var_12745_end_0, end_mask = var_12745_end_mask_0, x = var_12688_cast)[name = tensor<string, []>("op_12745_cast")];
            tensor<int32, [4]> var_12746_begin_0 = const()[name = tensor<string, []>("op_12746_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_12746_end_0 = const()[name = tensor<string, []>("op_12746_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_12746_end_mask_0 = const()[name = tensor<string, []>("op_12746_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12746_cast = slice_by_index(begin = var_12746_begin_0, end = var_12746_end_0, end_mask = var_12746_end_mask_0, x = var_12688_cast)[name = tensor<string, []>("op_12746_cast")];
            tensor<int32, [4]> var_12747_begin_0 = const()[name = tensor<string, []>("op_12747_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_12747_end_0 = const()[name = tensor<string, []>("op_12747_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_12747_end_mask_0 = const()[name = tensor<string, []>("op_12747_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12747_cast = slice_by_index(begin = var_12747_begin_0, end = var_12747_end_0, end_mask = var_12747_end_mask_0, x = var_12688_cast)[name = tensor<string, []>("op_12747_cast")];
            tensor<int32, [4]> var_12748_begin_0 = const()[name = tensor<string, []>("op_12748_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_12748_end_0 = const()[name = tensor<string, []>("op_12748_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_12748_end_mask_0 = const()[name = tensor<string, []>("op_12748_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12748_cast = slice_by_index(begin = var_12748_begin_0, end = var_12748_end_0, end_mask = var_12748_end_mask_0, x = var_12688_cast)[name = tensor<string, []>("op_12748_cast")];
            tensor<int32, [4]> var_12749_begin_0 = const()[name = tensor<string, []>("op_12749_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_12749_end_0 = const()[name = tensor<string, []>("op_12749_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_12749_end_mask_0 = const()[name = tensor<string, []>("op_12749_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12749_cast = slice_by_index(begin = var_12749_begin_0, end = var_12749_end_0, end_mask = var_12749_end_mask_0, x = var_12688_cast)[name = tensor<string, []>("op_12749_cast")];
            tensor<int32, [4]> var_12750_begin_0 = const()[name = tensor<string, []>("op_12750_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_12750_end_0 = const()[name = tensor<string, []>("op_12750_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_12750_end_mask_0 = const()[name = tensor<string, []>("op_12750_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12750_cast = slice_by_index(begin = var_12750_begin_0, end = var_12750_end_0, end_mask = var_12750_end_mask_0, x = var_12688_cast)[name = tensor<string, []>("op_12750_cast")];
            tensor<int32, [4]> var_12751_begin_0 = const()[name = tensor<string, []>("op_12751_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12751_end_0 = const()[name = tensor<string, []>("op_12751_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_12751_end_mask_0 = const()[name = tensor<string, []>("op_12751_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12751_cast = slice_by_index(begin = var_12751_begin_0, end = var_12751_end_0, end_mask = var_12751_end_mask_0, x = var_12692_cast)[name = tensor<string, []>("op_12751_cast")];
            tensor<int32, [4]> var_12752_begin_0 = const()[name = tensor<string, []>("op_12752_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12752_end_0 = const()[name = tensor<string, []>("op_12752_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_12752_end_mask_0 = const()[name = tensor<string, []>("op_12752_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12752_cast = slice_by_index(begin = var_12752_begin_0, end = var_12752_end_0, end_mask = var_12752_end_mask_0, x = var_12692_cast)[name = tensor<string, []>("op_12752_cast")];
            tensor<int32, [4]> var_12753_begin_0 = const()[name = tensor<string, []>("op_12753_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_12753_end_0 = const()[name = tensor<string, []>("op_12753_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_12753_end_mask_0 = const()[name = tensor<string, []>("op_12753_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12753_cast = slice_by_index(begin = var_12753_begin_0, end = var_12753_end_0, end_mask = var_12753_end_mask_0, x = var_12692_cast)[name = tensor<string, []>("op_12753_cast")];
            tensor<int32, [4]> var_12754_begin_0 = const()[name = tensor<string, []>("op_12754_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_12754_end_0 = const()[name = tensor<string, []>("op_12754_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_12754_end_mask_0 = const()[name = tensor<string, []>("op_12754_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12754_cast = slice_by_index(begin = var_12754_begin_0, end = var_12754_end_0, end_mask = var_12754_end_mask_0, x = var_12692_cast)[name = tensor<string, []>("op_12754_cast")];
            tensor<int32, [4]> var_12755_begin_0 = const()[name = tensor<string, []>("op_12755_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_12755_end_0 = const()[name = tensor<string, []>("op_12755_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_12755_end_mask_0 = const()[name = tensor<string, []>("op_12755_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12755_cast = slice_by_index(begin = var_12755_begin_0, end = var_12755_end_0, end_mask = var_12755_end_mask_0, x = var_12692_cast)[name = tensor<string, []>("op_12755_cast")];
            tensor<int32, [4]> var_12756_begin_0 = const()[name = tensor<string, []>("op_12756_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_12756_end_0 = const()[name = tensor<string, []>("op_12756_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_12756_end_mask_0 = const()[name = tensor<string, []>("op_12756_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12756_cast = slice_by_index(begin = var_12756_begin_0, end = var_12756_end_0, end_mask = var_12756_end_mask_0, x = var_12692_cast)[name = tensor<string, []>("op_12756_cast")];
            tensor<int32, [4]> var_12757_begin_0 = const()[name = tensor<string, []>("op_12757_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_12757_end_0 = const()[name = tensor<string, []>("op_12757_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_12757_end_mask_0 = const()[name = tensor<string, []>("op_12757_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12757_cast = slice_by_index(begin = var_12757_begin_0, end = var_12757_end_0, end_mask = var_12757_end_mask_0, x = var_12692_cast)[name = tensor<string, []>("op_12757_cast")];
            tensor<int32, [4]> var_12758_begin_0 = const()[name = tensor<string, []>("op_12758_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_12758_end_0 = const()[name = tensor<string, []>("op_12758_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_12758_end_mask_0 = const()[name = tensor<string, []>("op_12758_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_12758_cast = slice_by_index(begin = var_12758_begin_0, end = var_12758_end_0, end_mask = var_12758_end_mask_0, x = var_12692_cast)[name = tensor<string, []>("op_12758_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_12763_begin_0 = const()[name = tensor<string, []>("op_12763_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12763_end_0 = const()[name = tensor<string, []>("op_12763_end_0"), val = tensor<int32, [4]>([2, 77, 1, 40])];
            tensor<bool, [4]> var_12763_end_mask_0 = const()[name = tensor<string, []>("op_12763_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)[name = tensor<string, []>("transpose_2")];
            tensor<fp16, [2, 77, 1, 40]> var_12763_cast = slice_by_index(begin = var_12763_begin_0, end = var_12763_end_0, end_mask = var_12763_end_mask_0, x = transpose_2)[name = tensor<string, []>("op_12763_cast")];
            tensor<int32, [4]> var_12767_begin_0 = const()[name = tensor<string, []>("op_12767_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 40])];
            tensor<int32, [4]> var_12767_end_0 = const()[name = tensor<string, []>("op_12767_end_0"), val = tensor<int32, [4]>([2, 77, 1, 80])];
            tensor<bool, [4]> var_12767_end_mask_0 = const()[name = tensor<string, []>("op_12767_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 40]> var_12767_cast = slice_by_index(begin = var_12767_begin_0, end = var_12767_end_0, end_mask = var_12767_end_mask_0, x = transpose_2)[name = tensor<string, []>("op_12767_cast")];
            tensor<int32, [4]> var_12771_begin_0 = const()[name = tensor<string, []>("op_12771_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 80])];
            tensor<int32, [4]> var_12771_end_0 = const()[name = tensor<string, []>("op_12771_end_0"), val = tensor<int32, [4]>([2, 77, 1, 120])];
            tensor<bool, [4]> var_12771_end_mask_0 = const()[name = tensor<string, []>("op_12771_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 40]> var_12771_cast = slice_by_index(begin = var_12771_begin_0, end = var_12771_end_0, end_mask = var_12771_end_mask_0, x = transpose_2)[name = tensor<string, []>("op_12771_cast")];
            tensor<int32, [4]> var_12775_begin_0 = const()[name = tensor<string, []>("op_12775_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 120])];
            tensor<int32, [4]> var_12775_end_0 = const()[name = tensor<string, []>("op_12775_end_0"), val = tensor<int32, [4]>([2, 77, 1, 160])];
            tensor<bool, [4]> var_12775_end_mask_0 = const()[name = tensor<string, []>("op_12775_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 40]> var_12775_cast = slice_by_index(begin = var_12775_begin_0, end = var_12775_end_0, end_mask = var_12775_end_mask_0, x = transpose_2)[name = tensor<string, []>("op_12775_cast")];
            tensor<int32, [4]> var_12779_begin_0 = const()[name = tensor<string, []>("op_12779_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_12779_end_0 = const()[name = tensor<string, []>("op_12779_end_0"), val = tensor<int32, [4]>([2, 77, 1, 200])];
            tensor<bool, [4]> var_12779_end_mask_0 = const()[name = tensor<string, []>("op_12779_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 40]> var_12779_cast = slice_by_index(begin = var_12779_begin_0, end = var_12779_end_0, end_mask = var_12779_end_mask_0, x = transpose_2)[name = tensor<string, []>("op_12779_cast")];
            tensor<int32, [4]> var_12783_begin_0 = const()[name = tensor<string, []>("op_12783_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 200])];
            tensor<int32, [4]> var_12783_end_0 = const()[name = tensor<string, []>("op_12783_end_0"), val = tensor<int32, [4]>([2, 77, 1, 240])];
            tensor<bool, [4]> var_12783_end_mask_0 = const()[name = tensor<string, []>("op_12783_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 40]> var_12783_cast = slice_by_index(begin = var_12783_begin_0, end = var_12783_end_0, end_mask = var_12783_end_mask_0, x = transpose_2)[name = tensor<string, []>("op_12783_cast")];
            tensor<int32, [4]> var_12787_begin_0 = const()[name = tensor<string, []>("op_12787_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 240])];
            tensor<int32, [4]> var_12787_end_0 = const()[name = tensor<string, []>("op_12787_end_0"), val = tensor<int32, [4]>([2, 77, 1, 280])];
            tensor<bool, [4]> var_12787_end_mask_0 = const()[name = tensor<string, []>("op_12787_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 40]> var_12787_cast = slice_by_index(begin = var_12787_begin_0, end = var_12787_end_0, end_mask = var_12787_end_mask_0, x = transpose_2)[name = tensor<string, []>("op_12787_cast")];
            tensor<int32, [4]> var_12791_begin_0 = const()[name = tensor<string, []>("op_12791_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 280])];
            tensor<int32, [4]> var_12791_end_0 = const()[name = tensor<string, []>("op_12791_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_12791_end_mask_0 = const()[name = tensor<string, []>("op_12791_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 40]> var_12791_cast = slice_by_index(begin = var_12791_begin_0, end = var_12791_end_0, end_mask = var_12791_end_mask_0, x = transpose_2)[name = tensor<string, []>("op_12791_cast")];
            tensor<int32, [4]> var_12793_begin_0 = const()[name = tensor<string, []>("op_12793_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12793_end_0 = const()[name = tensor<string, []>("op_12793_end_0"), val = tensor<int32, [4]>([2, 40, 1, 77])];
            tensor<bool, [4]> var_12793_end_mask_0 = const()[name = tensor<string, []>("op_12793_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_12793_cast = slice_by_index(begin = var_12793_begin_0, end = var_12793_end_0, end_mask = var_12793_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_12793_cast")];
            tensor<int32, [4]> var_12797_begin_0 = const()[name = tensor<string, []>("op_12797_begin_0"), val = tensor<int32, [4]>([0, 40, 0, 0])];
            tensor<int32, [4]> var_12797_end_0 = const()[name = tensor<string, []>("op_12797_end_0"), val = tensor<int32, [4]>([2, 80, 1, 77])];
            tensor<bool, [4]> var_12797_end_mask_0 = const()[name = tensor<string, []>("op_12797_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_12797_cast = slice_by_index(begin = var_12797_begin_0, end = var_12797_end_0, end_mask = var_12797_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_12797_cast")];
            tensor<int32, [4]> var_12801_begin_0 = const()[name = tensor<string, []>("op_12801_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_12801_end_0 = const()[name = tensor<string, []>("op_12801_end_0"), val = tensor<int32, [4]>([2, 120, 1, 77])];
            tensor<bool, [4]> var_12801_end_mask_0 = const()[name = tensor<string, []>("op_12801_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_12801_cast = slice_by_index(begin = var_12801_begin_0, end = var_12801_end_0, end_mask = var_12801_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_12801_cast")];
            tensor<int32, [4]> var_12805_begin_0 = const()[name = tensor<string, []>("op_12805_begin_0"), val = tensor<int32, [4]>([0, 120, 0, 0])];
            tensor<int32, [4]> var_12805_end_0 = const()[name = tensor<string, []>("op_12805_end_0"), val = tensor<int32, [4]>([2, 160, 1, 77])];
            tensor<bool, [4]> var_12805_end_mask_0 = const()[name = tensor<string, []>("op_12805_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_12805_cast = slice_by_index(begin = var_12805_begin_0, end = var_12805_end_0, end_mask = var_12805_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_12805_cast")];
            tensor<int32, [4]> var_12809_begin_0 = const()[name = tensor<string, []>("op_12809_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_12809_end_0 = const()[name = tensor<string, []>("op_12809_end_0"), val = tensor<int32, [4]>([2, 200, 1, 77])];
            tensor<bool, [4]> var_12809_end_mask_0 = const()[name = tensor<string, []>("op_12809_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_12809_cast = slice_by_index(begin = var_12809_begin_0, end = var_12809_end_0, end_mask = var_12809_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_12809_cast")];
            tensor<int32, [4]> var_12813_begin_0 = const()[name = tensor<string, []>("op_12813_begin_0"), val = tensor<int32, [4]>([0, 200, 0, 0])];
            tensor<int32, [4]> var_12813_end_0 = const()[name = tensor<string, []>("op_12813_end_0"), val = tensor<int32, [4]>([2, 240, 1, 77])];
            tensor<bool, [4]> var_12813_end_mask_0 = const()[name = tensor<string, []>("op_12813_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_12813_cast = slice_by_index(begin = var_12813_begin_0, end = var_12813_end_0, end_mask = var_12813_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_12813_cast")];
            tensor<int32, [4]> var_12817_begin_0 = const()[name = tensor<string, []>("op_12817_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_12817_end_0 = const()[name = tensor<string, []>("op_12817_end_0"), val = tensor<int32, [4]>([2, 280, 1, 77])];
            tensor<bool, [4]> var_12817_end_mask_0 = const()[name = tensor<string, []>("op_12817_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_12817_cast = slice_by_index(begin = var_12817_begin_0, end = var_12817_end_0, end_mask = var_12817_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_12817_cast")];
            tensor<int32, [4]> var_12821_begin_0 = const()[name = tensor<string, []>("op_12821_begin_0"), val = tensor<int32, [4]>([0, 280, 0, 0])];
            tensor<int32, [4]> var_12821_end_0 = const()[name = tensor<string, []>("op_12821_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_12821_end_mask_0 = const()[name = tensor<string, []>("op_12821_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_12821_cast = slice_by_index(begin = var_12821_begin_0, end = var_12821_end_0, end_mask = var_12821_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_12821_cast")];
            tensor<string, []> var_12825_equation_0 = const()[name = tensor<string, []>("op_12825_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12825_cast = einsum(equation = var_12825_equation_0, values = (var_12763_cast, var_12695_cast))[name = tensor<string, []>("op_12825_cast")];
            tensor<fp16, []> var_12826_to_fp16 = const()[name = tensor<string, []>("op_12826_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1217_cast = mul(x = var_12825_cast, y = var_12826_to_fp16)[name = tensor<string, []>("aw_chunk_1217_cast")];
            tensor<string, []> var_12829_equation_0 = const()[name = tensor<string, []>("op_12829_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12829_cast = einsum(equation = var_12829_equation_0, values = (var_12763_cast, var_12696_cast))[name = tensor<string, []>("op_12829_cast")];
            tensor<fp16, []> var_12830_to_fp16 = const()[name = tensor<string, []>("op_12830_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1219_cast = mul(x = var_12829_cast, y = var_12830_to_fp16)[name = tensor<string, []>("aw_chunk_1219_cast")];
            tensor<string, []> var_12833_equation_0 = const()[name = tensor<string, []>("op_12833_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12833_cast = einsum(equation = var_12833_equation_0, values = (var_12763_cast, var_12697_cast))[name = tensor<string, []>("op_12833_cast")];
            tensor<fp16, []> var_12834_to_fp16 = const()[name = tensor<string, []>("op_12834_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1221_cast = mul(x = var_12833_cast, y = var_12834_to_fp16)[name = tensor<string, []>("aw_chunk_1221_cast")];
            tensor<string, []> var_12837_equation_0 = const()[name = tensor<string, []>("op_12837_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12837_cast = einsum(equation = var_12837_equation_0, values = (var_12763_cast, var_12698_cast))[name = tensor<string, []>("op_12837_cast")];
            tensor<fp16, []> var_12838_to_fp16 = const()[name = tensor<string, []>("op_12838_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1223_cast = mul(x = var_12837_cast, y = var_12838_to_fp16)[name = tensor<string, []>("aw_chunk_1223_cast")];
            tensor<string, []> var_12841_equation_0 = const()[name = tensor<string, []>("op_12841_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12841_cast = einsum(equation = var_12841_equation_0, values = (var_12763_cast, var_12699_cast))[name = tensor<string, []>("op_12841_cast")];
            tensor<fp16, []> var_12842_to_fp16 = const()[name = tensor<string, []>("op_12842_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1225_cast = mul(x = var_12841_cast, y = var_12842_to_fp16)[name = tensor<string, []>("aw_chunk_1225_cast")];
            tensor<string, []> var_12845_equation_0 = const()[name = tensor<string, []>("op_12845_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12845_cast = einsum(equation = var_12845_equation_0, values = (var_12763_cast, var_12700_cast))[name = tensor<string, []>("op_12845_cast")];
            tensor<fp16, []> var_12846_to_fp16 = const()[name = tensor<string, []>("op_12846_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1227_cast = mul(x = var_12845_cast, y = var_12846_to_fp16)[name = tensor<string, []>("aw_chunk_1227_cast")];
            tensor<string, []> var_12849_equation_0 = const()[name = tensor<string, []>("op_12849_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12849_cast = einsum(equation = var_12849_equation_0, values = (var_12763_cast, var_12701_cast))[name = tensor<string, []>("op_12849_cast")];
            tensor<fp16, []> var_12850_to_fp16 = const()[name = tensor<string, []>("op_12850_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1229_cast = mul(x = var_12849_cast, y = var_12850_to_fp16)[name = tensor<string, []>("aw_chunk_1229_cast")];
            tensor<string, []> var_12853_equation_0 = const()[name = tensor<string, []>("op_12853_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12853_cast = einsum(equation = var_12853_equation_0, values = (var_12763_cast, var_12702_cast))[name = tensor<string, []>("op_12853_cast")];
            tensor<fp16, []> var_12854_to_fp16 = const()[name = tensor<string, []>("op_12854_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1231_cast = mul(x = var_12853_cast, y = var_12854_to_fp16)[name = tensor<string, []>("aw_chunk_1231_cast")];
            tensor<string, []> var_12857_equation_0 = const()[name = tensor<string, []>("op_12857_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12857_cast = einsum(equation = var_12857_equation_0, values = (var_12767_cast, var_12703_cast))[name = tensor<string, []>("op_12857_cast")];
            tensor<fp16, []> var_12858_to_fp16 = const()[name = tensor<string, []>("op_12858_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1233_cast = mul(x = var_12857_cast, y = var_12858_to_fp16)[name = tensor<string, []>("aw_chunk_1233_cast")];
            tensor<string, []> var_12861_equation_0 = const()[name = tensor<string, []>("op_12861_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12861_cast = einsum(equation = var_12861_equation_0, values = (var_12767_cast, var_12704_cast))[name = tensor<string, []>("op_12861_cast")];
            tensor<fp16, []> var_12862_to_fp16 = const()[name = tensor<string, []>("op_12862_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1235_cast = mul(x = var_12861_cast, y = var_12862_to_fp16)[name = tensor<string, []>("aw_chunk_1235_cast")];
            tensor<string, []> var_12865_equation_0 = const()[name = tensor<string, []>("op_12865_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12865_cast = einsum(equation = var_12865_equation_0, values = (var_12767_cast, var_12705_cast))[name = tensor<string, []>("op_12865_cast")];
            tensor<fp16, []> var_12866_to_fp16 = const()[name = tensor<string, []>("op_12866_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1237_cast = mul(x = var_12865_cast, y = var_12866_to_fp16)[name = tensor<string, []>("aw_chunk_1237_cast")];
            tensor<string, []> var_12869_equation_0 = const()[name = tensor<string, []>("op_12869_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12869_cast = einsum(equation = var_12869_equation_0, values = (var_12767_cast, var_12706_cast))[name = tensor<string, []>("op_12869_cast")];
            tensor<fp16, []> var_12870_to_fp16 = const()[name = tensor<string, []>("op_12870_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1239_cast = mul(x = var_12869_cast, y = var_12870_to_fp16)[name = tensor<string, []>("aw_chunk_1239_cast")];
            tensor<string, []> var_12873_equation_0 = const()[name = tensor<string, []>("op_12873_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12873_cast = einsum(equation = var_12873_equation_0, values = (var_12767_cast, var_12707_cast))[name = tensor<string, []>("op_12873_cast")];
            tensor<fp16, []> var_12874_to_fp16 = const()[name = tensor<string, []>("op_12874_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1241_cast = mul(x = var_12873_cast, y = var_12874_to_fp16)[name = tensor<string, []>("aw_chunk_1241_cast")];
            tensor<string, []> var_12877_equation_0 = const()[name = tensor<string, []>("op_12877_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12877_cast = einsum(equation = var_12877_equation_0, values = (var_12767_cast, var_12708_cast))[name = tensor<string, []>("op_12877_cast")];
            tensor<fp16, []> var_12878_to_fp16 = const()[name = tensor<string, []>("op_12878_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1243_cast = mul(x = var_12877_cast, y = var_12878_to_fp16)[name = tensor<string, []>("aw_chunk_1243_cast")];
            tensor<string, []> var_12881_equation_0 = const()[name = tensor<string, []>("op_12881_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12881_cast = einsum(equation = var_12881_equation_0, values = (var_12767_cast, var_12709_cast))[name = tensor<string, []>("op_12881_cast")];
            tensor<fp16, []> var_12882_to_fp16 = const()[name = tensor<string, []>("op_12882_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1245_cast = mul(x = var_12881_cast, y = var_12882_to_fp16)[name = tensor<string, []>("aw_chunk_1245_cast")];
            tensor<string, []> var_12885_equation_0 = const()[name = tensor<string, []>("op_12885_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12885_cast = einsum(equation = var_12885_equation_0, values = (var_12767_cast, var_12710_cast))[name = tensor<string, []>("op_12885_cast")];
            tensor<fp16, []> var_12886_to_fp16 = const()[name = tensor<string, []>("op_12886_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1247_cast = mul(x = var_12885_cast, y = var_12886_to_fp16)[name = tensor<string, []>("aw_chunk_1247_cast")];
            tensor<string, []> var_12889_equation_0 = const()[name = tensor<string, []>("op_12889_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12889_cast = einsum(equation = var_12889_equation_0, values = (var_12771_cast, var_12711_cast))[name = tensor<string, []>("op_12889_cast")];
            tensor<fp16, []> var_12890_to_fp16 = const()[name = tensor<string, []>("op_12890_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1249_cast = mul(x = var_12889_cast, y = var_12890_to_fp16)[name = tensor<string, []>("aw_chunk_1249_cast")];
            tensor<string, []> var_12893_equation_0 = const()[name = tensor<string, []>("op_12893_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12893_cast = einsum(equation = var_12893_equation_0, values = (var_12771_cast, var_12712_cast))[name = tensor<string, []>("op_12893_cast")];
            tensor<fp16, []> var_12894_to_fp16 = const()[name = tensor<string, []>("op_12894_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1251_cast = mul(x = var_12893_cast, y = var_12894_to_fp16)[name = tensor<string, []>("aw_chunk_1251_cast")];
            tensor<string, []> var_12897_equation_0 = const()[name = tensor<string, []>("op_12897_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12897_cast = einsum(equation = var_12897_equation_0, values = (var_12771_cast, var_12713_cast))[name = tensor<string, []>("op_12897_cast")];
            tensor<fp16, []> var_12898_to_fp16 = const()[name = tensor<string, []>("op_12898_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1253_cast = mul(x = var_12897_cast, y = var_12898_to_fp16)[name = tensor<string, []>("aw_chunk_1253_cast")];
            tensor<string, []> var_12901_equation_0 = const()[name = tensor<string, []>("op_12901_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12901_cast = einsum(equation = var_12901_equation_0, values = (var_12771_cast, var_12714_cast))[name = tensor<string, []>("op_12901_cast")];
            tensor<fp16, []> var_12902_to_fp16 = const()[name = tensor<string, []>("op_12902_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1255_cast = mul(x = var_12901_cast, y = var_12902_to_fp16)[name = tensor<string, []>("aw_chunk_1255_cast")];
            tensor<string, []> var_12905_equation_0 = const()[name = tensor<string, []>("op_12905_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12905_cast = einsum(equation = var_12905_equation_0, values = (var_12771_cast, var_12715_cast))[name = tensor<string, []>("op_12905_cast")];
            tensor<fp16, []> var_12906_to_fp16 = const()[name = tensor<string, []>("op_12906_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1257_cast = mul(x = var_12905_cast, y = var_12906_to_fp16)[name = tensor<string, []>("aw_chunk_1257_cast")];
            tensor<string, []> var_12909_equation_0 = const()[name = tensor<string, []>("op_12909_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12909_cast = einsum(equation = var_12909_equation_0, values = (var_12771_cast, var_12716_cast))[name = tensor<string, []>("op_12909_cast")];
            tensor<fp16, []> var_12910_to_fp16 = const()[name = tensor<string, []>("op_12910_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1259_cast = mul(x = var_12909_cast, y = var_12910_to_fp16)[name = tensor<string, []>("aw_chunk_1259_cast")];
            tensor<string, []> var_12913_equation_0 = const()[name = tensor<string, []>("op_12913_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12913_cast = einsum(equation = var_12913_equation_0, values = (var_12771_cast, var_12717_cast))[name = tensor<string, []>("op_12913_cast")];
            tensor<fp16, []> var_12914_to_fp16 = const()[name = tensor<string, []>("op_12914_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1261_cast = mul(x = var_12913_cast, y = var_12914_to_fp16)[name = tensor<string, []>("aw_chunk_1261_cast")];
            tensor<string, []> var_12917_equation_0 = const()[name = tensor<string, []>("op_12917_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12917_cast = einsum(equation = var_12917_equation_0, values = (var_12771_cast, var_12718_cast))[name = tensor<string, []>("op_12917_cast")];
            tensor<fp16, []> var_12918_to_fp16 = const()[name = tensor<string, []>("op_12918_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1263_cast = mul(x = var_12917_cast, y = var_12918_to_fp16)[name = tensor<string, []>("aw_chunk_1263_cast")];
            tensor<string, []> var_12921_equation_0 = const()[name = tensor<string, []>("op_12921_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12921_cast = einsum(equation = var_12921_equation_0, values = (var_12775_cast, var_12719_cast))[name = tensor<string, []>("op_12921_cast")];
            tensor<fp16, []> var_12922_to_fp16 = const()[name = tensor<string, []>("op_12922_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1265_cast = mul(x = var_12921_cast, y = var_12922_to_fp16)[name = tensor<string, []>("aw_chunk_1265_cast")];
            tensor<string, []> var_12925_equation_0 = const()[name = tensor<string, []>("op_12925_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12925_cast = einsum(equation = var_12925_equation_0, values = (var_12775_cast, var_12720_cast))[name = tensor<string, []>("op_12925_cast")];
            tensor<fp16, []> var_12926_to_fp16 = const()[name = tensor<string, []>("op_12926_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1267_cast = mul(x = var_12925_cast, y = var_12926_to_fp16)[name = tensor<string, []>("aw_chunk_1267_cast")];
            tensor<string, []> var_12929_equation_0 = const()[name = tensor<string, []>("op_12929_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12929_cast = einsum(equation = var_12929_equation_0, values = (var_12775_cast, var_12721_cast))[name = tensor<string, []>("op_12929_cast")];
            tensor<fp16, []> var_12930_to_fp16 = const()[name = tensor<string, []>("op_12930_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1269_cast = mul(x = var_12929_cast, y = var_12930_to_fp16)[name = tensor<string, []>("aw_chunk_1269_cast")];
            tensor<string, []> var_12933_equation_0 = const()[name = tensor<string, []>("op_12933_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12933_cast = einsum(equation = var_12933_equation_0, values = (var_12775_cast, var_12722_cast))[name = tensor<string, []>("op_12933_cast")];
            tensor<fp16, []> var_12934_to_fp16 = const()[name = tensor<string, []>("op_12934_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1271_cast = mul(x = var_12933_cast, y = var_12934_to_fp16)[name = tensor<string, []>("aw_chunk_1271_cast")];
            tensor<string, []> var_12937_equation_0 = const()[name = tensor<string, []>("op_12937_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12937_cast = einsum(equation = var_12937_equation_0, values = (var_12775_cast, var_12723_cast))[name = tensor<string, []>("op_12937_cast")];
            tensor<fp16, []> var_12938_to_fp16 = const()[name = tensor<string, []>("op_12938_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1273_cast = mul(x = var_12937_cast, y = var_12938_to_fp16)[name = tensor<string, []>("aw_chunk_1273_cast")];
            tensor<string, []> var_12941_equation_0 = const()[name = tensor<string, []>("op_12941_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12941_cast = einsum(equation = var_12941_equation_0, values = (var_12775_cast, var_12724_cast))[name = tensor<string, []>("op_12941_cast")];
            tensor<fp16, []> var_12942_to_fp16 = const()[name = tensor<string, []>("op_12942_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1275_cast = mul(x = var_12941_cast, y = var_12942_to_fp16)[name = tensor<string, []>("aw_chunk_1275_cast")];
            tensor<string, []> var_12945_equation_0 = const()[name = tensor<string, []>("op_12945_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12945_cast = einsum(equation = var_12945_equation_0, values = (var_12775_cast, var_12725_cast))[name = tensor<string, []>("op_12945_cast")];
            tensor<fp16, []> var_12946_to_fp16 = const()[name = tensor<string, []>("op_12946_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1277_cast = mul(x = var_12945_cast, y = var_12946_to_fp16)[name = tensor<string, []>("aw_chunk_1277_cast")];
            tensor<string, []> var_12949_equation_0 = const()[name = tensor<string, []>("op_12949_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12949_cast = einsum(equation = var_12949_equation_0, values = (var_12775_cast, var_12726_cast))[name = tensor<string, []>("op_12949_cast")];
            tensor<fp16, []> var_12950_to_fp16 = const()[name = tensor<string, []>("op_12950_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1279_cast = mul(x = var_12949_cast, y = var_12950_to_fp16)[name = tensor<string, []>("aw_chunk_1279_cast")];
            tensor<string, []> var_12953_equation_0 = const()[name = tensor<string, []>("op_12953_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12953_cast = einsum(equation = var_12953_equation_0, values = (var_12779_cast, var_12727_cast))[name = tensor<string, []>("op_12953_cast")];
            tensor<fp16, []> var_12954_to_fp16 = const()[name = tensor<string, []>("op_12954_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1281_cast = mul(x = var_12953_cast, y = var_12954_to_fp16)[name = tensor<string, []>("aw_chunk_1281_cast")];
            tensor<string, []> var_12957_equation_0 = const()[name = tensor<string, []>("op_12957_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12957_cast = einsum(equation = var_12957_equation_0, values = (var_12779_cast, var_12728_cast))[name = tensor<string, []>("op_12957_cast")];
            tensor<fp16, []> var_12958_to_fp16 = const()[name = tensor<string, []>("op_12958_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1283_cast = mul(x = var_12957_cast, y = var_12958_to_fp16)[name = tensor<string, []>("aw_chunk_1283_cast")];
            tensor<string, []> var_12961_equation_0 = const()[name = tensor<string, []>("op_12961_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12961_cast = einsum(equation = var_12961_equation_0, values = (var_12779_cast, var_12729_cast))[name = tensor<string, []>("op_12961_cast")];
            tensor<fp16, []> var_12962_to_fp16 = const()[name = tensor<string, []>("op_12962_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1285_cast = mul(x = var_12961_cast, y = var_12962_to_fp16)[name = tensor<string, []>("aw_chunk_1285_cast")];
            tensor<string, []> var_12965_equation_0 = const()[name = tensor<string, []>("op_12965_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12965_cast = einsum(equation = var_12965_equation_0, values = (var_12779_cast, var_12730_cast))[name = tensor<string, []>("op_12965_cast")];
            tensor<fp16, []> var_12966_to_fp16 = const()[name = tensor<string, []>("op_12966_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1287_cast = mul(x = var_12965_cast, y = var_12966_to_fp16)[name = tensor<string, []>("aw_chunk_1287_cast")];
            tensor<string, []> var_12969_equation_0 = const()[name = tensor<string, []>("op_12969_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12969_cast = einsum(equation = var_12969_equation_0, values = (var_12779_cast, var_12731_cast))[name = tensor<string, []>("op_12969_cast")];
            tensor<fp16, []> var_12970_to_fp16 = const()[name = tensor<string, []>("op_12970_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1289_cast = mul(x = var_12969_cast, y = var_12970_to_fp16)[name = tensor<string, []>("aw_chunk_1289_cast")];
            tensor<string, []> var_12973_equation_0 = const()[name = tensor<string, []>("op_12973_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12973_cast = einsum(equation = var_12973_equation_0, values = (var_12779_cast, var_12732_cast))[name = tensor<string, []>("op_12973_cast")];
            tensor<fp16, []> var_12974_to_fp16 = const()[name = tensor<string, []>("op_12974_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1291_cast = mul(x = var_12973_cast, y = var_12974_to_fp16)[name = tensor<string, []>("aw_chunk_1291_cast")];
            tensor<string, []> var_12977_equation_0 = const()[name = tensor<string, []>("op_12977_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12977_cast = einsum(equation = var_12977_equation_0, values = (var_12779_cast, var_12733_cast))[name = tensor<string, []>("op_12977_cast")];
            tensor<fp16, []> var_12978_to_fp16 = const()[name = tensor<string, []>("op_12978_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1293_cast = mul(x = var_12977_cast, y = var_12978_to_fp16)[name = tensor<string, []>("aw_chunk_1293_cast")];
            tensor<string, []> var_12981_equation_0 = const()[name = tensor<string, []>("op_12981_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12981_cast = einsum(equation = var_12981_equation_0, values = (var_12779_cast, var_12734_cast))[name = tensor<string, []>("op_12981_cast")];
            tensor<fp16, []> var_12982_to_fp16 = const()[name = tensor<string, []>("op_12982_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1295_cast = mul(x = var_12981_cast, y = var_12982_to_fp16)[name = tensor<string, []>("aw_chunk_1295_cast")];
            tensor<string, []> var_12985_equation_0 = const()[name = tensor<string, []>("op_12985_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12985_cast = einsum(equation = var_12985_equation_0, values = (var_12783_cast, var_12735_cast))[name = tensor<string, []>("op_12985_cast")];
            tensor<fp16, []> var_12986_to_fp16 = const()[name = tensor<string, []>("op_12986_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1297_cast = mul(x = var_12985_cast, y = var_12986_to_fp16)[name = tensor<string, []>("aw_chunk_1297_cast")];
            tensor<string, []> var_12989_equation_0 = const()[name = tensor<string, []>("op_12989_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12989_cast = einsum(equation = var_12989_equation_0, values = (var_12783_cast, var_12736_cast))[name = tensor<string, []>("op_12989_cast")];
            tensor<fp16, []> var_12990_to_fp16 = const()[name = tensor<string, []>("op_12990_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1299_cast = mul(x = var_12989_cast, y = var_12990_to_fp16)[name = tensor<string, []>("aw_chunk_1299_cast")];
            tensor<string, []> var_12993_equation_0 = const()[name = tensor<string, []>("op_12993_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12993_cast = einsum(equation = var_12993_equation_0, values = (var_12783_cast, var_12737_cast))[name = tensor<string, []>("op_12993_cast")];
            tensor<fp16, []> var_12994_to_fp16 = const()[name = tensor<string, []>("op_12994_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1301_cast = mul(x = var_12993_cast, y = var_12994_to_fp16)[name = tensor<string, []>("aw_chunk_1301_cast")];
            tensor<string, []> var_12997_equation_0 = const()[name = tensor<string, []>("op_12997_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12997_cast = einsum(equation = var_12997_equation_0, values = (var_12783_cast, var_12738_cast))[name = tensor<string, []>("op_12997_cast")];
            tensor<fp16, []> var_12998_to_fp16 = const()[name = tensor<string, []>("op_12998_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1303_cast = mul(x = var_12997_cast, y = var_12998_to_fp16)[name = tensor<string, []>("aw_chunk_1303_cast")];
            tensor<string, []> var_13001_equation_0 = const()[name = tensor<string, []>("op_13001_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13001_cast = einsum(equation = var_13001_equation_0, values = (var_12783_cast, var_12739_cast))[name = tensor<string, []>("op_13001_cast")];
            tensor<fp16, []> var_13002_to_fp16 = const()[name = tensor<string, []>("op_13002_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1305_cast = mul(x = var_13001_cast, y = var_13002_to_fp16)[name = tensor<string, []>("aw_chunk_1305_cast")];
            tensor<string, []> var_13005_equation_0 = const()[name = tensor<string, []>("op_13005_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13005_cast = einsum(equation = var_13005_equation_0, values = (var_12783_cast, var_12740_cast))[name = tensor<string, []>("op_13005_cast")];
            tensor<fp16, []> var_13006_to_fp16 = const()[name = tensor<string, []>("op_13006_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1307_cast = mul(x = var_13005_cast, y = var_13006_to_fp16)[name = tensor<string, []>("aw_chunk_1307_cast")];
            tensor<string, []> var_13009_equation_0 = const()[name = tensor<string, []>("op_13009_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13009_cast = einsum(equation = var_13009_equation_0, values = (var_12783_cast, var_12741_cast))[name = tensor<string, []>("op_13009_cast")];
            tensor<fp16, []> var_13010_to_fp16 = const()[name = tensor<string, []>("op_13010_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1309_cast = mul(x = var_13009_cast, y = var_13010_to_fp16)[name = tensor<string, []>("aw_chunk_1309_cast")];
            tensor<string, []> var_13013_equation_0 = const()[name = tensor<string, []>("op_13013_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13013_cast = einsum(equation = var_13013_equation_0, values = (var_12783_cast, var_12742_cast))[name = tensor<string, []>("op_13013_cast")];
            tensor<fp16, []> var_13014_to_fp16 = const()[name = tensor<string, []>("op_13014_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1311_cast = mul(x = var_13013_cast, y = var_13014_to_fp16)[name = tensor<string, []>("aw_chunk_1311_cast")];
            tensor<string, []> var_13017_equation_0 = const()[name = tensor<string, []>("op_13017_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13017_cast = einsum(equation = var_13017_equation_0, values = (var_12787_cast, var_12743_cast))[name = tensor<string, []>("op_13017_cast")];
            tensor<fp16, []> var_13018_to_fp16 = const()[name = tensor<string, []>("op_13018_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1313_cast = mul(x = var_13017_cast, y = var_13018_to_fp16)[name = tensor<string, []>("aw_chunk_1313_cast")];
            tensor<string, []> var_13021_equation_0 = const()[name = tensor<string, []>("op_13021_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13021_cast = einsum(equation = var_13021_equation_0, values = (var_12787_cast, var_12744_cast))[name = tensor<string, []>("op_13021_cast")];
            tensor<fp16, []> var_13022_to_fp16 = const()[name = tensor<string, []>("op_13022_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1315_cast = mul(x = var_13021_cast, y = var_13022_to_fp16)[name = tensor<string, []>("aw_chunk_1315_cast")];
            tensor<string, []> var_13025_equation_0 = const()[name = tensor<string, []>("op_13025_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13025_cast = einsum(equation = var_13025_equation_0, values = (var_12787_cast, var_12745_cast))[name = tensor<string, []>("op_13025_cast")];
            tensor<fp16, []> var_13026_to_fp16 = const()[name = tensor<string, []>("op_13026_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1317_cast = mul(x = var_13025_cast, y = var_13026_to_fp16)[name = tensor<string, []>("aw_chunk_1317_cast")];
            tensor<string, []> var_13029_equation_0 = const()[name = tensor<string, []>("op_13029_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13029_cast = einsum(equation = var_13029_equation_0, values = (var_12787_cast, var_12746_cast))[name = tensor<string, []>("op_13029_cast")];
            tensor<fp16, []> var_13030_to_fp16 = const()[name = tensor<string, []>("op_13030_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1319_cast = mul(x = var_13029_cast, y = var_13030_to_fp16)[name = tensor<string, []>("aw_chunk_1319_cast")];
            tensor<string, []> var_13033_equation_0 = const()[name = tensor<string, []>("op_13033_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13033_cast = einsum(equation = var_13033_equation_0, values = (var_12787_cast, var_12747_cast))[name = tensor<string, []>("op_13033_cast")];
            tensor<fp16, []> var_13034_to_fp16 = const()[name = tensor<string, []>("op_13034_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1321_cast = mul(x = var_13033_cast, y = var_13034_to_fp16)[name = tensor<string, []>("aw_chunk_1321_cast")];
            tensor<string, []> var_13037_equation_0 = const()[name = tensor<string, []>("op_13037_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13037_cast = einsum(equation = var_13037_equation_0, values = (var_12787_cast, var_12748_cast))[name = tensor<string, []>("op_13037_cast")];
            tensor<fp16, []> var_13038_to_fp16 = const()[name = tensor<string, []>("op_13038_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1323_cast = mul(x = var_13037_cast, y = var_13038_to_fp16)[name = tensor<string, []>("aw_chunk_1323_cast")];
            tensor<string, []> var_13041_equation_0 = const()[name = tensor<string, []>("op_13041_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13041_cast = einsum(equation = var_13041_equation_0, values = (var_12787_cast, var_12749_cast))[name = tensor<string, []>("op_13041_cast")];
            tensor<fp16, []> var_13042_to_fp16 = const()[name = tensor<string, []>("op_13042_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1325_cast = mul(x = var_13041_cast, y = var_13042_to_fp16)[name = tensor<string, []>("aw_chunk_1325_cast")];
            tensor<string, []> var_13045_equation_0 = const()[name = tensor<string, []>("op_13045_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13045_cast = einsum(equation = var_13045_equation_0, values = (var_12787_cast, var_12750_cast))[name = tensor<string, []>("op_13045_cast")];
            tensor<fp16, []> var_13046_to_fp16 = const()[name = tensor<string, []>("op_13046_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1327_cast = mul(x = var_13045_cast, y = var_13046_to_fp16)[name = tensor<string, []>("aw_chunk_1327_cast")];
            tensor<string, []> var_13049_equation_0 = const()[name = tensor<string, []>("op_13049_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13049_cast = einsum(equation = var_13049_equation_0, values = (var_12791_cast, var_12751_cast))[name = tensor<string, []>("op_13049_cast")];
            tensor<fp16, []> var_13050_to_fp16 = const()[name = tensor<string, []>("op_13050_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1329_cast = mul(x = var_13049_cast, y = var_13050_to_fp16)[name = tensor<string, []>("aw_chunk_1329_cast")];
            tensor<string, []> var_13053_equation_0 = const()[name = tensor<string, []>("op_13053_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13053_cast = einsum(equation = var_13053_equation_0, values = (var_12791_cast, var_12752_cast))[name = tensor<string, []>("op_13053_cast")];
            tensor<fp16, []> var_13054_to_fp16 = const()[name = tensor<string, []>("op_13054_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1331_cast = mul(x = var_13053_cast, y = var_13054_to_fp16)[name = tensor<string, []>("aw_chunk_1331_cast")];
            tensor<string, []> var_13057_equation_0 = const()[name = tensor<string, []>("op_13057_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13057_cast = einsum(equation = var_13057_equation_0, values = (var_12791_cast, var_12753_cast))[name = tensor<string, []>("op_13057_cast")];
            tensor<fp16, []> var_13058_to_fp16 = const()[name = tensor<string, []>("op_13058_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1333_cast = mul(x = var_13057_cast, y = var_13058_to_fp16)[name = tensor<string, []>("aw_chunk_1333_cast")];
            tensor<string, []> var_13061_equation_0 = const()[name = tensor<string, []>("op_13061_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13061_cast = einsum(equation = var_13061_equation_0, values = (var_12791_cast, var_12754_cast))[name = tensor<string, []>("op_13061_cast")];
            tensor<fp16, []> var_13062_to_fp16 = const()[name = tensor<string, []>("op_13062_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1335_cast = mul(x = var_13061_cast, y = var_13062_to_fp16)[name = tensor<string, []>("aw_chunk_1335_cast")];
            tensor<string, []> var_13065_equation_0 = const()[name = tensor<string, []>("op_13065_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13065_cast = einsum(equation = var_13065_equation_0, values = (var_12791_cast, var_12755_cast))[name = tensor<string, []>("op_13065_cast")];
            tensor<fp16, []> var_13066_to_fp16 = const()[name = tensor<string, []>("op_13066_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1337_cast = mul(x = var_13065_cast, y = var_13066_to_fp16)[name = tensor<string, []>("aw_chunk_1337_cast")];
            tensor<string, []> var_13069_equation_0 = const()[name = tensor<string, []>("op_13069_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13069_cast = einsum(equation = var_13069_equation_0, values = (var_12791_cast, var_12756_cast))[name = tensor<string, []>("op_13069_cast")];
            tensor<fp16, []> var_13070_to_fp16 = const()[name = tensor<string, []>("op_13070_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1339_cast = mul(x = var_13069_cast, y = var_13070_to_fp16)[name = tensor<string, []>("aw_chunk_1339_cast")];
            tensor<string, []> var_13073_equation_0 = const()[name = tensor<string, []>("op_13073_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13073_cast = einsum(equation = var_13073_equation_0, values = (var_12791_cast, var_12757_cast))[name = tensor<string, []>("op_13073_cast")];
            tensor<fp16, []> var_13074_to_fp16 = const()[name = tensor<string, []>("op_13074_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1341_cast = mul(x = var_13073_cast, y = var_13074_to_fp16)[name = tensor<string, []>("aw_chunk_1341_cast")];
            tensor<string, []> var_13077_equation_0 = const()[name = tensor<string, []>("op_13077_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13077_cast = einsum(equation = var_13077_equation_0, values = (var_12791_cast, var_12758_cast))[name = tensor<string, []>("op_13077_cast")];
            tensor<fp16, []> var_13078_to_fp16 = const()[name = tensor<string, []>("op_13078_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1343_cast = mul(x = var_13077_cast, y = var_13078_to_fp16)[name = tensor<string, []>("aw_chunk_1343_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13080_cast = softmax(axis = var_10375, x = aw_chunk_1217_cast)[name = tensor<string, []>("op_13080_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13081_cast = softmax(axis = var_10375, x = aw_chunk_1219_cast)[name = tensor<string, []>("op_13081_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13082_cast = softmax(axis = var_10375, x = aw_chunk_1221_cast)[name = tensor<string, []>("op_13082_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13083_cast = softmax(axis = var_10375, x = aw_chunk_1223_cast)[name = tensor<string, []>("op_13083_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13084_cast = softmax(axis = var_10375, x = aw_chunk_1225_cast)[name = tensor<string, []>("op_13084_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13085_cast = softmax(axis = var_10375, x = aw_chunk_1227_cast)[name = tensor<string, []>("op_13085_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13086_cast = softmax(axis = var_10375, x = aw_chunk_1229_cast)[name = tensor<string, []>("op_13086_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13087_cast = softmax(axis = var_10375, x = aw_chunk_1231_cast)[name = tensor<string, []>("op_13087_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13088_cast = softmax(axis = var_10375, x = aw_chunk_1233_cast)[name = tensor<string, []>("op_13088_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13089_cast = softmax(axis = var_10375, x = aw_chunk_1235_cast)[name = tensor<string, []>("op_13089_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13090_cast = softmax(axis = var_10375, x = aw_chunk_1237_cast)[name = tensor<string, []>("op_13090_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13091_cast = softmax(axis = var_10375, x = aw_chunk_1239_cast)[name = tensor<string, []>("op_13091_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13092_cast = softmax(axis = var_10375, x = aw_chunk_1241_cast)[name = tensor<string, []>("op_13092_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13093_cast = softmax(axis = var_10375, x = aw_chunk_1243_cast)[name = tensor<string, []>("op_13093_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13094_cast = softmax(axis = var_10375, x = aw_chunk_1245_cast)[name = tensor<string, []>("op_13094_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13095_cast = softmax(axis = var_10375, x = aw_chunk_1247_cast)[name = tensor<string, []>("op_13095_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13096_cast = softmax(axis = var_10375, x = aw_chunk_1249_cast)[name = tensor<string, []>("op_13096_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13097_cast = softmax(axis = var_10375, x = aw_chunk_1251_cast)[name = tensor<string, []>("op_13097_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13098_cast = softmax(axis = var_10375, x = aw_chunk_1253_cast)[name = tensor<string, []>("op_13098_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13099_cast = softmax(axis = var_10375, x = aw_chunk_1255_cast)[name = tensor<string, []>("op_13099_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13100_cast = softmax(axis = var_10375, x = aw_chunk_1257_cast)[name = tensor<string, []>("op_13100_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13101_cast = softmax(axis = var_10375, x = aw_chunk_1259_cast)[name = tensor<string, []>("op_13101_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13102_cast = softmax(axis = var_10375, x = aw_chunk_1261_cast)[name = tensor<string, []>("op_13102_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13103_cast = softmax(axis = var_10375, x = aw_chunk_1263_cast)[name = tensor<string, []>("op_13103_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13104_cast = softmax(axis = var_10375, x = aw_chunk_1265_cast)[name = tensor<string, []>("op_13104_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13105_cast = softmax(axis = var_10375, x = aw_chunk_1267_cast)[name = tensor<string, []>("op_13105_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13106_cast = softmax(axis = var_10375, x = aw_chunk_1269_cast)[name = tensor<string, []>("op_13106_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13107_cast = softmax(axis = var_10375, x = aw_chunk_1271_cast)[name = tensor<string, []>("op_13107_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13108_cast = softmax(axis = var_10375, x = aw_chunk_1273_cast)[name = tensor<string, []>("op_13108_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13109_cast = softmax(axis = var_10375, x = aw_chunk_1275_cast)[name = tensor<string, []>("op_13109_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13110_cast = softmax(axis = var_10375, x = aw_chunk_1277_cast)[name = tensor<string, []>("op_13110_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13111_cast = softmax(axis = var_10375, x = aw_chunk_1279_cast)[name = tensor<string, []>("op_13111_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13112_cast = softmax(axis = var_10375, x = aw_chunk_1281_cast)[name = tensor<string, []>("op_13112_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13113_cast = softmax(axis = var_10375, x = aw_chunk_1283_cast)[name = tensor<string, []>("op_13113_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13114_cast = softmax(axis = var_10375, x = aw_chunk_1285_cast)[name = tensor<string, []>("op_13114_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13115_cast = softmax(axis = var_10375, x = aw_chunk_1287_cast)[name = tensor<string, []>("op_13115_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13116_cast = softmax(axis = var_10375, x = aw_chunk_1289_cast)[name = tensor<string, []>("op_13116_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13117_cast = softmax(axis = var_10375, x = aw_chunk_1291_cast)[name = tensor<string, []>("op_13117_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13118_cast = softmax(axis = var_10375, x = aw_chunk_1293_cast)[name = tensor<string, []>("op_13118_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13119_cast = softmax(axis = var_10375, x = aw_chunk_1295_cast)[name = tensor<string, []>("op_13119_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13120_cast = softmax(axis = var_10375, x = aw_chunk_1297_cast)[name = tensor<string, []>("op_13120_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13121_cast = softmax(axis = var_10375, x = aw_chunk_1299_cast)[name = tensor<string, []>("op_13121_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13122_cast = softmax(axis = var_10375, x = aw_chunk_1301_cast)[name = tensor<string, []>("op_13122_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13123_cast = softmax(axis = var_10375, x = aw_chunk_1303_cast)[name = tensor<string, []>("op_13123_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13124_cast = softmax(axis = var_10375, x = aw_chunk_1305_cast)[name = tensor<string, []>("op_13124_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13125_cast = softmax(axis = var_10375, x = aw_chunk_1307_cast)[name = tensor<string, []>("op_13125_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13126_cast = softmax(axis = var_10375, x = aw_chunk_1309_cast)[name = tensor<string, []>("op_13126_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13127_cast = softmax(axis = var_10375, x = aw_chunk_1311_cast)[name = tensor<string, []>("op_13127_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13128_cast = softmax(axis = var_10375, x = aw_chunk_1313_cast)[name = tensor<string, []>("op_13128_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13129_cast = softmax(axis = var_10375, x = aw_chunk_1315_cast)[name = tensor<string, []>("op_13129_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13130_cast = softmax(axis = var_10375, x = aw_chunk_1317_cast)[name = tensor<string, []>("op_13130_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13131_cast = softmax(axis = var_10375, x = aw_chunk_1319_cast)[name = tensor<string, []>("op_13131_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13132_cast = softmax(axis = var_10375, x = aw_chunk_1321_cast)[name = tensor<string, []>("op_13132_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13133_cast = softmax(axis = var_10375, x = aw_chunk_1323_cast)[name = tensor<string, []>("op_13133_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13134_cast = softmax(axis = var_10375, x = aw_chunk_1325_cast)[name = tensor<string, []>("op_13134_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13135_cast = softmax(axis = var_10375, x = aw_chunk_1327_cast)[name = tensor<string, []>("op_13135_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13136_cast = softmax(axis = var_10375, x = aw_chunk_1329_cast)[name = tensor<string, []>("op_13136_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13137_cast = softmax(axis = var_10375, x = aw_chunk_1331_cast)[name = tensor<string, []>("op_13137_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13138_cast = softmax(axis = var_10375, x = aw_chunk_1333_cast)[name = tensor<string, []>("op_13138_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13139_cast = softmax(axis = var_10375, x = aw_chunk_1335_cast)[name = tensor<string, []>("op_13139_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13140_cast = softmax(axis = var_10375, x = aw_chunk_1337_cast)[name = tensor<string, []>("op_13140_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13141_cast = softmax(axis = var_10375, x = aw_chunk_1339_cast)[name = tensor<string, []>("op_13141_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13142_cast = softmax(axis = var_10375, x = aw_chunk_1341_cast)[name = tensor<string, []>("op_13142_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13143_cast = softmax(axis = var_10375, x = aw_chunk_1343_cast)[name = tensor<string, []>("op_13143_cast")];
            tensor<string, []> var_13145_equation_0 = const()[name = tensor<string, []>("op_13145_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13145_cast = einsum(equation = var_13145_equation_0, values = (var_12793_cast, var_13080_cast))[name = tensor<string, []>("op_13145_cast")];
            tensor<string, []> var_13147_equation_0 = const()[name = tensor<string, []>("op_13147_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13147_cast = einsum(equation = var_13147_equation_0, values = (var_12793_cast, var_13081_cast))[name = tensor<string, []>("op_13147_cast")];
            tensor<string, []> var_13149_equation_0 = const()[name = tensor<string, []>("op_13149_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13149_cast = einsum(equation = var_13149_equation_0, values = (var_12793_cast, var_13082_cast))[name = tensor<string, []>("op_13149_cast")];
            tensor<string, []> var_13151_equation_0 = const()[name = tensor<string, []>("op_13151_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13151_cast = einsum(equation = var_13151_equation_0, values = (var_12793_cast, var_13083_cast))[name = tensor<string, []>("op_13151_cast")];
            tensor<string, []> var_13153_equation_0 = const()[name = tensor<string, []>("op_13153_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13153_cast = einsum(equation = var_13153_equation_0, values = (var_12793_cast, var_13084_cast))[name = tensor<string, []>("op_13153_cast")];
            tensor<string, []> var_13155_equation_0 = const()[name = tensor<string, []>("op_13155_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13155_cast = einsum(equation = var_13155_equation_0, values = (var_12793_cast, var_13085_cast))[name = tensor<string, []>("op_13155_cast")];
            tensor<string, []> var_13157_equation_0 = const()[name = tensor<string, []>("op_13157_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13157_cast = einsum(equation = var_13157_equation_0, values = (var_12793_cast, var_13086_cast))[name = tensor<string, []>("op_13157_cast")];
            tensor<string, []> var_13159_equation_0 = const()[name = tensor<string, []>("op_13159_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13159_cast = einsum(equation = var_13159_equation_0, values = (var_12793_cast, var_13087_cast))[name = tensor<string, []>("op_13159_cast")];
            tensor<string, []> var_13161_equation_0 = const()[name = tensor<string, []>("op_13161_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13161_cast = einsum(equation = var_13161_equation_0, values = (var_12797_cast, var_13088_cast))[name = tensor<string, []>("op_13161_cast")];
            tensor<string, []> var_13163_equation_0 = const()[name = tensor<string, []>("op_13163_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13163_cast = einsum(equation = var_13163_equation_0, values = (var_12797_cast, var_13089_cast))[name = tensor<string, []>("op_13163_cast")];
            tensor<string, []> var_13165_equation_0 = const()[name = tensor<string, []>("op_13165_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13165_cast = einsum(equation = var_13165_equation_0, values = (var_12797_cast, var_13090_cast))[name = tensor<string, []>("op_13165_cast")];
            tensor<string, []> var_13167_equation_0 = const()[name = tensor<string, []>("op_13167_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13167_cast = einsum(equation = var_13167_equation_0, values = (var_12797_cast, var_13091_cast))[name = tensor<string, []>("op_13167_cast")];
            tensor<string, []> var_13169_equation_0 = const()[name = tensor<string, []>("op_13169_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13169_cast = einsum(equation = var_13169_equation_0, values = (var_12797_cast, var_13092_cast))[name = tensor<string, []>("op_13169_cast")];
            tensor<string, []> var_13171_equation_0 = const()[name = tensor<string, []>("op_13171_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13171_cast = einsum(equation = var_13171_equation_0, values = (var_12797_cast, var_13093_cast))[name = tensor<string, []>("op_13171_cast")];
            tensor<string, []> var_13173_equation_0 = const()[name = tensor<string, []>("op_13173_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13173_cast = einsum(equation = var_13173_equation_0, values = (var_12797_cast, var_13094_cast))[name = tensor<string, []>("op_13173_cast")];
            tensor<string, []> var_13175_equation_0 = const()[name = tensor<string, []>("op_13175_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13175_cast = einsum(equation = var_13175_equation_0, values = (var_12797_cast, var_13095_cast))[name = tensor<string, []>("op_13175_cast")];
            tensor<string, []> var_13177_equation_0 = const()[name = tensor<string, []>("op_13177_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13177_cast = einsum(equation = var_13177_equation_0, values = (var_12801_cast, var_13096_cast))[name = tensor<string, []>("op_13177_cast")];
            tensor<string, []> var_13179_equation_0 = const()[name = tensor<string, []>("op_13179_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13179_cast = einsum(equation = var_13179_equation_0, values = (var_12801_cast, var_13097_cast))[name = tensor<string, []>("op_13179_cast")];
            tensor<string, []> var_13181_equation_0 = const()[name = tensor<string, []>("op_13181_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13181_cast = einsum(equation = var_13181_equation_0, values = (var_12801_cast, var_13098_cast))[name = tensor<string, []>("op_13181_cast")];
            tensor<string, []> var_13183_equation_0 = const()[name = tensor<string, []>("op_13183_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13183_cast = einsum(equation = var_13183_equation_0, values = (var_12801_cast, var_13099_cast))[name = tensor<string, []>("op_13183_cast")];
            tensor<string, []> var_13185_equation_0 = const()[name = tensor<string, []>("op_13185_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13185_cast = einsum(equation = var_13185_equation_0, values = (var_12801_cast, var_13100_cast))[name = tensor<string, []>("op_13185_cast")];
            tensor<string, []> var_13187_equation_0 = const()[name = tensor<string, []>("op_13187_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13187_cast = einsum(equation = var_13187_equation_0, values = (var_12801_cast, var_13101_cast))[name = tensor<string, []>("op_13187_cast")];
            tensor<string, []> var_13189_equation_0 = const()[name = tensor<string, []>("op_13189_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13189_cast = einsum(equation = var_13189_equation_0, values = (var_12801_cast, var_13102_cast))[name = tensor<string, []>("op_13189_cast")];
            tensor<string, []> var_13191_equation_0 = const()[name = tensor<string, []>("op_13191_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13191_cast = einsum(equation = var_13191_equation_0, values = (var_12801_cast, var_13103_cast))[name = tensor<string, []>("op_13191_cast")];
            tensor<string, []> var_13193_equation_0 = const()[name = tensor<string, []>("op_13193_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13193_cast = einsum(equation = var_13193_equation_0, values = (var_12805_cast, var_13104_cast))[name = tensor<string, []>("op_13193_cast")];
            tensor<string, []> var_13195_equation_0 = const()[name = tensor<string, []>("op_13195_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13195_cast = einsum(equation = var_13195_equation_0, values = (var_12805_cast, var_13105_cast))[name = tensor<string, []>("op_13195_cast")];
            tensor<string, []> var_13197_equation_0 = const()[name = tensor<string, []>("op_13197_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13197_cast = einsum(equation = var_13197_equation_0, values = (var_12805_cast, var_13106_cast))[name = tensor<string, []>("op_13197_cast")];
            tensor<string, []> var_13199_equation_0 = const()[name = tensor<string, []>("op_13199_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13199_cast = einsum(equation = var_13199_equation_0, values = (var_12805_cast, var_13107_cast))[name = tensor<string, []>("op_13199_cast")];
            tensor<string, []> var_13201_equation_0 = const()[name = tensor<string, []>("op_13201_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13201_cast = einsum(equation = var_13201_equation_0, values = (var_12805_cast, var_13108_cast))[name = tensor<string, []>("op_13201_cast")];
            tensor<string, []> var_13203_equation_0 = const()[name = tensor<string, []>("op_13203_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13203_cast = einsum(equation = var_13203_equation_0, values = (var_12805_cast, var_13109_cast))[name = tensor<string, []>("op_13203_cast")];
            tensor<string, []> var_13205_equation_0 = const()[name = tensor<string, []>("op_13205_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13205_cast = einsum(equation = var_13205_equation_0, values = (var_12805_cast, var_13110_cast))[name = tensor<string, []>("op_13205_cast")];
            tensor<string, []> var_13207_equation_0 = const()[name = tensor<string, []>("op_13207_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13207_cast = einsum(equation = var_13207_equation_0, values = (var_12805_cast, var_13111_cast))[name = tensor<string, []>("op_13207_cast")];
            tensor<string, []> var_13209_equation_0 = const()[name = tensor<string, []>("op_13209_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13209_cast = einsum(equation = var_13209_equation_0, values = (var_12809_cast, var_13112_cast))[name = tensor<string, []>("op_13209_cast")];
            tensor<string, []> var_13211_equation_0 = const()[name = tensor<string, []>("op_13211_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13211_cast = einsum(equation = var_13211_equation_0, values = (var_12809_cast, var_13113_cast))[name = tensor<string, []>("op_13211_cast")];
            tensor<string, []> var_13213_equation_0 = const()[name = tensor<string, []>("op_13213_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13213_cast = einsum(equation = var_13213_equation_0, values = (var_12809_cast, var_13114_cast))[name = tensor<string, []>("op_13213_cast")];
            tensor<string, []> var_13215_equation_0 = const()[name = tensor<string, []>("op_13215_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13215_cast = einsum(equation = var_13215_equation_0, values = (var_12809_cast, var_13115_cast))[name = tensor<string, []>("op_13215_cast")];
            tensor<string, []> var_13217_equation_0 = const()[name = tensor<string, []>("op_13217_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13217_cast = einsum(equation = var_13217_equation_0, values = (var_12809_cast, var_13116_cast))[name = tensor<string, []>("op_13217_cast")];
            tensor<string, []> var_13219_equation_0 = const()[name = tensor<string, []>("op_13219_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13219_cast = einsum(equation = var_13219_equation_0, values = (var_12809_cast, var_13117_cast))[name = tensor<string, []>("op_13219_cast")];
            tensor<string, []> var_13221_equation_0 = const()[name = tensor<string, []>("op_13221_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13221_cast = einsum(equation = var_13221_equation_0, values = (var_12809_cast, var_13118_cast))[name = tensor<string, []>("op_13221_cast")];
            tensor<string, []> var_13223_equation_0 = const()[name = tensor<string, []>("op_13223_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13223_cast = einsum(equation = var_13223_equation_0, values = (var_12809_cast, var_13119_cast))[name = tensor<string, []>("op_13223_cast")];
            tensor<string, []> var_13225_equation_0 = const()[name = tensor<string, []>("op_13225_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13225_cast = einsum(equation = var_13225_equation_0, values = (var_12813_cast, var_13120_cast))[name = tensor<string, []>("op_13225_cast")];
            tensor<string, []> var_13227_equation_0 = const()[name = tensor<string, []>("op_13227_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13227_cast = einsum(equation = var_13227_equation_0, values = (var_12813_cast, var_13121_cast))[name = tensor<string, []>("op_13227_cast")];
            tensor<string, []> var_13229_equation_0 = const()[name = tensor<string, []>("op_13229_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13229_cast = einsum(equation = var_13229_equation_0, values = (var_12813_cast, var_13122_cast))[name = tensor<string, []>("op_13229_cast")];
            tensor<string, []> var_13231_equation_0 = const()[name = tensor<string, []>("op_13231_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13231_cast = einsum(equation = var_13231_equation_0, values = (var_12813_cast, var_13123_cast))[name = tensor<string, []>("op_13231_cast")];
            tensor<string, []> var_13233_equation_0 = const()[name = tensor<string, []>("op_13233_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13233_cast = einsum(equation = var_13233_equation_0, values = (var_12813_cast, var_13124_cast))[name = tensor<string, []>("op_13233_cast")];
            tensor<string, []> var_13235_equation_0 = const()[name = tensor<string, []>("op_13235_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13235_cast = einsum(equation = var_13235_equation_0, values = (var_12813_cast, var_13125_cast))[name = tensor<string, []>("op_13235_cast")];
            tensor<string, []> var_13237_equation_0 = const()[name = tensor<string, []>("op_13237_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13237_cast = einsum(equation = var_13237_equation_0, values = (var_12813_cast, var_13126_cast))[name = tensor<string, []>("op_13237_cast")];
            tensor<string, []> var_13239_equation_0 = const()[name = tensor<string, []>("op_13239_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13239_cast = einsum(equation = var_13239_equation_0, values = (var_12813_cast, var_13127_cast))[name = tensor<string, []>("op_13239_cast")];
            tensor<string, []> var_13241_equation_0 = const()[name = tensor<string, []>("op_13241_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13241_cast = einsum(equation = var_13241_equation_0, values = (var_12817_cast, var_13128_cast))[name = tensor<string, []>("op_13241_cast")];
            tensor<string, []> var_13243_equation_0 = const()[name = tensor<string, []>("op_13243_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13243_cast = einsum(equation = var_13243_equation_0, values = (var_12817_cast, var_13129_cast))[name = tensor<string, []>("op_13243_cast")];
            tensor<string, []> var_13245_equation_0 = const()[name = tensor<string, []>("op_13245_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13245_cast = einsum(equation = var_13245_equation_0, values = (var_12817_cast, var_13130_cast))[name = tensor<string, []>("op_13245_cast")];
            tensor<string, []> var_13247_equation_0 = const()[name = tensor<string, []>("op_13247_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13247_cast = einsum(equation = var_13247_equation_0, values = (var_12817_cast, var_13131_cast))[name = tensor<string, []>("op_13247_cast")];
            tensor<string, []> var_13249_equation_0 = const()[name = tensor<string, []>("op_13249_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13249_cast = einsum(equation = var_13249_equation_0, values = (var_12817_cast, var_13132_cast))[name = tensor<string, []>("op_13249_cast")];
            tensor<string, []> var_13251_equation_0 = const()[name = tensor<string, []>("op_13251_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13251_cast = einsum(equation = var_13251_equation_0, values = (var_12817_cast, var_13133_cast))[name = tensor<string, []>("op_13251_cast")];
            tensor<string, []> var_13253_equation_0 = const()[name = tensor<string, []>("op_13253_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13253_cast = einsum(equation = var_13253_equation_0, values = (var_12817_cast, var_13134_cast))[name = tensor<string, []>("op_13253_cast")];
            tensor<string, []> var_13255_equation_0 = const()[name = tensor<string, []>("op_13255_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13255_cast = einsum(equation = var_13255_equation_0, values = (var_12817_cast, var_13135_cast))[name = tensor<string, []>("op_13255_cast")];
            tensor<string, []> var_13257_equation_0 = const()[name = tensor<string, []>("op_13257_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13257_cast = einsum(equation = var_13257_equation_0, values = (var_12821_cast, var_13136_cast))[name = tensor<string, []>("op_13257_cast")];
            tensor<string, []> var_13259_equation_0 = const()[name = tensor<string, []>("op_13259_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13259_cast = einsum(equation = var_13259_equation_0, values = (var_12821_cast, var_13137_cast))[name = tensor<string, []>("op_13259_cast")];
            tensor<string, []> var_13261_equation_0 = const()[name = tensor<string, []>("op_13261_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13261_cast = einsum(equation = var_13261_equation_0, values = (var_12821_cast, var_13138_cast))[name = tensor<string, []>("op_13261_cast")];
            tensor<string, []> var_13263_equation_0 = const()[name = tensor<string, []>("op_13263_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13263_cast = einsum(equation = var_13263_equation_0, values = (var_12821_cast, var_13139_cast))[name = tensor<string, []>("op_13263_cast")];
            tensor<string, []> var_13265_equation_0 = const()[name = tensor<string, []>("op_13265_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13265_cast = einsum(equation = var_13265_equation_0, values = (var_12821_cast, var_13140_cast))[name = tensor<string, []>("op_13265_cast")];
            tensor<string, []> var_13267_equation_0 = const()[name = tensor<string, []>("op_13267_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13267_cast = einsum(equation = var_13267_equation_0, values = (var_12821_cast, var_13141_cast))[name = tensor<string, []>("op_13267_cast")];
            tensor<string, []> var_13269_equation_0 = const()[name = tensor<string, []>("op_13269_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13269_cast = einsum(equation = var_13269_equation_0, values = (var_12821_cast, var_13142_cast))[name = tensor<string, []>("op_13269_cast")];
            tensor<string, []> var_13271_equation_0 = const()[name = tensor<string, []>("op_13271_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13271_cast = einsum(equation = var_13271_equation_0, values = (var_12821_cast, var_13143_cast))[name = tensor<string, []>("op_13271_cast")];
            tensor<bool, []> var_13273_interleave_0 = const()[name = tensor<string, []>("op_13273_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_13273_cast = concat(axis = var_10347, interleave = var_13273_interleave_0, values = (var_13145_cast, var_13147_cast, var_13149_cast, var_13151_cast, var_13153_cast, var_13155_cast, var_13157_cast, var_13159_cast))[name = tensor<string, []>("op_13273_cast")];
            tensor<bool, []> var_13275_interleave_0 = const()[name = tensor<string, []>("op_13275_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_13275_cast = concat(axis = var_10347, interleave = var_13275_interleave_0, values = (var_13161_cast, var_13163_cast, var_13165_cast, var_13167_cast, var_13169_cast, var_13171_cast, var_13173_cast, var_13175_cast))[name = tensor<string, []>("op_13275_cast")];
            tensor<bool, []> var_13277_interleave_0 = const()[name = tensor<string, []>("op_13277_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_13277_cast = concat(axis = var_10347, interleave = var_13277_interleave_0, values = (var_13177_cast, var_13179_cast, var_13181_cast, var_13183_cast, var_13185_cast, var_13187_cast, var_13189_cast, var_13191_cast))[name = tensor<string, []>("op_13277_cast")];
            tensor<bool, []> var_13279_interleave_0 = const()[name = tensor<string, []>("op_13279_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_13279_cast = concat(axis = var_10347, interleave = var_13279_interleave_0, values = (var_13193_cast, var_13195_cast, var_13197_cast, var_13199_cast, var_13201_cast, var_13203_cast, var_13205_cast, var_13207_cast))[name = tensor<string, []>("op_13279_cast")];
            tensor<bool, []> var_13281_interleave_0 = const()[name = tensor<string, []>("op_13281_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_13281_cast = concat(axis = var_10347, interleave = var_13281_interleave_0, values = (var_13209_cast, var_13211_cast, var_13213_cast, var_13215_cast, var_13217_cast, var_13219_cast, var_13221_cast, var_13223_cast))[name = tensor<string, []>("op_13281_cast")];
            tensor<bool, []> var_13283_interleave_0 = const()[name = tensor<string, []>("op_13283_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_13283_cast = concat(axis = var_10347, interleave = var_13283_interleave_0, values = (var_13225_cast, var_13227_cast, var_13229_cast, var_13231_cast, var_13233_cast, var_13235_cast, var_13237_cast, var_13239_cast))[name = tensor<string, []>("op_13283_cast")];
            tensor<bool, []> var_13285_interleave_0 = const()[name = tensor<string, []>("op_13285_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_13285_cast = concat(axis = var_10347, interleave = var_13285_interleave_0, values = (var_13241_cast, var_13243_cast, var_13245_cast, var_13247_cast, var_13249_cast, var_13251_cast, var_13253_cast, var_13255_cast))[name = tensor<string, []>("op_13285_cast")];
            tensor<bool, []> var_13287_interleave_0 = const()[name = tensor<string, []>("op_13287_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_13287_cast = concat(axis = var_10347, interleave = var_13287_interleave_0, values = (var_13257_cast, var_13259_cast, var_13261_cast, var_13263_cast, var_13265_cast, var_13267_cast, var_13269_cast, var_13271_cast))[name = tensor<string, []>("op_13287_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_10375, interleave = input_493_interleave_0, values = (var_13273_cast, var_13275_cast, var_13277_cast, var_13279_cast, var_13281_cast, var_13283_cast, var_13285_cast, var_13287_cast))[name = tensor<string, []>("input_493_cast")];
            tensor<int32, [2]> var_13293 = const()[name = tensor<string, []>("op_13293"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13295 = const()[name = tensor<string, []>("op_13295"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_13297_pad_type_0 = const()[name = tensor<string, []>("op_13297_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_13297_pad_0 = const()[name = tensor<string, []>("op_13297_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(639614016))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(639690880))), name = tensor<string, []>("up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            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, []>(639691072)))];
            tensor<fp16, [2, 320, 1, 4096]> var_13297_cast = conv(bias = up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_13295, groups = var_10375, pad = var_13297_pad_0, pad_type = var_13297_pad_type_0, strides = var_13293, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_493_cast)[name = tensor<string, []>("op_13297_cast")];
            tensor<fp16, [2, 320, 1, 4096]> inputs_89_cast = add(x = var_13297_cast, y = inputs_87_cast)[name = tensor<string, []>("inputs_89_cast")];
            tensor<int32, [1]> var_13301 = const()[name = tensor<string, []>("op_13301"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_89_cast = reduce_mean(axes = var_13301, keep_dims = var_10370, x = inputs_89_cast)[name = tensor<string, []>("channels_mean_89_cast")];
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_89_cast = sub(x = inputs_89_cast, y = channels_mean_89_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_89_cast")];
            tensor<int32, [1]> var_13305 = const()[name = tensor<string, []>("op_13305"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_13306_cast = reduce_mean(axes = var_13305, keep_dims = var_10370, x = zero_mean_sq_89_cast)[name = tensor<string, []>("op_13306_cast")];
            tensor<fp16, []> var_13307_to_fp16 = const()[name = tensor<string, []>("op_13307_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_13308_cast = add(x = var_13306_cast, y = var_13307_to_fp16)[name = tensor<string, []>("op_13308_cast")];
            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_13308_cast)[name = tensor<string, []>("denom_89_cast")];
            tensor<fp16, [2, 320, 1, 4096]> out_89_cast = mul(x = zero_mean_89_cast, y = denom_89_cast)[name = tensor<string, []>("out_89_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_13312_to_fp16 = const()[name = tensor<string, []>("op_13312_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(639691776)))];
            tensor<fp16, [2, 320, 1, 4096]> var_13313_cast = add(x = out_89_cast, y = var_13312_to_fp16)[name = tensor<string, []>("op_13313_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_13315_to_fp16 = const()[name = tensor<string, []>("op_13315_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(639692480)))];
            tensor<fp16, [2, 320, 1, 4096]> input_495_cast = mul(x = var_13313_cast, y = var_13315_to_fp16)[name = tensor<string, []>("input_495_cast")];
            tensor<int32, [2]> var_13323 = const()[name = tensor<string, []>("op_13323"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13325 = const()[name = tensor<string, []>("op_13325"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_13327_pad_type_0 = const()[name = tensor<string, []>("op_13327_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_13327_pad_0 = const()[name = tensor<string, []>("op_13327_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [614400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(639693184))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(640307648))), name = tensor<string, []>("up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([2560, 320, 1, 1])];
            tensor<fp16, [2560]> up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(640307840))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(640309824))), name = tensor<string, []>("up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([2560])];
            tensor<fp16, [2, 2560, 1, 4096]> var_13327_cast = conv(bias = up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_13325, groups = var_10375, pad = var_13327_pad_0, pad_type = var_13327_pad_type_0, strides = var_13323, weight = up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_495_cast)[name = tensor<string, []>("op_13327_cast")];
            tensor<int32, [2]> var_13328_split_sizes_0 = const()[name = tensor<string, []>("op_13328_split_sizes_0"), val = tensor<int32, [2]>([1280, 1280])];
            tensor<int32, []> var_13328_axis_0 = const()[name = tensor<string, []>("op_13328_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 1280, 1, 4096]> var_13328_cast_0, tensor<fp16, [2, 1280, 1, 4096]> var_13328_cast_1 = split(axis = var_13328_axis_0, split_sizes = var_13328_split_sizes_0, x = var_13327_cast)[name = tensor<string, []>("op_13328_cast")];
            tensor<string, []> var_13330_mode_0 = const()[name = tensor<string, []>("op_13330_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 1280, 1, 4096]> var_13330_cast = gelu(mode = var_13330_mode_0, x = var_13328_cast_1)[name = tensor<string, []>("op_13330_cast")];
            tensor<fp16, [2, 1280, 1, 4096]> input_497_cast = mul(x = var_13328_cast_0, y = var_13330_cast)[name = tensor<string, []>("input_497_cast")];
            tensor<int32, [2]> var_13334 = const()[name = tensor<string, []>("op_13334"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13336 = const()[name = tensor<string, []>("op_13336"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_13338_pad_type_0 = const()[name = tensor<string, []>("op_13338_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_13338_pad_0 = const()[name = tensor<string, []>("op_13338_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(640310016))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(640617280))), name = tensor<string, []>("up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 1280, 1, 1])];
            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, []>(640617472)))];
            tensor<fp16, [2, 320, 1, 4096]> var_13338_cast = conv(bias = up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_13336, groups = var_10375, pad = var_13338_pad_0, pad_type = var_13338_pad_type_0, strides = var_13334, weight = up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_497_cast)[name = tensor<string, []>("op_13338_cast")];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_313_cast = add(x = var_13338_cast, y = inputs_89_cast)[name = tensor<string, []>("hidden_states_313_cast")];
            tensor<int32, [4]> var_13340 = const()[name = tensor<string, []>("op_13340"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> input_499_cast = reshape(shape = var_13340, x = hidden_states_313_cast)[name = tensor<string, []>("input_499_cast")];
            tensor<int32, [2]> var_13344 = const()[name = tensor<string, []>("op_13344"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13346 = const()[name = tensor<string, []>("op_13346"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(640618176))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(640695040))), name = tensor<string, []>("up_blocks_3_attentions_1_proj_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            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, []>(640695232)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_315_cast = conv(bias = up_blocks_3_attentions_1_proj_out_bias_to_fp16, dilations = var_13346, groups = var_10375, pad = hidden_states_315_pad_0, pad_type = hidden_states_315_pad_type_0, strides = var_13344, weight = up_blocks_3_attentions_1_proj_out_weight_to_fp16_palettized, x = input_499_cast)[name = tensor<string, []>("hidden_states_315_cast")];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_317_cast = add(x = hidden_states_315_cast, y = hidden_states_303_cast)[name = tensor<string, []>("hidden_states_317_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_10375, interleave = input_501_interleave_0, values = (hidden_states_317_cast, input_7_cast))[name = tensor<string, []>("input_501_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)[name = tensor<string, []>("reshape_228_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)[name = tensor<string, []>("reduce_mean_171_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> sub_114_cast = sub(x = reshape_228_cast, y = reduce_mean_171_cast)[name = tensor<string, []>("sub_114_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> square_57_cast = square(x = sub_114_cast)[name = tensor<string, []>("square_57_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)[name = tensor<string, []>("reduce_mean_173_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)[name = tensor<string, []>("add_114_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_57_cast = sqrt(x = add_114_cast)[name = tensor<string, []>("sqrt_57_cast")];
            tensor<fp16, [2, 32, 20, 64, 64]> real_div_57_cast = real_div(x = sub_114_cast, y = sqrt_57_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_229_cast")];
            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, []>(640695936)))];
            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, []>(640697280)))];
            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_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_229_cast)[name = tensor<string, []>("add_115_cast")];
            tensor<fp16, [2, 640, 64, 64]> input_505_cast = silu(x = add_115_cast)[name = tensor<string, []>("input_505_cast")];
            tensor<int32, [2]> var_13364 = const()[name = tensor<string, []>("op_13364"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13366 = const()[name = tensor<string, []>("op_13366"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1382400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(640698624))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(642081088))), name = tensor<string, []>("up_blocks_3_resnets_2_conv1_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 640, 3, 3])];
            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, []>(642081280)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_319_cast = conv(bias = up_blocks_3_resnets_2_conv1_bias_to_fp16, dilations = var_13366, groups = var_10375, pad = hidden_states_319_pad_0, pad_type = hidden_states_319_pad_type_0, strides = var_13364, weight = up_blocks_3_resnets_2_conv1_weight_to_fp16_palettized, x = input_505_cast)[name = tensor<string, []>("hidden_states_319_cast")];
            tensor<int32, [2]> var_13372 = const()[name = tensor<string, []>("op_13372"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13374 = const()[name = tensor<string, []>("op_13374"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(642081984))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(642389248))), name = tensor<string, []>("up_blocks_3_resnets_2_time_emb_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 1280, 1, 1])];
            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, []>(642389440)))];
            tensor<fp16, [2, 320, 1, 1]> temb_cast = conv(bias = up_blocks_3_resnets_2_time_emb_proj_bias_to_fp16, dilations = var_13374, groups = var_10375, pad = temb_pad_0, pad_type = temb_pad_type_0, strides = var_13372, weight = up_blocks_3_resnets_2_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor<string, []>("temb_cast")];
            tensor<fp16, [2, 320, 64, 64]> input_509_cast = add(x = hidden_states_319_cast, y = temb_cast)[name = tensor<string, []>("input_509_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)[name = tensor<string, []>("reshape_232_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)[name = tensor<string, []>("reduce_mean_174_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> sub_116_cast = sub(x = reshape_232_cast, y = reduce_mean_174_cast)[name = tensor<string, []>("sub_116_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> square_58_cast = square(x = sub_116_cast)[name = tensor<string, []>("square_58_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)[name = tensor<string, []>("reduce_mean_176_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)[name = tensor<string, []>("add_116_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_58_cast = sqrt(x = add_116_cast)[name = tensor<string, []>("sqrt_58_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> real_div_58_cast = real_div(x = sub_116_cast, y = sqrt_58_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_233_cast")];
            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, []>(642390144)))];
            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, []>(642390848)))];
            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_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_233_cast)[name = tensor<string, []>("add_117_cast")];
            tensor<fp16, [2, 320, 64, 64]> input_513_cast = silu(x = add_117_cast)[name = tensor<string, []>("input_513_cast")];
            tensor<int32, [2]> var_13384 = const()[name = tensor<string, []>("op_13384"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13386 = const()[name = tensor<string, []>("op_13386"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [691200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(642391552))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643082816))), name = tensor<string, []>("up_blocks_3_resnets_2_conv2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 3, 3])];
            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, []>(643083008)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_321_cast = conv(bias = up_blocks_3_resnets_2_conv2_bias_to_fp16, dilations = var_13386, groups = var_10375, pad = hidden_states_321_pad_0, pad_type = hidden_states_321_pad_type_0, strides = var_13384, weight = up_blocks_3_resnets_2_conv2_weight_to_fp16_palettized, x = input_513_cast)[name = tensor<string, []>("hidden_states_321_cast")];
            tensor<int32, [2]> var_13391 = const()[name = tensor<string, []>("op_13391"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13393 = const()[name = tensor<string, []>("op_13393"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [153600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643083712))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643237376))), name = tensor<string, []>("up_blocks_3_resnets_2_conv_shortcut_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 640, 1, 1])];
            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, []>(643237568)))];
            tensor<fp16, [2, 320, 64, 64]> x_cast = conv(bias = up_blocks_3_resnets_2_conv_shortcut_bias_to_fp16, dilations = var_13393, groups = var_10375, pad = x_pad_0, pad_type = x_pad_type_0, strides = var_13391, weight = up_blocks_3_resnets_2_conv_shortcut_weight_to_fp16_palettized, x = input_501_cast)[name = tensor<string, []>("x_cast")];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_323_cast = add(x = x_cast, y = hidden_states_321_cast)[name = tensor<string, []>("hidden_states_323_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)[name = tensor<string, []>("reshape_236_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)[name = tensor<string, []>("reduce_mean_177_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> sub_118_cast = sub(x = reshape_236_cast, y = reduce_mean_177_cast)[name = tensor<string, []>("sub_118_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> square_59_cast = square(x = sub_118_cast)[name = tensor<string, []>("square_59_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)[name = tensor<string, []>("reduce_mean_179_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)[name = tensor<string, []>("add_118_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_59_cast = sqrt(x = add_118_cast)[name = tensor<string, []>("sqrt_59_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> real_div_59_cast = real_div(x = sub_118_cast, y = sqrt_59_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_237_cast")];
            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, []>(643238272)))];
            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, []>(643238976)))];
            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_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_237_cast)[name = tensor<string, []>("add_119_cast")];
            tensor<int32, [2]> var_13413 = const()[name = tensor<string, []>("op_13413"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13415 = const()[name = tensor<string, []>("op_13415"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643239680))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643316544))), name = tensor<string, []>("up_blocks_3_attentions_2_proj_in_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            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, []>(643316736)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_325_cast = conv(bias = up_blocks_3_attentions_2_proj_in_bias_to_fp16, dilations = var_13415, groups = var_10375, pad = hidden_states_325_pad_0, pad_type = hidden_states_325_pad_type_0, strides = var_13413, weight = up_blocks_3_attentions_2_proj_in_weight_to_fp16_palettized, x = add_119_cast)[name = tensor<string, []>("hidden_states_325_cast")];
            tensor<int32, [4]> var_13420 = const()[name = tensor<string, []>("op_13420"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<fp16, [2, 320, 1, 4096]> inputs_91_cast = reshape(shape = var_13420, x = hidden_states_325_cast)[name = tensor<string, []>("inputs_91_cast")];
            tensor<int32, [1]> var_13430 = const()[name = tensor<string, []>("op_13430"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_91_cast = reduce_mean(axes = var_13430, keep_dims = var_10370, x = inputs_91_cast)[name = tensor<string, []>("channels_mean_91_cast")];
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_91_cast = sub(x = inputs_91_cast, y = channels_mean_91_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_91_cast")];
            tensor<int32, [1]> var_13434 = const()[name = tensor<string, []>("op_13434"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_13435_cast = reduce_mean(axes = var_13434, keep_dims = var_10370, x = zero_mean_sq_91_cast)[name = tensor<string, []>("op_13435_cast")];
            tensor<fp16, []> var_13436_to_fp16 = const()[name = tensor<string, []>("op_13436_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_13437_cast = add(x = var_13435_cast, y = var_13436_to_fp16)[name = tensor<string, []>("op_13437_cast")];
            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_13437_cast)[name = tensor<string, []>("denom_91_cast")];
            tensor<fp16, [2, 320, 1, 4096]> out_91_cast = mul(x = zero_mean_91_cast, y = denom_91_cast)[name = tensor<string, []>("out_91_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_13441_to_fp16 = const()[name = tensor<string, []>("op_13441_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643317440)))];
            tensor<fp16, [2, 320, 1, 4096]> var_13442_cast = add(x = out_91_cast, y = var_13441_to_fp16)[name = tensor<string, []>("op_13442_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_13444_to_fp16 = const()[name = tensor<string, []>("op_13444_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643318144)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_327_cast = mul(x = var_13442_cast, y = var_13444_to_fp16)[name = tensor<string, []>("hidden_states_327_cast")];
            tensor<int32, [2]> var_13451 = const()[name = tensor<string, []>("op_13451"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13453 = const()[name = tensor<string, []>("op_13453"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643318848))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643395712))), name = tensor<string, []>("up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            tensor<fp16, [2, 320, 1, 4096]> q_61_cast = conv(dilations = var_13453, groups = var_10375, pad = q_61_pad_0, pad_type = q_61_pad_type_0, strides = var_13451, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_327_cast)[name = tensor<string, []>("q_61_cast")];
            tensor<int32, [2]> var_13457 = const()[name = tensor<string, []>("op_13457"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13459 = const()[name = tensor<string, []>("op_13459"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643395904))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643472768))), name = tensor<string, []>("up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            tensor<fp16, [2, 320, 1, 4096]> k_121_cast = conv(dilations = var_13459, groups = var_10375, pad = k_121_pad_0, pad_type = k_121_pad_type_0, strides = var_13457, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_327_cast)[name = tensor<string, []>("k_121_cast")];
            tensor<int32, [2]> var_13463 = const()[name = tensor<string, []>("op_13463"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13465 = const()[name = tensor<string, []>("op_13465"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643472960))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643549824))), name = tensor<string, []>("up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            tensor<fp16, [2, 320, 1, 4096]> v_61_cast = conv(dilations = var_13465, groups = var_10375, pad = v_61_pad_0, pad_type = v_61_pad_type_0, strides = var_13463, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_327_cast)[name = tensor<string, []>("v_61_cast")];
            tensor<int32, [4]> var_13469_begin_0 = const()[name = tensor<string, []>("op_13469_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13469_end_0 = const()[name = tensor<string, []>("op_13469_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_13469_end_mask_0 = const()[name = tensor<string, []>("op_13469_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_13469_cast = slice_by_index(begin = var_13469_begin_0, end = var_13469_end_0, end_mask = var_13469_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_13469_cast")];
            tensor<int32, [4]> var_13473_begin_0 = const()[name = tensor<string, []>("op_13473_begin_0"), val = tensor<int32, [4]>([0, 40, 0, 0])];
            tensor<int32, [4]> var_13473_end_0 = const()[name = tensor<string, []>("op_13473_end_0"), val = tensor<int32, [4]>([2, 80, 1, 4096])];
            tensor<bool, [4]> var_13473_end_mask_0 = const()[name = tensor<string, []>("op_13473_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_13473_cast = slice_by_index(begin = var_13473_begin_0, end = var_13473_end_0, end_mask = var_13473_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_13473_cast")];
            tensor<int32, [4]> var_13477_begin_0 = const()[name = tensor<string, []>("op_13477_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_13477_end_0 = const()[name = tensor<string, []>("op_13477_end_0"), val = tensor<int32, [4]>([2, 120, 1, 4096])];
            tensor<bool, [4]> var_13477_end_mask_0 = const()[name = tensor<string, []>("op_13477_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_13477_cast = slice_by_index(begin = var_13477_begin_0, end = var_13477_end_0, end_mask = var_13477_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_13477_cast")];
            tensor<int32, [4]> var_13481_begin_0 = const()[name = tensor<string, []>("op_13481_begin_0"), val = tensor<int32, [4]>([0, 120, 0, 0])];
            tensor<int32, [4]> var_13481_end_0 = const()[name = tensor<string, []>("op_13481_end_0"), val = tensor<int32, [4]>([2, 160, 1, 4096])];
            tensor<bool, [4]> var_13481_end_mask_0 = const()[name = tensor<string, []>("op_13481_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_13481_cast = slice_by_index(begin = var_13481_begin_0, end = var_13481_end_0, end_mask = var_13481_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_13481_cast")];
            tensor<int32, [4]> var_13485_begin_0 = const()[name = tensor<string, []>("op_13485_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_13485_end_0 = const()[name = tensor<string, []>("op_13485_end_0"), val = tensor<int32, [4]>([2, 200, 1, 4096])];
            tensor<bool, [4]> var_13485_end_mask_0 = const()[name = tensor<string, []>("op_13485_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_13485_cast = slice_by_index(begin = var_13485_begin_0, end = var_13485_end_0, end_mask = var_13485_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_13485_cast")];
            tensor<int32, [4]> var_13489_begin_0 = const()[name = tensor<string, []>("op_13489_begin_0"), val = tensor<int32, [4]>([0, 200, 0, 0])];
            tensor<int32, [4]> var_13489_end_0 = const()[name = tensor<string, []>("op_13489_end_0"), val = tensor<int32, [4]>([2, 240, 1, 4096])];
            tensor<bool, [4]> var_13489_end_mask_0 = const()[name = tensor<string, []>("op_13489_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_13489_cast = slice_by_index(begin = var_13489_begin_0, end = var_13489_end_0, end_mask = var_13489_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_13489_cast")];
            tensor<int32, [4]> var_13493_begin_0 = const()[name = tensor<string, []>("op_13493_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_13493_end_0 = const()[name = tensor<string, []>("op_13493_end_0"), val = tensor<int32, [4]>([2, 280, 1, 4096])];
            tensor<bool, [4]> var_13493_end_mask_0 = const()[name = tensor<string, []>("op_13493_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_13493_cast = slice_by_index(begin = var_13493_begin_0, end = var_13493_end_0, end_mask = var_13493_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_13493_cast")];
            tensor<int32, [4]> var_13497_begin_0 = const()[name = tensor<string, []>("op_13497_begin_0"), val = tensor<int32, [4]>([0, 280, 0, 0])];
            tensor<int32, [4]> var_13497_end_0 = const()[name = tensor<string, []>("op_13497_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_13497_end_mask_0 = const()[name = tensor<string, []>("op_13497_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_13497_cast = slice_by_index(begin = var_13497_begin_0, end = var_13497_end_0, end_mask = var_13497_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_13497_cast")];
            tensor<int32, [4]> var_13500_begin_0 = const()[name = tensor<string, []>("op_13500_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13500_end_0 = const()[name = tensor<string, []>("op_13500_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_13500_end_mask_0 = const()[name = tensor<string, []>("op_13500_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13500_cast = slice_by_index(begin = var_13500_begin_0, end = var_13500_end_0, end_mask = var_13500_end_mask_0, x = var_13469_cast)[name = tensor<string, []>("op_13500_cast")];
            tensor<int32, [4]> var_13501_begin_0 = const()[name = tensor<string, []>("op_13501_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_13501_end_0 = const()[name = tensor<string, []>("op_13501_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_13501_end_mask_0 = const()[name = tensor<string, []>("op_13501_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13501_cast = slice_by_index(begin = var_13501_begin_0, end = var_13501_end_0, end_mask = var_13501_end_mask_0, x = var_13469_cast)[name = tensor<string, []>("op_13501_cast")];
            tensor<int32, [4]> var_13502_begin_0 = const()[name = tensor<string, []>("op_13502_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_13502_end_0 = const()[name = tensor<string, []>("op_13502_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_13502_end_mask_0 = const()[name = tensor<string, []>("op_13502_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13502_cast = slice_by_index(begin = var_13502_begin_0, end = var_13502_end_0, end_mask = var_13502_end_mask_0, x = var_13469_cast)[name = tensor<string, []>("op_13502_cast")];
            tensor<int32, [4]> var_13503_begin_0 = const()[name = tensor<string, []>("op_13503_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_13503_end_0 = const()[name = tensor<string, []>("op_13503_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_13503_end_mask_0 = const()[name = tensor<string, []>("op_13503_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13503_cast = slice_by_index(begin = var_13503_begin_0, end = var_13503_end_0, end_mask = var_13503_end_mask_0, x = var_13469_cast)[name = tensor<string, []>("op_13503_cast")];
            tensor<int32, [4]> var_13504_begin_0 = const()[name = tensor<string, []>("op_13504_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_13504_end_0 = const()[name = tensor<string, []>("op_13504_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_13504_end_mask_0 = const()[name = tensor<string, []>("op_13504_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13504_cast = slice_by_index(begin = var_13504_begin_0, end = var_13504_end_0, end_mask = var_13504_end_mask_0, x = var_13469_cast)[name = tensor<string, []>("op_13504_cast")];
            tensor<int32, [4]> var_13505_begin_0 = const()[name = tensor<string, []>("op_13505_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_13505_end_0 = const()[name = tensor<string, []>("op_13505_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_13505_end_mask_0 = const()[name = tensor<string, []>("op_13505_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13505_cast = slice_by_index(begin = var_13505_begin_0, end = var_13505_end_0, end_mask = var_13505_end_mask_0, x = var_13469_cast)[name = tensor<string, []>("op_13505_cast")];
            tensor<int32, [4]> var_13506_begin_0 = const()[name = tensor<string, []>("op_13506_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_13506_end_0 = const()[name = tensor<string, []>("op_13506_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_13506_end_mask_0 = const()[name = tensor<string, []>("op_13506_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13506_cast = slice_by_index(begin = var_13506_begin_0, end = var_13506_end_0, end_mask = var_13506_end_mask_0, x = var_13469_cast)[name = tensor<string, []>("op_13506_cast")];
            tensor<int32, [4]> var_13507_begin_0 = const()[name = tensor<string, []>("op_13507_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_13507_end_0 = const()[name = tensor<string, []>("op_13507_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_13507_end_mask_0 = const()[name = tensor<string, []>("op_13507_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13507_cast = slice_by_index(begin = var_13507_begin_0, end = var_13507_end_0, end_mask = var_13507_end_mask_0, x = var_13469_cast)[name = tensor<string, []>("op_13507_cast")];
            tensor<int32, [4]> var_13508_begin_0 = const()[name = tensor<string, []>("op_13508_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13508_end_0 = const()[name = tensor<string, []>("op_13508_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_13508_end_mask_0 = const()[name = tensor<string, []>("op_13508_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13508_cast = slice_by_index(begin = var_13508_begin_0, end = var_13508_end_0, end_mask = var_13508_end_mask_0, x = var_13473_cast)[name = tensor<string, []>("op_13508_cast")];
            tensor<int32, [4]> var_13509_begin_0 = const()[name = tensor<string, []>("op_13509_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_13509_end_0 = const()[name = tensor<string, []>("op_13509_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_13509_end_mask_0 = const()[name = tensor<string, []>("op_13509_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13509_cast = slice_by_index(begin = var_13509_begin_0, end = var_13509_end_0, end_mask = var_13509_end_mask_0, x = var_13473_cast)[name = tensor<string, []>("op_13509_cast")];
            tensor<int32, [4]> var_13510_begin_0 = const()[name = tensor<string, []>("op_13510_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_13510_end_0 = const()[name = tensor<string, []>("op_13510_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_13510_end_mask_0 = const()[name = tensor<string, []>("op_13510_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13510_cast = slice_by_index(begin = var_13510_begin_0, end = var_13510_end_0, end_mask = var_13510_end_mask_0, x = var_13473_cast)[name = tensor<string, []>("op_13510_cast")];
            tensor<int32, [4]> var_13511_begin_0 = const()[name = tensor<string, []>("op_13511_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_13511_end_0 = const()[name = tensor<string, []>("op_13511_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_13511_end_mask_0 = const()[name = tensor<string, []>("op_13511_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13511_cast = slice_by_index(begin = var_13511_begin_0, end = var_13511_end_0, end_mask = var_13511_end_mask_0, x = var_13473_cast)[name = tensor<string, []>("op_13511_cast")];
            tensor<int32, [4]> var_13512_begin_0 = const()[name = tensor<string, []>("op_13512_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_13512_end_0 = const()[name = tensor<string, []>("op_13512_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_13512_end_mask_0 = const()[name = tensor<string, []>("op_13512_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13512_cast = slice_by_index(begin = var_13512_begin_0, end = var_13512_end_0, end_mask = var_13512_end_mask_0, x = var_13473_cast)[name = tensor<string, []>("op_13512_cast")];
            tensor<int32, [4]> var_13513_begin_0 = const()[name = tensor<string, []>("op_13513_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_13513_end_0 = const()[name = tensor<string, []>("op_13513_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_13513_end_mask_0 = const()[name = tensor<string, []>("op_13513_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13513_cast = slice_by_index(begin = var_13513_begin_0, end = var_13513_end_0, end_mask = var_13513_end_mask_0, x = var_13473_cast)[name = tensor<string, []>("op_13513_cast")];
            tensor<int32, [4]> var_13514_begin_0 = const()[name = tensor<string, []>("op_13514_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_13514_end_0 = const()[name = tensor<string, []>("op_13514_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_13514_end_mask_0 = const()[name = tensor<string, []>("op_13514_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13514_cast = slice_by_index(begin = var_13514_begin_0, end = var_13514_end_0, end_mask = var_13514_end_mask_0, x = var_13473_cast)[name = tensor<string, []>("op_13514_cast")];
            tensor<int32, [4]> var_13515_begin_0 = const()[name = tensor<string, []>("op_13515_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_13515_end_0 = const()[name = tensor<string, []>("op_13515_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_13515_end_mask_0 = const()[name = tensor<string, []>("op_13515_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13515_cast = slice_by_index(begin = var_13515_begin_0, end = var_13515_end_0, end_mask = var_13515_end_mask_0, x = var_13473_cast)[name = tensor<string, []>("op_13515_cast")];
            tensor<int32, [4]> var_13516_begin_0 = const()[name = tensor<string, []>("op_13516_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13516_end_0 = const()[name = tensor<string, []>("op_13516_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_13516_end_mask_0 = const()[name = tensor<string, []>("op_13516_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13516_cast = slice_by_index(begin = var_13516_begin_0, end = var_13516_end_0, end_mask = var_13516_end_mask_0, x = var_13477_cast)[name = tensor<string, []>("op_13516_cast")];
            tensor<int32, [4]> var_13517_begin_0 = const()[name = tensor<string, []>("op_13517_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_13517_end_0 = const()[name = tensor<string, []>("op_13517_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_13517_end_mask_0 = const()[name = tensor<string, []>("op_13517_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13517_cast = slice_by_index(begin = var_13517_begin_0, end = var_13517_end_0, end_mask = var_13517_end_mask_0, x = var_13477_cast)[name = tensor<string, []>("op_13517_cast")];
            tensor<int32, [4]> var_13518_begin_0 = const()[name = tensor<string, []>("op_13518_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_13518_end_0 = const()[name = tensor<string, []>("op_13518_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_13518_end_mask_0 = const()[name = tensor<string, []>("op_13518_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13518_cast = slice_by_index(begin = var_13518_begin_0, end = var_13518_end_0, end_mask = var_13518_end_mask_0, x = var_13477_cast)[name = tensor<string, []>("op_13518_cast")];
            tensor<int32, [4]> var_13519_begin_0 = const()[name = tensor<string, []>("op_13519_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_13519_end_0 = const()[name = tensor<string, []>("op_13519_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_13519_end_mask_0 = const()[name = tensor<string, []>("op_13519_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13519_cast = slice_by_index(begin = var_13519_begin_0, end = var_13519_end_0, end_mask = var_13519_end_mask_0, x = var_13477_cast)[name = tensor<string, []>("op_13519_cast")];
            tensor<int32, [4]> var_13520_begin_0 = const()[name = tensor<string, []>("op_13520_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_13520_end_0 = const()[name = tensor<string, []>("op_13520_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_13520_end_mask_0 = const()[name = tensor<string, []>("op_13520_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13520_cast = slice_by_index(begin = var_13520_begin_0, end = var_13520_end_0, end_mask = var_13520_end_mask_0, x = var_13477_cast)[name = tensor<string, []>("op_13520_cast")];
            tensor<int32, [4]> var_13521_begin_0 = const()[name = tensor<string, []>("op_13521_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_13521_end_0 = const()[name = tensor<string, []>("op_13521_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_13521_end_mask_0 = const()[name = tensor<string, []>("op_13521_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13521_cast = slice_by_index(begin = var_13521_begin_0, end = var_13521_end_0, end_mask = var_13521_end_mask_0, x = var_13477_cast)[name = tensor<string, []>("op_13521_cast")];
            tensor<int32, [4]> var_13522_begin_0 = const()[name = tensor<string, []>("op_13522_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_13522_end_0 = const()[name = tensor<string, []>("op_13522_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_13522_end_mask_0 = const()[name = tensor<string, []>("op_13522_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13522_cast = slice_by_index(begin = var_13522_begin_0, end = var_13522_end_0, end_mask = var_13522_end_mask_0, x = var_13477_cast)[name = tensor<string, []>("op_13522_cast")];
            tensor<int32, [4]> var_13523_begin_0 = const()[name = tensor<string, []>("op_13523_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_13523_end_0 = const()[name = tensor<string, []>("op_13523_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_13523_end_mask_0 = const()[name = tensor<string, []>("op_13523_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13523_cast = slice_by_index(begin = var_13523_begin_0, end = var_13523_end_0, end_mask = var_13523_end_mask_0, x = var_13477_cast)[name = tensor<string, []>("op_13523_cast")];
            tensor<int32, [4]> var_13524_begin_0 = const()[name = tensor<string, []>("op_13524_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13524_end_0 = const()[name = tensor<string, []>("op_13524_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_13524_end_mask_0 = const()[name = tensor<string, []>("op_13524_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13524_cast = slice_by_index(begin = var_13524_begin_0, end = var_13524_end_0, end_mask = var_13524_end_mask_0, x = var_13481_cast)[name = tensor<string, []>("op_13524_cast")];
            tensor<int32, [4]> var_13525_begin_0 = const()[name = tensor<string, []>("op_13525_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_13525_end_0 = const()[name = tensor<string, []>("op_13525_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_13525_end_mask_0 = const()[name = tensor<string, []>("op_13525_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13525_cast = slice_by_index(begin = var_13525_begin_0, end = var_13525_end_0, end_mask = var_13525_end_mask_0, x = var_13481_cast)[name = tensor<string, []>("op_13525_cast")];
            tensor<int32, [4]> var_13526_begin_0 = const()[name = tensor<string, []>("op_13526_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_13526_end_0 = const()[name = tensor<string, []>("op_13526_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_13526_end_mask_0 = const()[name = tensor<string, []>("op_13526_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13526_cast = slice_by_index(begin = var_13526_begin_0, end = var_13526_end_0, end_mask = var_13526_end_mask_0, x = var_13481_cast)[name = tensor<string, []>("op_13526_cast")];
            tensor<int32, [4]> var_13527_begin_0 = const()[name = tensor<string, []>("op_13527_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_13527_end_0 = const()[name = tensor<string, []>("op_13527_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_13527_end_mask_0 = const()[name = tensor<string, []>("op_13527_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13527_cast = slice_by_index(begin = var_13527_begin_0, end = var_13527_end_0, end_mask = var_13527_end_mask_0, x = var_13481_cast)[name = tensor<string, []>("op_13527_cast")];
            tensor<int32, [4]> var_13528_begin_0 = const()[name = tensor<string, []>("op_13528_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_13528_end_0 = const()[name = tensor<string, []>("op_13528_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_13528_end_mask_0 = const()[name = tensor<string, []>("op_13528_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13528_cast = slice_by_index(begin = var_13528_begin_0, end = var_13528_end_0, end_mask = var_13528_end_mask_0, x = var_13481_cast)[name = tensor<string, []>("op_13528_cast")];
            tensor<int32, [4]> var_13529_begin_0 = const()[name = tensor<string, []>("op_13529_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_13529_end_0 = const()[name = tensor<string, []>("op_13529_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_13529_end_mask_0 = const()[name = tensor<string, []>("op_13529_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13529_cast = slice_by_index(begin = var_13529_begin_0, end = var_13529_end_0, end_mask = var_13529_end_mask_0, x = var_13481_cast)[name = tensor<string, []>("op_13529_cast")];
            tensor<int32, [4]> var_13530_begin_0 = const()[name = tensor<string, []>("op_13530_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_13530_end_0 = const()[name = tensor<string, []>("op_13530_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_13530_end_mask_0 = const()[name = tensor<string, []>("op_13530_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13530_cast = slice_by_index(begin = var_13530_begin_0, end = var_13530_end_0, end_mask = var_13530_end_mask_0, x = var_13481_cast)[name = tensor<string, []>("op_13530_cast")];
            tensor<int32, [4]> var_13531_begin_0 = const()[name = tensor<string, []>("op_13531_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_13531_end_0 = const()[name = tensor<string, []>("op_13531_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_13531_end_mask_0 = const()[name = tensor<string, []>("op_13531_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13531_cast = slice_by_index(begin = var_13531_begin_0, end = var_13531_end_0, end_mask = var_13531_end_mask_0, x = var_13481_cast)[name = tensor<string, []>("op_13531_cast")];
            tensor<int32, [4]> var_13532_begin_0 = const()[name = tensor<string, []>("op_13532_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13532_end_0 = const()[name = tensor<string, []>("op_13532_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_13532_end_mask_0 = const()[name = tensor<string, []>("op_13532_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13532_cast = slice_by_index(begin = var_13532_begin_0, end = var_13532_end_0, end_mask = var_13532_end_mask_0, x = var_13485_cast)[name = tensor<string, []>("op_13532_cast")];
            tensor<int32, [4]> var_13533_begin_0 = const()[name = tensor<string, []>("op_13533_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_13533_end_0 = const()[name = tensor<string, []>("op_13533_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_13533_end_mask_0 = const()[name = tensor<string, []>("op_13533_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13533_cast = slice_by_index(begin = var_13533_begin_0, end = var_13533_end_0, end_mask = var_13533_end_mask_0, x = var_13485_cast)[name = tensor<string, []>("op_13533_cast")];
            tensor<int32, [4]> var_13534_begin_0 = const()[name = tensor<string, []>("op_13534_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_13534_end_0 = const()[name = tensor<string, []>("op_13534_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_13534_end_mask_0 = const()[name = tensor<string, []>("op_13534_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13534_cast = slice_by_index(begin = var_13534_begin_0, end = var_13534_end_0, end_mask = var_13534_end_mask_0, x = var_13485_cast)[name = tensor<string, []>("op_13534_cast")];
            tensor<int32, [4]> var_13535_begin_0 = const()[name = tensor<string, []>("op_13535_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_13535_end_0 = const()[name = tensor<string, []>("op_13535_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_13535_end_mask_0 = const()[name = tensor<string, []>("op_13535_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13535_cast = slice_by_index(begin = var_13535_begin_0, end = var_13535_end_0, end_mask = var_13535_end_mask_0, x = var_13485_cast)[name = tensor<string, []>("op_13535_cast")];
            tensor<int32, [4]> var_13536_begin_0 = const()[name = tensor<string, []>("op_13536_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_13536_end_0 = const()[name = tensor<string, []>("op_13536_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_13536_end_mask_0 = const()[name = tensor<string, []>("op_13536_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13536_cast = slice_by_index(begin = var_13536_begin_0, end = var_13536_end_0, end_mask = var_13536_end_mask_0, x = var_13485_cast)[name = tensor<string, []>("op_13536_cast")];
            tensor<int32, [4]> var_13537_begin_0 = const()[name = tensor<string, []>("op_13537_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_13537_end_0 = const()[name = tensor<string, []>("op_13537_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_13537_end_mask_0 = const()[name = tensor<string, []>("op_13537_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13537_cast = slice_by_index(begin = var_13537_begin_0, end = var_13537_end_0, end_mask = var_13537_end_mask_0, x = var_13485_cast)[name = tensor<string, []>("op_13537_cast")];
            tensor<int32, [4]> var_13538_begin_0 = const()[name = tensor<string, []>("op_13538_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_13538_end_0 = const()[name = tensor<string, []>("op_13538_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_13538_end_mask_0 = const()[name = tensor<string, []>("op_13538_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13538_cast = slice_by_index(begin = var_13538_begin_0, end = var_13538_end_0, end_mask = var_13538_end_mask_0, x = var_13485_cast)[name = tensor<string, []>("op_13538_cast")];
            tensor<int32, [4]> var_13539_begin_0 = const()[name = tensor<string, []>("op_13539_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_13539_end_0 = const()[name = tensor<string, []>("op_13539_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_13539_end_mask_0 = const()[name = tensor<string, []>("op_13539_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13539_cast = slice_by_index(begin = var_13539_begin_0, end = var_13539_end_0, end_mask = var_13539_end_mask_0, x = var_13485_cast)[name = tensor<string, []>("op_13539_cast")];
            tensor<int32, [4]> var_13540_begin_0 = const()[name = tensor<string, []>("op_13540_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13540_end_0 = const()[name = tensor<string, []>("op_13540_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_13540_end_mask_0 = const()[name = tensor<string, []>("op_13540_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13540_cast = slice_by_index(begin = var_13540_begin_0, end = var_13540_end_0, end_mask = var_13540_end_mask_0, x = var_13489_cast)[name = tensor<string, []>("op_13540_cast")];
            tensor<int32, [4]> var_13541_begin_0 = const()[name = tensor<string, []>("op_13541_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_13541_end_0 = const()[name = tensor<string, []>("op_13541_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_13541_end_mask_0 = const()[name = tensor<string, []>("op_13541_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13541_cast = slice_by_index(begin = var_13541_begin_0, end = var_13541_end_0, end_mask = var_13541_end_mask_0, x = var_13489_cast)[name = tensor<string, []>("op_13541_cast")];
            tensor<int32, [4]> var_13542_begin_0 = const()[name = tensor<string, []>("op_13542_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_13542_end_0 = const()[name = tensor<string, []>("op_13542_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_13542_end_mask_0 = const()[name = tensor<string, []>("op_13542_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13542_cast = slice_by_index(begin = var_13542_begin_0, end = var_13542_end_0, end_mask = var_13542_end_mask_0, x = var_13489_cast)[name = tensor<string, []>("op_13542_cast")];
            tensor<int32, [4]> var_13543_begin_0 = const()[name = tensor<string, []>("op_13543_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_13543_end_0 = const()[name = tensor<string, []>("op_13543_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_13543_end_mask_0 = const()[name = tensor<string, []>("op_13543_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13543_cast = slice_by_index(begin = var_13543_begin_0, end = var_13543_end_0, end_mask = var_13543_end_mask_0, x = var_13489_cast)[name = tensor<string, []>("op_13543_cast")];
            tensor<int32, [4]> var_13544_begin_0 = const()[name = tensor<string, []>("op_13544_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_13544_end_0 = const()[name = tensor<string, []>("op_13544_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_13544_end_mask_0 = const()[name = tensor<string, []>("op_13544_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13544_cast = slice_by_index(begin = var_13544_begin_0, end = var_13544_end_0, end_mask = var_13544_end_mask_0, x = var_13489_cast)[name = tensor<string, []>("op_13544_cast")];
            tensor<int32, [4]> var_13545_begin_0 = const()[name = tensor<string, []>("op_13545_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_13545_end_0 = const()[name = tensor<string, []>("op_13545_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_13545_end_mask_0 = const()[name = tensor<string, []>("op_13545_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13545_cast = slice_by_index(begin = var_13545_begin_0, end = var_13545_end_0, end_mask = var_13545_end_mask_0, x = var_13489_cast)[name = tensor<string, []>("op_13545_cast")];
            tensor<int32, [4]> var_13546_begin_0 = const()[name = tensor<string, []>("op_13546_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_13546_end_0 = const()[name = tensor<string, []>("op_13546_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_13546_end_mask_0 = const()[name = tensor<string, []>("op_13546_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13546_cast = slice_by_index(begin = var_13546_begin_0, end = var_13546_end_0, end_mask = var_13546_end_mask_0, x = var_13489_cast)[name = tensor<string, []>("op_13546_cast")];
            tensor<int32, [4]> var_13547_begin_0 = const()[name = tensor<string, []>("op_13547_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_13547_end_0 = const()[name = tensor<string, []>("op_13547_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_13547_end_mask_0 = const()[name = tensor<string, []>("op_13547_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13547_cast = slice_by_index(begin = var_13547_begin_0, end = var_13547_end_0, end_mask = var_13547_end_mask_0, x = var_13489_cast)[name = tensor<string, []>("op_13547_cast")];
            tensor<int32, [4]> var_13548_begin_0 = const()[name = tensor<string, []>("op_13548_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13548_end_0 = const()[name = tensor<string, []>("op_13548_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_13548_end_mask_0 = const()[name = tensor<string, []>("op_13548_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13548_cast = slice_by_index(begin = var_13548_begin_0, end = var_13548_end_0, end_mask = var_13548_end_mask_0, x = var_13493_cast)[name = tensor<string, []>("op_13548_cast")];
            tensor<int32, [4]> var_13549_begin_0 = const()[name = tensor<string, []>("op_13549_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_13549_end_0 = const()[name = tensor<string, []>("op_13549_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_13549_end_mask_0 = const()[name = tensor<string, []>("op_13549_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13549_cast = slice_by_index(begin = var_13549_begin_0, end = var_13549_end_0, end_mask = var_13549_end_mask_0, x = var_13493_cast)[name = tensor<string, []>("op_13549_cast")];
            tensor<int32, [4]> var_13550_begin_0 = const()[name = tensor<string, []>("op_13550_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_13550_end_0 = const()[name = tensor<string, []>("op_13550_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_13550_end_mask_0 = const()[name = tensor<string, []>("op_13550_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13550_cast = slice_by_index(begin = var_13550_begin_0, end = var_13550_end_0, end_mask = var_13550_end_mask_0, x = var_13493_cast)[name = tensor<string, []>("op_13550_cast")];
            tensor<int32, [4]> var_13551_begin_0 = const()[name = tensor<string, []>("op_13551_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_13551_end_0 = const()[name = tensor<string, []>("op_13551_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_13551_end_mask_0 = const()[name = tensor<string, []>("op_13551_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13551_cast = slice_by_index(begin = var_13551_begin_0, end = var_13551_end_0, end_mask = var_13551_end_mask_0, x = var_13493_cast)[name = tensor<string, []>("op_13551_cast")];
            tensor<int32, [4]> var_13552_begin_0 = const()[name = tensor<string, []>("op_13552_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_13552_end_0 = const()[name = tensor<string, []>("op_13552_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_13552_end_mask_0 = const()[name = tensor<string, []>("op_13552_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13552_cast = slice_by_index(begin = var_13552_begin_0, end = var_13552_end_0, end_mask = var_13552_end_mask_0, x = var_13493_cast)[name = tensor<string, []>("op_13552_cast")];
            tensor<int32, [4]> var_13553_begin_0 = const()[name = tensor<string, []>("op_13553_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_13553_end_0 = const()[name = tensor<string, []>("op_13553_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_13553_end_mask_0 = const()[name = tensor<string, []>("op_13553_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13553_cast = slice_by_index(begin = var_13553_begin_0, end = var_13553_end_0, end_mask = var_13553_end_mask_0, x = var_13493_cast)[name = tensor<string, []>("op_13553_cast")];
            tensor<int32, [4]> var_13554_begin_0 = const()[name = tensor<string, []>("op_13554_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_13554_end_0 = const()[name = tensor<string, []>("op_13554_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_13554_end_mask_0 = const()[name = tensor<string, []>("op_13554_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13554_cast = slice_by_index(begin = var_13554_begin_0, end = var_13554_end_0, end_mask = var_13554_end_mask_0, x = var_13493_cast)[name = tensor<string, []>("op_13554_cast")];
            tensor<int32, [4]> var_13555_begin_0 = const()[name = tensor<string, []>("op_13555_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_13555_end_0 = const()[name = tensor<string, []>("op_13555_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_13555_end_mask_0 = const()[name = tensor<string, []>("op_13555_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13555_cast = slice_by_index(begin = var_13555_begin_0, end = var_13555_end_0, end_mask = var_13555_end_mask_0, x = var_13493_cast)[name = tensor<string, []>("op_13555_cast")];
            tensor<int32, [4]> var_13556_begin_0 = const()[name = tensor<string, []>("op_13556_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13556_end_0 = const()[name = tensor<string, []>("op_13556_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_13556_end_mask_0 = const()[name = tensor<string, []>("op_13556_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13556_cast = slice_by_index(begin = var_13556_begin_0, end = var_13556_end_0, end_mask = var_13556_end_mask_0, x = var_13497_cast)[name = tensor<string, []>("op_13556_cast")];
            tensor<int32, [4]> var_13557_begin_0 = const()[name = tensor<string, []>("op_13557_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_13557_end_0 = const()[name = tensor<string, []>("op_13557_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_13557_end_mask_0 = const()[name = tensor<string, []>("op_13557_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13557_cast = slice_by_index(begin = var_13557_begin_0, end = var_13557_end_0, end_mask = var_13557_end_mask_0, x = var_13497_cast)[name = tensor<string, []>("op_13557_cast")];
            tensor<int32, [4]> var_13558_begin_0 = const()[name = tensor<string, []>("op_13558_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_13558_end_0 = const()[name = tensor<string, []>("op_13558_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_13558_end_mask_0 = const()[name = tensor<string, []>("op_13558_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13558_cast = slice_by_index(begin = var_13558_begin_0, end = var_13558_end_0, end_mask = var_13558_end_mask_0, x = var_13497_cast)[name = tensor<string, []>("op_13558_cast")];
            tensor<int32, [4]> var_13559_begin_0 = const()[name = tensor<string, []>("op_13559_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_13559_end_0 = const()[name = tensor<string, []>("op_13559_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_13559_end_mask_0 = const()[name = tensor<string, []>("op_13559_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13559_cast = slice_by_index(begin = var_13559_begin_0, end = var_13559_end_0, end_mask = var_13559_end_mask_0, x = var_13497_cast)[name = tensor<string, []>("op_13559_cast")];
            tensor<int32, [4]> var_13560_begin_0 = const()[name = tensor<string, []>("op_13560_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_13560_end_0 = const()[name = tensor<string, []>("op_13560_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_13560_end_mask_0 = const()[name = tensor<string, []>("op_13560_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13560_cast = slice_by_index(begin = var_13560_begin_0, end = var_13560_end_0, end_mask = var_13560_end_mask_0, x = var_13497_cast)[name = tensor<string, []>("op_13560_cast")];
            tensor<int32, [4]> var_13561_begin_0 = const()[name = tensor<string, []>("op_13561_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_13561_end_0 = const()[name = tensor<string, []>("op_13561_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_13561_end_mask_0 = const()[name = tensor<string, []>("op_13561_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13561_cast = slice_by_index(begin = var_13561_begin_0, end = var_13561_end_0, end_mask = var_13561_end_mask_0, x = var_13497_cast)[name = tensor<string, []>("op_13561_cast")];
            tensor<int32, [4]> var_13562_begin_0 = const()[name = tensor<string, []>("op_13562_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_13562_end_0 = const()[name = tensor<string, []>("op_13562_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_13562_end_mask_0 = const()[name = tensor<string, []>("op_13562_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13562_cast = slice_by_index(begin = var_13562_begin_0, end = var_13562_end_0, end_mask = var_13562_end_mask_0, x = var_13497_cast)[name = tensor<string, []>("op_13562_cast")];
            tensor<int32, [4]> var_13563_begin_0 = const()[name = tensor<string, []>("op_13563_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_13563_end_0 = const()[name = tensor<string, []>("op_13563_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_13563_end_mask_0 = const()[name = tensor<string, []>("op_13563_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_13563_cast = slice_by_index(begin = var_13563_begin_0, end = var_13563_end_0, end_mask = var_13563_end_mask_0, x = var_13497_cast)[name = tensor<string, []>("op_13563_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_13568_begin_0 = const()[name = tensor<string, []>("op_13568_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13568_end_0 = const()[name = tensor<string, []>("op_13568_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 40])];
            tensor<bool, [4]> var_13568_end_mask_0 = const()[name = tensor<string, []>("op_13568_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)[name = tensor<string, []>("transpose_1")];
            tensor<fp16, [2, 4096, 1, 40]> var_13568_cast = slice_by_index(begin = var_13568_begin_0, end = var_13568_end_0, end_mask = var_13568_end_mask_0, x = transpose_1)[name = tensor<string, []>("op_13568_cast")];
            tensor<int32, [4]> var_13572_begin_0 = const()[name = tensor<string, []>("op_13572_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 40])];
            tensor<int32, [4]> var_13572_end_0 = const()[name = tensor<string, []>("op_13572_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 80])];
            tensor<bool, [4]> var_13572_end_mask_0 = const()[name = tensor<string, []>("op_13572_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_13572_cast = slice_by_index(begin = var_13572_begin_0, end = var_13572_end_0, end_mask = var_13572_end_mask_0, x = transpose_1)[name = tensor<string, []>("op_13572_cast")];
            tensor<int32, [4]> var_13576_begin_0 = const()[name = tensor<string, []>("op_13576_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 80])];
            tensor<int32, [4]> var_13576_end_0 = const()[name = tensor<string, []>("op_13576_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 120])];
            tensor<bool, [4]> var_13576_end_mask_0 = const()[name = tensor<string, []>("op_13576_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_13576_cast = slice_by_index(begin = var_13576_begin_0, end = var_13576_end_0, end_mask = var_13576_end_mask_0, x = transpose_1)[name = tensor<string, []>("op_13576_cast")];
            tensor<int32, [4]> var_13580_begin_0 = const()[name = tensor<string, []>("op_13580_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 120])];
            tensor<int32, [4]> var_13580_end_0 = const()[name = tensor<string, []>("op_13580_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 160])];
            tensor<bool, [4]> var_13580_end_mask_0 = const()[name = tensor<string, []>("op_13580_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_13580_cast = slice_by_index(begin = var_13580_begin_0, end = var_13580_end_0, end_mask = var_13580_end_mask_0, x = transpose_1)[name = tensor<string, []>("op_13580_cast")];
            tensor<int32, [4]> var_13584_begin_0 = const()[name = tensor<string, []>("op_13584_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_13584_end_0 = const()[name = tensor<string, []>("op_13584_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 200])];
            tensor<bool, [4]> var_13584_end_mask_0 = const()[name = tensor<string, []>("op_13584_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_13584_cast = slice_by_index(begin = var_13584_begin_0, end = var_13584_end_0, end_mask = var_13584_end_mask_0, x = transpose_1)[name = tensor<string, []>("op_13584_cast")];
            tensor<int32, [4]> var_13588_begin_0 = const()[name = tensor<string, []>("op_13588_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 200])];
            tensor<int32, [4]> var_13588_end_0 = const()[name = tensor<string, []>("op_13588_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 240])];
            tensor<bool, [4]> var_13588_end_mask_0 = const()[name = tensor<string, []>("op_13588_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_13588_cast = slice_by_index(begin = var_13588_begin_0, end = var_13588_end_0, end_mask = var_13588_end_mask_0, x = transpose_1)[name = tensor<string, []>("op_13588_cast")];
            tensor<int32, [4]> var_13592_begin_0 = const()[name = tensor<string, []>("op_13592_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 240])];
            tensor<int32, [4]> var_13592_end_0 = const()[name = tensor<string, []>("op_13592_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 280])];
            tensor<bool, [4]> var_13592_end_mask_0 = const()[name = tensor<string, []>("op_13592_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_13592_cast = slice_by_index(begin = var_13592_begin_0, end = var_13592_end_0, end_mask = var_13592_end_mask_0, x = transpose_1)[name = tensor<string, []>("op_13592_cast")];
            tensor<int32, [4]> var_13596_begin_0 = const()[name = tensor<string, []>("op_13596_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 280])];
            tensor<int32, [4]> var_13596_end_0 = const()[name = tensor<string, []>("op_13596_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_13596_end_mask_0 = const()[name = tensor<string, []>("op_13596_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 40]> var_13596_cast = slice_by_index(begin = var_13596_begin_0, end = var_13596_end_0, end_mask = var_13596_end_mask_0, x = transpose_1)[name = tensor<string, []>("op_13596_cast")];
            tensor<int32, [4]> var_13598_begin_0 = const()[name = tensor<string, []>("op_13598_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13598_end_0 = const()[name = tensor<string, []>("op_13598_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_13598_end_mask_0 = const()[name = tensor<string, []>("op_13598_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_13598_cast = slice_by_index(begin = var_13598_begin_0, end = var_13598_end_0, end_mask = var_13598_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_13598_cast")];
            tensor<int32, [4]> var_13602_begin_0 = const()[name = tensor<string, []>("op_13602_begin_0"), val = tensor<int32, [4]>([0, 40, 0, 0])];
            tensor<int32, [4]> var_13602_end_0 = const()[name = tensor<string, []>("op_13602_end_0"), val = tensor<int32, [4]>([2, 80, 1, 4096])];
            tensor<bool, [4]> var_13602_end_mask_0 = const()[name = tensor<string, []>("op_13602_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_13602_cast = slice_by_index(begin = var_13602_begin_0, end = var_13602_end_0, end_mask = var_13602_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_13602_cast")];
            tensor<int32, [4]> var_13606_begin_0 = const()[name = tensor<string, []>("op_13606_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_13606_end_0 = const()[name = tensor<string, []>("op_13606_end_0"), val = tensor<int32, [4]>([2, 120, 1, 4096])];
            tensor<bool, [4]> var_13606_end_mask_0 = const()[name = tensor<string, []>("op_13606_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_13606_cast = slice_by_index(begin = var_13606_begin_0, end = var_13606_end_0, end_mask = var_13606_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_13606_cast")];
            tensor<int32, [4]> var_13610_begin_0 = const()[name = tensor<string, []>("op_13610_begin_0"), val = tensor<int32, [4]>([0, 120, 0, 0])];
            tensor<int32, [4]> var_13610_end_0 = const()[name = tensor<string, []>("op_13610_end_0"), val = tensor<int32, [4]>([2, 160, 1, 4096])];
            tensor<bool, [4]> var_13610_end_mask_0 = const()[name = tensor<string, []>("op_13610_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_13610_cast = slice_by_index(begin = var_13610_begin_0, end = var_13610_end_0, end_mask = var_13610_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_13610_cast")];
            tensor<int32, [4]> var_13614_begin_0 = const()[name = tensor<string, []>("op_13614_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_13614_end_0 = const()[name = tensor<string, []>("op_13614_end_0"), val = tensor<int32, [4]>([2, 200, 1, 4096])];
            tensor<bool, [4]> var_13614_end_mask_0 = const()[name = tensor<string, []>("op_13614_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_13614_cast = slice_by_index(begin = var_13614_begin_0, end = var_13614_end_0, end_mask = var_13614_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_13614_cast")];
            tensor<int32, [4]> var_13618_begin_0 = const()[name = tensor<string, []>("op_13618_begin_0"), val = tensor<int32, [4]>([0, 200, 0, 0])];
            tensor<int32, [4]> var_13618_end_0 = const()[name = tensor<string, []>("op_13618_end_0"), val = tensor<int32, [4]>([2, 240, 1, 4096])];
            tensor<bool, [4]> var_13618_end_mask_0 = const()[name = tensor<string, []>("op_13618_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_13618_cast = slice_by_index(begin = var_13618_begin_0, end = var_13618_end_0, end_mask = var_13618_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_13618_cast")];
            tensor<int32, [4]> var_13622_begin_0 = const()[name = tensor<string, []>("op_13622_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_13622_end_0 = const()[name = tensor<string, []>("op_13622_end_0"), val = tensor<int32, [4]>([2, 280, 1, 4096])];
            tensor<bool, [4]> var_13622_end_mask_0 = const()[name = tensor<string, []>("op_13622_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_13622_cast = slice_by_index(begin = var_13622_begin_0, end = var_13622_end_0, end_mask = var_13622_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_13622_cast")];
            tensor<int32, [4]> var_13626_begin_0 = const()[name = tensor<string, []>("op_13626_begin_0"), val = tensor<int32, [4]>([0, 280, 0, 0])];
            tensor<int32, [4]> var_13626_end_0 = const()[name = tensor<string, []>("op_13626_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_13626_end_mask_0 = const()[name = tensor<string, []>("op_13626_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_13626_cast = slice_by_index(begin = var_13626_begin_0, end = var_13626_end_0, end_mask = var_13626_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_13626_cast")];
            tensor<string, []> var_13630_equation_0 = const()[name = tensor<string, []>("op_13630_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13630_cast = einsum(equation = var_13630_equation_0, values = (var_13568_cast, var_13500_cast))[name = tensor<string, []>("op_13630_cast")];
            tensor<fp16, []> var_13631_to_fp16 = const()[name = tensor<string, []>("op_13631_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1345_cast = mul(x = var_13630_cast, y = var_13631_to_fp16)[name = tensor<string, []>("aw_chunk_1345_cast")];
            tensor<string, []> var_13634_equation_0 = const()[name = tensor<string, []>("op_13634_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13634_cast = einsum(equation = var_13634_equation_0, values = (var_13568_cast, var_13501_cast))[name = tensor<string, []>("op_13634_cast")];
            tensor<fp16, []> var_13635_to_fp16 = const()[name = tensor<string, []>("op_13635_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1347_cast = mul(x = var_13634_cast, y = var_13635_to_fp16)[name = tensor<string, []>("aw_chunk_1347_cast")];
            tensor<string, []> var_13638_equation_0 = const()[name = tensor<string, []>("op_13638_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13638_cast = einsum(equation = var_13638_equation_0, values = (var_13568_cast, var_13502_cast))[name = tensor<string, []>("op_13638_cast")];
            tensor<fp16, []> var_13639_to_fp16 = const()[name = tensor<string, []>("op_13639_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1349_cast = mul(x = var_13638_cast, y = var_13639_to_fp16)[name = tensor<string, []>("aw_chunk_1349_cast")];
            tensor<string, []> var_13642_equation_0 = const()[name = tensor<string, []>("op_13642_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13642_cast = einsum(equation = var_13642_equation_0, values = (var_13568_cast, var_13503_cast))[name = tensor<string, []>("op_13642_cast")];
            tensor<fp16, []> var_13643_to_fp16 = const()[name = tensor<string, []>("op_13643_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1351_cast = mul(x = var_13642_cast, y = var_13643_to_fp16)[name = tensor<string, []>("aw_chunk_1351_cast")];
            tensor<string, []> var_13646_equation_0 = const()[name = tensor<string, []>("op_13646_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13646_cast = einsum(equation = var_13646_equation_0, values = (var_13568_cast, var_13504_cast))[name = tensor<string, []>("op_13646_cast")];
            tensor<fp16, []> var_13647_to_fp16 = const()[name = tensor<string, []>("op_13647_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1353_cast = mul(x = var_13646_cast, y = var_13647_to_fp16)[name = tensor<string, []>("aw_chunk_1353_cast")];
            tensor<string, []> var_13650_equation_0 = const()[name = tensor<string, []>("op_13650_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13650_cast = einsum(equation = var_13650_equation_0, values = (var_13568_cast, var_13505_cast))[name = tensor<string, []>("op_13650_cast")];
            tensor<fp16, []> var_13651_to_fp16 = const()[name = tensor<string, []>("op_13651_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1355_cast = mul(x = var_13650_cast, y = var_13651_to_fp16)[name = tensor<string, []>("aw_chunk_1355_cast")];
            tensor<string, []> var_13654_equation_0 = const()[name = tensor<string, []>("op_13654_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13654_cast = einsum(equation = var_13654_equation_0, values = (var_13568_cast, var_13506_cast))[name = tensor<string, []>("op_13654_cast")];
            tensor<fp16, []> var_13655_to_fp16 = const()[name = tensor<string, []>("op_13655_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1357_cast = mul(x = var_13654_cast, y = var_13655_to_fp16)[name = tensor<string, []>("aw_chunk_1357_cast")];
            tensor<string, []> var_13658_equation_0 = const()[name = tensor<string, []>("op_13658_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13658_cast = einsum(equation = var_13658_equation_0, values = (var_13568_cast, var_13507_cast))[name = tensor<string, []>("op_13658_cast")];
            tensor<fp16, []> var_13659_to_fp16 = const()[name = tensor<string, []>("op_13659_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1359_cast = mul(x = var_13658_cast, y = var_13659_to_fp16)[name = tensor<string, []>("aw_chunk_1359_cast")];
            tensor<string, []> var_13662_equation_0 = const()[name = tensor<string, []>("op_13662_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13662_cast = einsum(equation = var_13662_equation_0, values = (var_13572_cast, var_13508_cast))[name = tensor<string, []>("op_13662_cast")];
            tensor<fp16, []> var_13663_to_fp16 = const()[name = tensor<string, []>("op_13663_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1361_cast = mul(x = var_13662_cast, y = var_13663_to_fp16)[name = tensor<string, []>("aw_chunk_1361_cast")];
            tensor<string, []> var_13666_equation_0 = const()[name = tensor<string, []>("op_13666_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13666_cast = einsum(equation = var_13666_equation_0, values = (var_13572_cast, var_13509_cast))[name = tensor<string, []>("op_13666_cast")];
            tensor<fp16, []> var_13667_to_fp16 = const()[name = tensor<string, []>("op_13667_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1363_cast = mul(x = var_13666_cast, y = var_13667_to_fp16)[name = tensor<string, []>("aw_chunk_1363_cast")];
            tensor<string, []> var_13670_equation_0 = const()[name = tensor<string, []>("op_13670_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13670_cast = einsum(equation = var_13670_equation_0, values = (var_13572_cast, var_13510_cast))[name = tensor<string, []>("op_13670_cast")];
            tensor<fp16, []> var_13671_to_fp16 = const()[name = tensor<string, []>("op_13671_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1365_cast = mul(x = var_13670_cast, y = var_13671_to_fp16)[name = tensor<string, []>("aw_chunk_1365_cast")];
            tensor<string, []> var_13674_equation_0 = const()[name = tensor<string, []>("op_13674_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13674_cast = einsum(equation = var_13674_equation_0, values = (var_13572_cast, var_13511_cast))[name = tensor<string, []>("op_13674_cast")];
            tensor<fp16, []> var_13675_to_fp16 = const()[name = tensor<string, []>("op_13675_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1367_cast = mul(x = var_13674_cast, y = var_13675_to_fp16)[name = tensor<string, []>("aw_chunk_1367_cast")];
            tensor<string, []> var_13678_equation_0 = const()[name = tensor<string, []>("op_13678_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13678_cast = einsum(equation = var_13678_equation_0, values = (var_13572_cast, var_13512_cast))[name = tensor<string, []>("op_13678_cast")];
            tensor<fp16, []> var_13679_to_fp16 = const()[name = tensor<string, []>("op_13679_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1369_cast = mul(x = var_13678_cast, y = var_13679_to_fp16)[name = tensor<string, []>("aw_chunk_1369_cast")];
            tensor<string, []> var_13682_equation_0 = const()[name = tensor<string, []>("op_13682_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13682_cast = einsum(equation = var_13682_equation_0, values = (var_13572_cast, var_13513_cast))[name = tensor<string, []>("op_13682_cast")];
            tensor<fp16, []> var_13683_to_fp16 = const()[name = tensor<string, []>("op_13683_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1371_cast = mul(x = var_13682_cast, y = var_13683_to_fp16)[name = tensor<string, []>("aw_chunk_1371_cast")];
            tensor<string, []> var_13686_equation_0 = const()[name = tensor<string, []>("op_13686_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13686_cast = einsum(equation = var_13686_equation_0, values = (var_13572_cast, var_13514_cast))[name = tensor<string, []>("op_13686_cast")];
            tensor<fp16, []> var_13687_to_fp16 = const()[name = tensor<string, []>("op_13687_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1373_cast = mul(x = var_13686_cast, y = var_13687_to_fp16)[name = tensor<string, []>("aw_chunk_1373_cast")];
            tensor<string, []> var_13690_equation_0 = const()[name = tensor<string, []>("op_13690_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13690_cast = einsum(equation = var_13690_equation_0, values = (var_13572_cast, var_13515_cast))[name = tensor<string, []>("op_13690_cast")];
            tensor<fp16, []> var_13691_to_fp16 = const()[name = tensor<string, []>("op_13691_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1375_cast = mul(x = var_13690_cast, y = var_13691_to_fp16)[name = tensor<string, []>("aw_chunk_1375_cast")];
            tensor<string, []> var_13694_equation_0 = const()[name = tensor<string, []>("op_13694_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13694_cast = einsum(equation = var_13694_equation_0, values = (var_13576_cast, var_13516_cast))[name = tensor<string, []>("op_13694_cast")];
            tensor<fp16, []> var_13695_to_fp16 = const()[name = tensor<string, []>("op_13695_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1377_cast = mul(x = var_13694_cast, y = var_13695_to_fp16)[name = tensor<string, []>("aw_chunk_1377_cast")];
            tensor<string, []> var_13698_equation_0 = const()[name = tensor<string, []>("op_13698_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13698_cast = einsum(equation = var_13698_equation_0, values = (var_13576_cast, var_13517_cast))[name = tensor<string, []>("op_13698_cast")];
            tensor<fp16, []> var_13699_to_fp16 = const()[name = tensor<string, []>("op_13699_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1379_cast = mul(x = var_13698_cast, y = var_13699_to_fp16)[name = tensor<string, []>("aw_chunk_1379_cast")];
            tensor<string, []> var_13702_equation_0 = const()[name = tensor<string, []>("op_13702_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13702_cast = einsum(equation = var_13702_equation_0, values = (var_13576_cast, var_13518_cast))[name = tensor<string, []>("op_13702_cast")];
            tensor<fp16, []> var_13703_to_fp16 = const()[name = tensor<string, []>("op_13703_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1381_cast = mul(x = var_13702_cast, y = var_13703_to_fp16)[name = tensor<string, []>("aw_chunk_1381_cast")];
            tensor<string, []> var_13706_equation_0 = const()[name = tensor<string, []>("op_13706_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13706_cast = einsum(equation = var_13706_equation_0, values = (var_13576_cast, var_13519_cast))[name = tensor<string, []>("op_13706_cast")];
            tensor<fp16, []> var_13707_to_fp16 = const()[name = tensor<string, []>("op_13707_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1383_cast = mul(x = var_13706_cast, y = var_13707_to_fp16)[name = tensor<string, []>("aw_chunk_1383_cast")];
            tensor<string, []> var_13710_equation_0 = const()[name = tensor<string, []>("op_13710_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13710_cast = einsum(equation = var_13710_equation_0, values = (var_13576_cast, var_13520_cast))[name = tensor<string, []>("op_13710_cast")];
            tensor<fp16, []> var_13711_to_fp16 = const()[name = tensor<string, []>("op_13711_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1385_cast = mul(x = var_13710_cast, y = var_13711_to_fp16)[name = tensor<string, []>("aw_chunk_1385_cast")];
            tensor<string, []> var_13714_equation_0 = const()[name = tensor<string, []>("op_13714_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13714_cast = einsum(equation = var_13714_equation_0, values = (var_13576_cast, var_13521_cast))[name = tensor<string, []>("op_13714_cast")];
            tensor<fp16, []> var_13715_to_fp16 = const()[name = tensor<string, []>("op_13715_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1387_cast = mul(x = var_13714_cast, y = var_13715_to_fp16)[name = tensor<string, []>("aw_chunk_1387_cast")];
            tensor<string, []> var_13718_equation_0 = const()[name = tensor<string, []>("op_13718_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13718_cast = einsum(equation = var_13718_equation_0, values = (var_13576_cast, var_13522_cast))[name = tensor<string, []>("op_13718_cast")];
            tensor<fp16, []> var_13719_to_fp16 = const()[name = tensor<string, []>("op_13719_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1389_cast = mul(x = var_13718_cast, y = var_13719_to_fp16)[name = tensor<string, []>("aw_chunk_1389_cast")];
            tensor<string, []> var_13722_equation_0 = const()[name = tensor<string, []>("op_13722_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13722_cast = einsum(equation = var_13722_equation_0, values = (var_13576_cast, var_13523_cast))[name = tensor<string, []>("op_13722_cast")];
            tensor<fp16, []> var_13723_to_fp16 = const()[name = tensor<string, []>("op_13723_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1391_cast = mul(x = var_13722_cast, y = var_13723_to_fp16)[name = tensor<string, []>("aw_chunk_1391_cast")];
            tensor<string, []> var_13726_equation_0 = const()[name = tensor<string, []>("op_13726_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13726_cast = einsum(equation = var_13726_equation_0, values = (var_13580_cast, var_13524_cast))[name = tensor<string, []>("op_13726_cast")];
            tensor<fp16, []> var_13727_to_fp16 = const()[name = tensor<string, []>("op_13727_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1393_cast = mul(x = var_13726_cast, y = var_13727_to_fp16)[name = tensor<string, []>("aw_chunk_1393_cast")];
            tensor<string, []> var_13730_equation_0 = const()[name = tensor<string, []>("op_13730_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13730_cast = einsum(equation = var_13730_equation_0, values = (var_13580_cast, var_13525_cast))[name = tensor<string, []>("op_13730_cast")];
            tensor<fp16, []> var_13731_to_fp16 = const()[name = tensor<string, []>("op_13731_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1395_cast = mul(x = var_13730_cast, y = var_13731_to_fp16)[name = tensor<string, []>("aw_chunk_1395_cast")];
            tensor<string, []> var_13734_equation_0 = const()[name = tensor<string, []>("op_13734_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13734_cast = einsum(equation = var_13734_equation_0, values = (var_13580_cast, var_13526_cast))[name = tensor<string, []>("op_13734_cast")];
            tensor<fp16, []> var_13735_to_fp16 = const()[name = tensor<string, []>("op_13735_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1397_cast = mul(x = var_13734_cast, y = var_13735_to_fp16)[name = tensor<string, []>("aw_chunk_1397_cast")];
            tensor<string, []> var_13738_equation_0 = const()[name = tensor<string, []>("op_13738_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13738_cast = einsum(equation = var_13738_equation_0, values = (var_13580_cast, var_13527_cast))[name = tensor<string, []>("op_13738_cast")];
            tensor<fp16, []> var_13739_to_fp16 = const()[name = tensor<string, []>("op_13739_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1399_cast = mul(x = var_13738_cast, y = var_13739_to_fp16)[name = tensor<string, []>("aw_chunk_1399_cast")];
            tensor<string, []> var_13742_equation_0 = const()[name = tensor<string, []>("op_13742_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13742_cast = einsum(equation = var_13742_equation_0, values = (var_13580_cast, var_13528_cast))[name = tensor<string, []>("op_13742_cast")];
            tensor<fp16, []> var_13743_to_fp16 = const()[name = tensor<string, []>("op_13743_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1401_cast = mul(x = var_13742_cast, y = var_13743_to_fp16)[name = tensor<string, []>("aw_chunk_1401_cast")];
            tensor<string, []> var_13746_equation_0 = const()[name = tensor<string, []>("op_13746_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13746_cast = einsum(equation = var_13746_equation_0, values = (var_13580_cast, var_13529_cast))[name = tensor<string, []>("op_13746_cast")];
            tensor<fp16, []> var_13747_to_fp16 = const()[name = tensor<string, []>("op_13747_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1403_cast = mul(x = var_13746_cast, y = var_13747_to_fp16)[name = tensor<string, []>("aw_chunk_1403_cast")];
            tensor<string, []> var_13750_equation_0 = const()[name = tensor<string, []>("op_13750_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13750_cast = einsum(equation = var_13750_equation_0, values = (var_13580_cast, var_13530_cast))[name = tensor<string, []>("op_13750_cast")];
            tensor<fp16, []> var_13751_to_fp16 = const()[name = tensor<string, []>("op_13751_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1405_cast = mul(x = var_13750_cast, y = var_13751_to_fp16)[name = tensor<string, []>("aw_chunk_1405_cast")];
            tensor<string, []> var_13754_equation_0 = const()[name = tensor<string, []>("op_13754_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13754_cast = einsum(equation = var_13754_equation_0, values = (var_13580_cast, var_13531_cast))[name = tensor<string, []>("op_13754_cast")];
            tensor<fp16, []> var_13755_to_fp16 = const()[name = tensor<string, []>("op_13755_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1407_cast = mul(x = var_13754_cast, y = var_13755_to_fp16)[name = tensor<string, []>("aw_chunk_1407_cast")];
            tensor<string, []> var_13758_equation_0 = const()[name = tensor<string, []>("op_13758_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13758_cast = einsum(equation = var_13758_equation_0, values = (var_13584_cast, var_13532_cast))[name = tensor<string, []>("op_13758_cast")];
            tensor<fp16, []> var_13759_to_fp16 = const()[name = tensor<string, []>("op_13759_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1409_cast = mul(x = var_13758_cast, y = var_13759_to_fp16)[name = tensor<string, []>("aw_chunk_1409_cast")];
            tensor<string, []> var_13762_equation_0 = const()[name = tensor<string, []>("op_13762_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13762_cast = einsum(equation = var_13762_equation_0, values = (var_13584_cast, var_13533_cast))[name = tensor<string, []>("op_13762_cast")];
            tensor<fp16, []> var_13763_to_fp16 = const()[name = tensor<string, []>("op_13763_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1411_cast = mul(x = var_13762_cast, y = var_13763_to_fp16)[name = tensor<string, []>("aw_chunk_1411_cast")];
            tensor<string, []> var_13766_equation_0 = const()[name = tensor<string, []>("op_13766_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13766_cast = einsum(equation = var_13766_equation_0, values = (var_13584_cast, var_13534_cast))[name = tensor<string, []>("op_13766_cast")];
            tensor<fp16, []> var_13767_to_fp16 = const()[name = tensor<string, []>("op_13767_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1413_cast = mul(x = var_13766_cast, y = var_13767_to_fp16)[name = tensor<string, []>("aw_chunk_1413_cast")];
            tensor<string, []> var_13770_equation_0 = const()[name = tensor<string, []>("op_13770_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13770_cast = einsum(equation = var_13770_equation_0, values = (var_13584_cast, var_13535_cast))[name = tensor<string, []>("op_13770_cast")];
            tensor<fp16, []> var_13771_to_fp16 = const()[name = tensor<string, []>("op_13771_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1415_cast = mul(x = var_13770_cast, y = var_13771_to_fp16)[name = tensor<string, []>("aw_chunk_1415_cast")];
            tensor<string, []> var_13774_equation_0 = const()[name = tensor<string, []>("op_13774_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13774_cast = einsum(equation = var_13774_equation_0, values = (var_13584_cast, var_13536_cast))[name = tensor<string, []>("op_13774_cast")];
            tensor<fp16, []> var_13775_to_fp16 = const()[name = tensor<string, []>("op_13775_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1417_cast = mul(x = var_13774_cast, y = var_13775_to_fp16)[name = tensor<string, []>("aw_chunk_1417_cast")];
            tensor<string, []> var_13778_equation_0 = const()[name = tensor<string, []>("op_13778_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13778_cast = einsum(equation = var_13778_equation_0, values = (var_13584_cast, var_13537_cast))[name = tensor<string, []>("op_13778_cast")];
            tensor<fp16, []> var_13779_to_fp16 = const()[name = tensor<string, []>("op_13779_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1419_cast = mul(x = var_13778_cast, y = var_13779_to_fp16)[name = tensor<string, []>("aw_chunk_1419_cast")];
            tensor<string, []> var_13782_equation_0 = const()[name = tensor<string, []>("op_13782_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13782_cast = einsum(equation = var_13782_equation_0, values = (var_13584_cast, var_13538_cast))[name = tensor<string, []>("op_13782_cast")];
            tensor<fp16, []> var_13783_to_fp16 = const()[name = tensor<string, []>("op_13783_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1421_cast = mul(x = var_13782_cast, y = var_13783_to_fp16)[name = tensor<string, []>("aw_chunk_1421_cast")];
            tensor<string, []> var_13786_equation_0 = const()[name = tensor<string, []>("op_13786_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13786_cast = einsum(equation = var_13786_equation_0, values = (var_13584_cast, var_13539_cast))[name = tensor<string, []>("op_13786_cast")];
            tensor<fp16, []> var_13787_to_fp16 = const()[name = tensor<string, []>("op_13787_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1423_cast = mul(x = var_13786_cast, y = var_13787_to_fp16)[name = tensor<string, []>("aw_chunk_1423_cast")];
            tensor<string, []> var_13790_equation_0 = const()[name = tensor<string, []>("op_13790_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13790_cast = einsum(equation = var_13790_equation_0, values = (var_13588_cast, var_13540_cast))[name = tensor<string, []>("op_13790_cast")];
            tensor<fp16, []> var_13791_to_fp16 = const()[name = tensor<string, []>("op_13791_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1425_cast = mul(x = var_13790_cast, y = var_13791_to_fp16)[name = tensor<string, []>("aw_chunk_1425_cast")];
            tensor<string, []> var_13794_equation_0 = const()[name = tensor<string, []>("op_13794_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13794_cast = einsum(equation = var_13794_equation_0, values = (var_13588_cast, var_13541_cast))[name = tensor<string, []>("op_13794_cast")];
            tensor<fp16, []> var_13795_to_fp16 = const()[name = tensor<string, []>("op_13795_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1427_cast = mul(x = var_13794_cast, y = var_13795_to_fp16)[name = tensor<string, []>("aw_chunk_1427_cast")];
            tensor<string, []> var_13798_equation_0 = const()[name = tensor<string, []>("op_13798_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13798_cast = einsum(equation = var_13798_equation_0, values = (var_13588_cast, var_13542_cast))[name = tensor<string, []>("op_13798_cast")];
            tensor<fp16, []> var_13799_to_fp16 = const()[name = tensor<string, []>("op_13799_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1429_cast = mul(x = var_13798_cast, y = var_13799_to_fp16)[name = tensor<string, []>("aw_chunk_1429_cast")];
            tensor<string, []> var_13802_equation_0 = const()[name = tensor<string, []>("op_13802_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13802_cast = einsum(equation = var_13802_equation_0, values = (var_13588_cast, var_13543_cast))[name = tensor<string, []>("op_13802_cast")];
            tensor<fp16, []> var_13803_to_fp16 = const()[name = tensor<string, []>("op_13803_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1431_cast = mul(x = var_13802_cast, y = var_13803_to_fp16)[name = tensor<string, []>("aw_chunk_1431_cast")];
            tensor<string, []> var_13806_equation_0 = const()[name = tensor<string, []>("op_13806_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13806_cast = einsum(equation = var_13806_equation_0, values = (var_13588_cast, var_13544_cast))[name = tensor<string, []>("op_13806_cast")];
            tensor<fp16, []> var_13807_to_fp16 = const()[name = tensor<string, []>("op_13807_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1433_cast = mul(x = var_13806_cast, y = var_13807_to_fp16)[name = tensor<string, []>("aw_chunk_1433_cast")];
            tensor<string, []> var_13810_equation_0 = const()[name = tensor<string, []>("op_13810_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13810_cast = einsum(equation = var_13810_equation_0, values = (var_13588_cast, var_13545_cast))[name = tensor<string, []>("op_13810_cast")];
            tensor<fp16, []> var_13811_to_fp16 = const()[name = tensor<string, []>("op_13811_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1435_cast = mul(x = var_13810_cast, y = var_13811_to_fp16)[name = tensor<string, []>("aw_chunk_1435_cast")];
            tensor<string, []> var_13814_equation_0 = const()[name = tensor<string, []>("op_13814_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13814_cast = einsum(equation = var_13814_equation_0, values = (var_13588_cast, var_13546_cast))[name = tensor<string, []>("op_13814_cast")];
            tensor<fp16, []> var_13815_to_fp16 = const()[name = tensor<string, []>("op_13815_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1437_cast = mul(x = var_13814_cast, y = var_13815_to_fp16)[name = tensor<string, []>("aw_chunk_1437_cast")];
            tensor<string, []> var_13818_equation_0 = const()[name = tensor<string, []>("op_13818_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13818_cast = einsum(equation = var_13818_equation_0, values = (var_13588_cast, var_13547_cast))[name = tensor<string, []>("op_13818_cast")];
            tensor<fp16, []> var_13819_to_fp16 = const()[name = tensor<string, []>("op_13819_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1439_cast = mul(x = var_13818_cast, y = var_13819_to_fp16)[name = tensor<string, []>("aw_chunk_1439_cast")];
            tensor<string, []> var_13822_equation_0 = const()[name = tensor<string, []>("op_13822_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13822_cast = einsum(equation = var_13822_equation_0, values = (var_13592_cast, var_13548_cast))[name = tensor<string, []>("op_13822_cast")];
            tensor<fp16, []> var_13823_to_fp16 = const()[name = tensor<string, []>("op_13823_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1441_cast = mul(x = var_13822_cast, y = var_13823_to_fp16)[name = tensor<string, []>("aw_chunk_1441_cast")];
            tensor<string, []> var_13826_equation_0 = const()[name = tensor<string, []>("op_13826_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13826_cast = einsum(equation = var_13826_equation_0, values = (var_13592_cast, var_13549_cast))[name = tensor<string, []>("op_13826_cast")];
            tensor<fp16, []> var_13827_to_fp16 = const()[name = tensor<string, []>("op_13827_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1443_cast = mul(x = var_13826_cast, y = var_13827_to_fp16)[name = tensor<string, []>("aw_chunk_1443_cast")];
            tensor<string, []> var_13830_equation_0 = const()[name = tensor<string, []>("op_13830_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13830_cast = einsum(equation = var_13830_equation_0, values = (var_13592_cast, var_13550_cast))[name = tensor<string, []>("op_13830_cast")];
            tensor<fp16, []> var_13831_to_fp16 = const()[name = tensor<string, []>("op_13831_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1445_cast = mul(x = var_13830_cast, y = var_13831_to_fp16)[name = tensor<string, []>("aw_chunk_1445_cast")];
            tensor<string, []> var_13834_equation_0 = const()[name = tensor<string, []>("op_13834_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13834_cast = einsum(equation = var_13834_equation_0, values = (var_13592_cast, var_13551_cast))[name = tensor<string, []>("op_13834_cast")];
            tensor<fp16, []> var_13835_to_fp16 = const()[name = tensor<string, []>("op_13835_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1447_cast = mul(x = var_13834_cast, y = var_13835_to_fp16)[name = tensor<string, []>("aw_chunk_1447_cast")];
            tensor<string, []> var_13838_equation_0 = const()[name = tensor<string, []>("op_13838_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13838_cast = einsum(equation = var_13838_equation_0, values = (var_13592_cast, var_13552_cast))[name = tensor<string, []>("op_13838_cast")];
            tensor<fp16, []> var_13839_to_fp16 = const()[name = tensor<string, []>("op_13839_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1449_cast = mul(x = var_13838_cast, y = var_13839_to_fp16)[name = tensor<string, []>("aw_chunk_1449_cast")];
            tensor<string, []> var_13842_equation_0 = const()[name = tensor<string, []>("op_13842_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13842_cast = einsum(equation = var_13842_equation_0, values = (var_13592_cast, var_13553_cast))[name = tensor<string, []>("op_13842_cast")];
            tensor<fp16, []> var_13843_to_fp16 = const()[name = tensor<string, []>("op_13843_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1451_cast = mul(x = var_13842_cast, y = var_13843_to_fp16)[name = tensor<string, []>("aw_chunk_1451_cast")];
            tensor<string, []> var_13846_equation_0 = const()[name = tensor<string, []>("op_13846_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13846_cast = einsum(equation = var_13846_equation_0, values = (var_13592_cast, var_13554_cast))[name = tensor<string, []>("op_13846_cast")];
            tensor<fp16, []> var_13847_to_fp16 = const()[name = tensor<string, []>("op_13847_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1453_cast = mul(x = var_13846_cast, y = var_13847_to_fp16)[name = tensor<string, []>("aw_chunk_1453_cast")];
            tensor<string, []> var_13850_equation_0 = const()[name = tensor<string, []>("op_13850_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13850_cast = einsum(equation = var_13850_equation_0, values = (var_13592_cast, var_13555_cast))[name = tensor<string, []>("op_13850_cast")];
            tensor<fp16, []> var_13851_to_fp16 = const()[name = tensor<string, []>("op_13851_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1455_cast = mul(x = var_13850_cast, y = var_13851_to_fp16)[name = tensor<string, []>("aw_chunk_1455_cast")];
            tensor<string, []> var_13854_equation_0 = const()[name = tensor<string, []>("op_13854_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13854_cast = einsum(equation = var_13854_equation_0, values = (var_13596_cast, var_13556_cast))[name = tensor<string, []>("op_13854_cast")];
            tensor<fp16, []> var_13855_to_fp16 = const()[name = tensor<string, []>("op_13855_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1457_cast = mul(x = var_13854_cast, y = var_13855_to_fp16)[name = tensor<string, []>("aw_chunk_1457_cast")];
            tensor<string, []> var_13858_equation_0 = const()[name = tensor<string, []>("op_13858_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13858_cast = einsum(equation = var_13858_equation_0, values = (var_13596_cast, var_13557_cast))[name = tensor<string, []>("op_13858_cast")];
            tensor<fp16, []> var_13859_to_fp16 = const()[name = tensor<string, []>("op_13859_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1459_cast = mul(x = var_13858_cast, y = var_13859_to_fp16)[name = tensor<string, []>("aw_chunk_1459_cast")];
            tensor<string, []> var_13862_equation_0 = const()[name = tensor<string, []>("op_13862_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13862_cast = einsum(equation = var_13862_equation_0, values = (var_13596_cast, var_13558_cast))[name = tensor<string, []>("op_13862_cast")];
            tensor<fp16, []> var_13863_to_fp16 = const()[name = tensor<string, []>("op_13863_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1461_cast = mul(x = var_13862_cast, y = var_13863_to_fp16)[name = tensor<string, []>("aw_chunk_1461_cast")];
            tensor<string, []> var_13866_equation_0 = const()[name = tensor<string, []>("op_13866_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13866_cast = einsum(equation = var_13866_equation_0, values = (var_13596_cast, var_13559_cast))[name = tensor<string, []>("op_13866_cast")];
            tensor<fp16, []> var_13867_to_fp16 = const()[name = tensor<string, []>("op_13867_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1463_cast = mul(x = var_13866_cast, y = var_13867_to_fp16)[name = tensor<string, []>("aw_chunk_1463_cast")];
            tensor<string, []> var_13870_equation_0 = const()[name = tensor<string, []>("op_13870_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13870_cast = einsum(equation = var_13870_equation_0, values = (var_13596_cast, var_13560_cast))[name = tensor<string, []>("op_13870_cast")];
            tensor<fp16, []> var_13871_to_fp16 = const()[name = tensor<string, []>("op_13871_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1465_cast = mul(x = var_13870_cast, y = var_13871_to_fp16)[name = tensor<string, []>("aw_chunk_1465_cast")];
            tensor<string, []> var_13874_equation_0 = const()[name = tensor<string, []>("op_13874_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13874_cast = einsum(equation = var_13874_equation_0, values = (var_13596_cast, var_13561_cast))[name = tensor<string, []>("op_13874_cast")];
            tensor<fp16, []> var_13875_to_fp16 = const()[name = tensor<string, []>("op_13875_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1467_cast = mul(x = var_13874_cast, y = var_13875_to_fp16)[name = tensor<string, []>("aw_chunk_1467_cast")];
            tensor<string, []> var_13878_equation_0 = const()[name = tensor<string, []>("op_13878_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13878_cast = einsum(equation = var_13878_equation_0, values = (var_13596_cast, var_13562_cast))[name = tensor<string, []>("op_13878_cast")];
            tensor<fp16, []> var_13879_to_fp16 = const()[name = tensor<string, []>("op_13879_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1469_cast = mul(x = var_13878_cast, y = var_13879_to_fp16)[name = tensor<string, []>("aw_chunk_1469_cast")];
            tensor<string, []> var_13882_equation_0 = const()[name = tensor<string, []>("op_13882_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13882_cast = einsum(equation = var_13882_equation_0, values = (var_13596_cast, var_13563_cast))[name = tensor<string, []>("op_13882_cast")];
            tensor<fp16, []> var_13883_to_fp16 = const()[name = tensor<string, []>("op_13883_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1471_cast = mul(x = var_13882_cast, y = var_13883_to_fp16)[name = tensor<string, []>("aw_chunk_1471_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13885_cast = softmax(axis = var_10375, x = aw_chunk_1345_cast)[name = tensor<string, []>("op_13885_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13886_cast = softmax(axis = var_10375, x = aw_chunk_1347_cast)[name = tensor<string, []>("op_13886_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13887_cast = softmax(axis = var_10375, x = aw_chunk_1349_cast)[name = tensor<string, []>("op_13887_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13888_cast = softmax(axis = var_10375, x = aw_chunk_1351_cast)[name = tensor<string, []>("op_13888_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13889_cast = softmax(axis = var_10375, x = aw_chunk_1353_cast)[name = tensor<string, []>("op_13889_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13890_cast = softmax(axis = var_10375, x = aw_chunk_1355_cast)[name = tensor<string, []>("op_13890_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13891_cast = softmax(axis = var_10375, x = aw_chunk_1357_cast)[name = tensor<string, []>("op_13891_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13892_cast = softmax(axis = var_10375, x = aw_chunk_1359_cast)[name = tensor<string, []>("op_13892_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13893_cast = softmax(axis = var_10375, x = aw_chunk_1361_cast)[name = tensor<string, []>("op_13893_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13894_cast = softmax(axis = var_10375, x = aw_chunk_1363_cast)[name = tensor<string, []>("op_13894_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13895_cast = softmax(axis = var_10375, x = aw_chunk_1365_cast)[name = tensor<string, []>("op_13895_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13896_cast = softmax(axis = var_10375, x = aw_chunk_1367_cast)[name = tensor<string, []>("op_13896_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13897_cast = softmax(axis = var_10375, x = aw_chunk_1369_cast)[name = tensor<string, []>("op_13897_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13898_cast = softmax(axis = var_10375, x = aw_chunk_1371_cast)[name = tensor<string, []>("op_13898_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13899_cast = softmax(axis = var_10375, x = aw_chunk_1373_cast)[name = tensor<string, []>("op_13899_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13900_cast = softmax(axis = var_10375, x = aw_chunk_1375_cast)[name = tensor<string, []>("op_13900_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13901_cast = softmax(axis = var_10375, x = aw_chunk_1377_cast)[name = tensor<string, []>("op_13901_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13902_cast = softmax(axis = var_10375, x = aw_chunk_1379_cast)[name = tensor<string, []>("op_13902_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13903_cast = softmax(axis = var_10375, x = aw_chunk_1381_cast)[name = tensor<string, []>("op_13903_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13904_cast = softmax(axis = var_10375, x = aw_chunk_1383_cast)[name = tensor<string, []>("op_13904_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13905_cast = softmax(axis = var_10375, x = aw_chunk_1385_cast)[name = tensor<string, []>("op_13905_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13906_cast = softmax(axis = var_10375, x = aw_chunk_1387_cast)[name = tensor<string, []>("op_13906_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13907_cast = softmax(axis = var_10375, x = aw_chunk_1389_cast)[name = tensor<string, []>("op_13907_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13908_cast = softmax(axis = var_10375, x = aw_chunk_1391_cast)[name = tensor<string, []>("op_13908_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13909_cast = softmax(axis = var_10375, x = aw_chunk_1393_cast)[name = tensor<string, []>("op_13909_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13910_cast = softmax(axis = var_10375, x = aw_chunk_1395_cast)[name = tensor<string, []>("op_13910_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13911_cast = softmax(axis = var_10375, x = aw_chunk_1397_cast)[name = tensor<string, []>("op_13911_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13912_cast = softmax(axis = var_10375, x = aw_chunk_1399_cast)[name = tensor<string, []>("op_13912_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13913_cast = softmax(axis = var_10375, x = aw_chunk_1401_cast)[name = tensor<string, []>("op_13913_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13914_cast = softmax(axis = var_10375, x = aw_chunk_1403_cast)[name = tensor<string, []>("op_13914_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13915_cast = softmax(axis = var_10375, x = aw_chunk_1405_cast)[name = tensor<string, []>("op_13915_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13916_cast = softmax(axis = var_10375, x = aw_chunk_1407_cast)[name = tensor<string, []>("op_13916_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13917_cast = softmax(axis = var_10375, x = aw_chunk_1409_cast)[name = tensor<string, []>("op_13917_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13918_cast = softmax(axis = var_10375, x = aw_chunk_1411_cast)[name = tensor<string, []>("op_13918_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13919_cast = softmax(axis = var_10375, x = aw_chunk_1413_cast)[name = tensor<string, []>("op_13919_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13920_cast = softmax(axis = var_10375, x = aw_chunk_1415_cast)[name = tensor<string, []>("op_13920_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13921_cast = softmax(axis = var_10375, x = aw_chunk_1417_cast)[name = tensor<string, []>("op_13921_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13922_cast = softmax(axis = var_10375, x = aw_chunk_1419_cast)[name = tensor<string, []>("op_13922_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13923_cast = softmax(axis = var_10375, x = aw_chunk_1421_cast)[name = tensor<string, []>("op_13923_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13924_cast = softmax(axis = var_10375, x = aw_chunk_1423_cast)[name = tensor<string, []>("op_13924_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13925_cast = softmax(axis = var_10375, x = aw_chunk_1425_cast)[name = tensor<string, []>("op_13925_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13926_cast = softmax(axis = var_10375, x = aw_chunk_1427_cast)[name = tensor<string, []>("op_13926_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13927_cast = softmax(axis = var_10375, x = aw_chunk_1429_cast)[name = tensor<string, []>("op_13927_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13928_cast = softmax(axis = var_10375, x = aw_chunk_1431_cast)[name = tensor<string, []>("op_13928_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13929_cast = softmax(axis = var_10375, x = aw_chunk_1433_cast)[name = tensor<string, []>("op_13929_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13930_cast = softmax(axis = var_10375, x = aw_chunk_1435_cast)[name = tensor<string, []>("op_13930_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13931_cast = softmax(axis = var_10375, x = aw_chunk_1437_cast)[name = tensor<string, []>("op_13931_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13932_cast = softmax(axis = var_10375, x = aw_chunk_1439_cast)[name = tensor<string, []>("op_13932_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13933_cast = softmax(axis = var_10375, x = aw_chunk_1441_cast)[name = tensor<string, []>("op_13933_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13934_cast = softmax(axis = var_10375, x = aw_chunk_1443_cast)[name = tensor<string, []>("op_13934_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13935_cast = softmax(axis = var_10375, x = aw_chunk_1445_cast)[name = tensor<string, []>("op_13935_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13936_cast = softmax(axis = var_10375, x = aw_chunk_1447_cast)[name = tensor<string, []>("op_13936_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13937_cast = softmax(axis = var_10375, x = aw_chunk_1449_cast)[name = tensor<string, []>("op_13937_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13938_cast = softmax(axis = var_10375, x = aw_chunk_1451_cast)[name = tensor<string, []>("op_13938_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13939_cast = softmax(axis = var_10375, x = aw_chunk_1453_cast)[name = tensor<string, []>("op_13939_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13940_cast = softmax(axis = var_10375, x = aw_chunk_1455_cast)[name = tensor<string, []>("op_13940_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13941_cast = softmax(axis = var_10375, x = aw_chunk_1457_cast)[name = tensor<string, []>("op_13941_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13942_cast = softmax(axis = var_10375, x = aw_chunk_1459_cast)[name = tensor<string, []>("op_13942_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13943_cast = softmax(axis = var_10375, x = aw_chunk_1461_cast)[name = tensor<string, []>("op_13943_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13944_cast = softmax(axis = var_10375, x = aw_chunk_1463_cast)[name = tensor<string, []>("op_13944_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13945_cast = softmax(axis = var_10375, x = aw_chunk_1465_cast)[name = tensor<string, []>("op_13945_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13946_cast = softmax(axis = var_10375, x = aw_chunk_1467_cast)[name = tensor<string, []>("op_13946_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13947_cast = softmax(axis = var_10375, x = aw_chunk_1469_cast)[name = tensor<string, []>("op_13947_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13948_cast = softmax(axis = var_10375, x = aw_chunk_1471_cast)[name = tensor<string, []>("op_13948_cast")];
            tensor<string, []> var_13950_equation_0 = const()[name = tensor<string, []>("op_13950_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13950_cast = einsum(equation = var_13950_equation_0, values = (var_13598_cast, var_13885_cast))[name = tensor<string, []>("op_13950_cast")];
            tensor<string, []> var_13952_equation_0 = const()[name = tensor<string, []>("op_13952_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13952_cast = einsum(equation = var_13952_equation_0, values = (var_13598_cast, var_13886_cast))[name = tensor<string, []>("op_13952_cast")];
            tensor<string, []> var_13954_equation_0 = const()[name = tensor<string, []>("op_13954_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13954_cast = einsum(equation = var_13954_equation_0, values = (var_13598_cast, var_13887_cast))[name = tensor<string, []>("op_13954_cast")];
            tensor<string, []> var_13956_equation_0 = const()[name = tensor<string, []>("op_13956_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13956_cast = einsum(equation = var_13956_equation_0, values = (var_13598_cast, var_13888_cast))[name = tensor<string, []>("op_13956_cast")];
            tensor<string, []> var_13958_equation_0 = const()[name = tensor<string, []>("op_13958_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13958_cast = einsum(equation = var_13958_equation_0, values = (var_13598_cast, var_13889_cast))[name = tensor<string, []>("op_13958_cast")];
            tensor<string, []> var_13960_equation_0 = const()[name = tensor<string, []>("op_13960_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13960_cast = einsum(equation = var_13960_equation_0, values = (var_13598_cast, var_13890_cast))[name = tensor<string, []>("op_13960_cast")];
            tensor<string, []> var_13962_equation_0 = const()[name = tensor<string, []>("op_13962_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13962_cast = einsum(equation = var_13962_equation_0, values = (var_13598_cast, var_13891_cast))[name = tensor<string, []>("op_13962_cast")];
            tensor<string, []> var_13964_equation_0 = const()[name = tensor<string, []>("op_13964_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13964_cast = einsum(equation = var_13964_equation_0, values = (var_13598_cast, var_13892_cast))[name = tensor<string, []>("op_13964_cast")];
            tensor<string, []> var_13966_equation_0 = const()[name = tensor<string, []>("op_13966_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13966_cast = einsum(equation = var_13966_equation_0, values = (var_13602_cast, var_13893_cast))[name = tensor<string, []>("op_13966_cast")];
            tensor<string, []> var_13968_equation_0 = const()[name = tensor<string, []>("op_13968_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13968_cast = einsum(equation = var_13968_equation_0, values = (var_13602_cast, var_13894_cast))[name = tensor<string, []>("op_13968_cast")];
            tensor<string, []> var_13970_equation_0 = const()[name = tensor<string, []>("op_13970_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13970_cast = einsum(equation = var_13970_equation_0, values = (var_13602_cast, var_13895_cast))[name = tensor<string, []>("op_13970_cast")];
            tensor<string, []> var_13972_equation_0 = const()[name = tensor<string, []>("op_13972_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13972_cast = einsum(equation = var_13972_equation_0, values = (var_13602_cast, var_13896_cast))[name = tensor<string, []>("op_13972_cast")];
            tensor<string, []> var_13974_equation_0 = const()[name = tensor<string, []>("op_13974_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13974_cast = einsum(equation = var_13974_equation_0, values = (var_13602_cast, var_13897_cast))[name = tensor<string, []>("op_13974_cast")];
            tensor<string, []> var_13976_equation_0 = const()[name = tensor<string, []>("op_13976_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13976_cast = einsum(equation = var_13976_equation_0, values = (var_13602_cast, var_13898_cast))[name = tensor<string, []>("op_13976_cast")];
            tensor<string, []> var_13978_equation_0 = const()[name = tensor<string, []>("op_13978_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13978_cast = einsum(equation = var_13978_equation_0, values = (var_13602_cast, var_13899_cast))[name = tensor<string, []>("op_13978_cast")];
            tensor<string, []> var_13980_equation_0 = const()[name = tensor<string, []>("op_13980_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13980_cast = einsum(equation = var_13980_equation_0, values = (var_13602_cast, var_13900_cast))[name = tensor<string, []>("op_13980_cast")];
            tensor<string, []> var_13982_equation_0 = const()[name = tensor<string, []>("op_13982_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13982_cast = einsum(equation = var_13982_equation_0, values = (var_13606_cast, var_13901_cast))[name = tensor<string, []>("op_13982_cast")];
            tensor<string, []> var_13984_equation_0 = const()[name = tensor<string, []>("op_13984_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13984_cast = einsum(equation = var_13984_equation_0, values = (var_13606_cast, var_13902_cast))[name = tensor<string, []>("op_13984_cast")];
            tensor<string, []> var_13986_equation_0 = const()[name = tensor<string, []>("op_13986_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13986_cast = einsum(equation = var_13986_equation_0, values = (var_13606_cast, var_13903_cast))[name = tensor<string, []>("op_13986_cast")];
            tensor<string, []> var_13988_equation_0 = const()[name = tensor<string, []>("op_13988_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13988_cast = einsum(equation = var_13988_equation_0, values = (var_13606_cast, var_13904_cast))[name = tensor<string, []>("op_13988_cast")];
            tensor<string, []> var_13990_equation_0 = const()[name = tensor<string, []>("op_13990_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13990_cast = einsum(equation = var_13990_equation_0, values = (var_13606_cast, var_13905_cast))[name = tensor<string, []>("op_13990_cast")];
            tensor<string, []> var_13992_equation_0 = const()[name = tensor<string, []>("op_13992_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13992_cast = einsum(equation = var_13992_equation_0, values = (var_13606_cast, var_13906_cast))[name = tensor<string, []>("op_13992_cast")];
            tensor<string, []> var_13994_equation_0 = const()[name = tensor<string, []>("op_13994_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13994_cast = einsum(equation = var_13994_equation_0, values = (var_13606_cast, var_13907_cast))[name = tensor<string, []>("op_13994_cast")];
            tensor<string, []> var_13996_equation_0 = const()[name = tensor<string, []>("op_13996_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13996_cast = einsum(equation = var_13996_equation_0, values = (var_13606_cast, var_13908_cast))[name = tensor<string, []>("op_13996_cast")];
            tensor<string, []> var_13998_equation_0 = const()[name = tensor<string, []>("op_13998_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_13998_cast = einsum(equation = var_13998_equation_0, values = (var_13610_cast, var_13909_cast))[name = tensor<string, []>("op_13998_cast")];
            tensor<string, []> var_14000_equation_0 = const()[name = tensor<string, []>("op_14000_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14000_cast = einsum(equation = var_14000_equation_0, values = (var_13610_cast, var_13910_cast))[name = tensor<string, []>("op_14000_cast")];
            tensor<string, []> var_14002_equation_0 = const()[name = tensor<string, []>("op_14002_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14002_cast = einsum(equation = var_14002_equation_0, values = (var_13610_cast, var_13911_cast))[name = tensor<string, []>("op_14002_cast")];
            tensor<string, []> var_14004_equation_0 = const()[name = tensor<string, []>("op_14004_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14004_cast = einsum(equation = var_14004_equation_0, values = (var_13610_cast, var_13912_cast))[name = tensor<string, []>("op_14004_cast")];
            tensor<string, []> var_14006_equation_0 = const()[name = tensor<string, []>("op_14006_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14006_cast = einsum(equation = var_14006_equation_0, values = (var_13610_cast, var_13913_cast))[name = tensor<string, []>("op_14006_cast")];
            tensor<string, []> var_14008_equation_0 = const()[name = tensor<string, []>("op_14008_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14008_cast = einsum(equation = var_14008_equation_0, values = (var_13610_cast, var_13914_cast))[name = tensor<string, []>("op_14008_cast")];
            tensor<string, []> var_14010_equation_0 = const()[name = tensor<string, []>("op_14010_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14010_cast = einsum(equation = var_14010_equation_0, values = (var_13610_cast, var_13915_cast))[name = tensor<string, []>("op_14010_cast")];
            tensor<string, []> var_14012_equation_0 = const()[name = tensor<string, []>("op_14012_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14012_cast = einsum(equation = var_14012_equation_0, values = (var_13610_cast, var_13916_cast))[name = tensor<string, []>("op_14012_cast")];
            tensor<string, []> var_14014_equation_0 = const()[name = tensor<string, []>("op_14014_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14014_cast = einsum(equation = var_14014_equation_0, values = (var_13614_cast, var_13917_cast))[name = tensor<string, []>("op_14014_cast")];
            tensor<string, []> var_14016_equation_0 = const()[name = tensor<string, []>("op_14016_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14016_cast = einsum(equation = var_14016_equation_0, values = (var_13614_cast, var_13918_cast))[name = tensor<string, []>("op_14016_cast")];
            tensor<string, []> var_14018_equation_0 = const()[name = tensor<string, []>("op_14018_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14018_cast = einsum(equation = var_14018_equation_0, values = (var_13614_cast, var_13919_cast))[name = tensor<string, []>("op_14018_cast")];
            tensor<string, []> var_14020_equation_0 = const()[name = tensor<string, []>("op_14020_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14020_cast = einsum(equation = var_14020_equation_0, values = (var_13614_cast, var_13920_cast))[name = tensor<string, []>("op_14020_cast")];
            tensor<string, []> var_14022_equation_0 = const()[name = tensor<string, []>("op_14022_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14022_cast = einsum(equation = var_14022_equation_0, values = (var_13614_cast, var_13921_cast))[name = tensor<string, []>("op_14022_cast")];
            tensor<string, []> var_14024_equation_0 = const()[name = tensor<string, []>("op_14024_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14024_cast = einsum(equation = var_14024_equation_0, values = (var_13614_cast, var_13922_cast))[name = tensor<string, []>("op_14024_cast")];
            tensor<string, []> var_14026_equation_0 = const()[name = tensor<string, []>("op_14026_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14026_cast = einsum(equation = var_14026_equation_0, values = (var_13614_cast, var_13923_cast))[name = tensor<string, []>("op_14026_cast")];
            tensor<string, []> var_14028_equation_0 = const()[name = tensor<string, []>("op_14028_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14028_cast = einsum(equation = var_14028_equation_0, values = (var_13614_cast, var_13924_cast))[name = tensor<string, []>("op_14028_cast")];
            tensor<string, []> var_14030_equation_0 = const()[name = tensor<string, []>("op_14030_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14030_cast = einsum(equation = var_14030_equation_0, values = (var_13618_cast, var_13925_cast))[name = tensor<string, []>("op_14030_cast")];
            tensor<string, []> var_14032_equation_0 = const()[name = tensor<string, []>("op_14032_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14032_cast = einsum(equation = var_14032_equation_0, values = (var_13618_cast, var_13926_cast))[name = tensor<string, []>("op_14032_cast")];
            tensor<string, []> var_14034_equation_0 = const()[name = tensor<string, []>("op_14034_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14034_cast = einsum(equation = var_14034_equation_0, values = (var_13618_cast, var_13927_cast))[name = tensor<string, []>("op_14034_cast")];
            tensor<string, []> var_14036_equation_0 = const()[name = tensor<string, []>("op_14036_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14036_cast = einsum(equation = var_14036_equation_0, values = (var_13618_cast, var_13928_cast))[name = tensor<string, []>("op_14036_cast")];
            tensor<string, []> var_14038_equation_0 = const()[name = tensor<string, []>("op_14038_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14038_cast = einsum(equation = var_14038_equation_0, values = (var_13618_cast, var_13929_cast))[name = tensor<string, []>("op_14038_cast")];
            tensor<string, []> var_14040_equation_0 = const()[name = tensor<string, []>("op_14040_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14040_cast = einsum(equation = var_14040_equation_0, values = (var_13618_cast, var_13930_cast))[name = tensor<string, []>("op_14040_cast")];
            tensor<string, []> var_14042_equation_0 = const()[name = tensor<string, []>("op_14042_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14042_cast = einsum(equation = var_14042_equation_0, values = (var_13618_cast, var_13931_cast))[name = tensor<string, []>("op_14042_cast")];
            tensor<string, []> var_14044_equation_0 = const()[name = tensor<string, []>("op_14044_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14044_cast = einsum(equation = var_14044_equation_0, values = (var_13618_cast, var_13932_cast))[name = tensor<string, []>("op_14044_cast")];
            tensor<string, []> var_14046_equation_0 = const()[name = tensor<string, []>("op_14046_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14046_cast = einsum(equation = var_14046_equation_0, values = (var_13622_cast, var_13933_cast))[name = tensor<string, []>("op_14046_cast")];
            tensor<string, []> var_14048_equation_0 = const()[name = tensor<string, []>("op_14048_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14048_cast = einsum(equation = var_14048_equation_0, values = (var_13622_cast, var_13934_cast))[name = tensor<string, []>("op_14048_cast")];
            tensor<string, []> var_14050_equation_0 = const()[name = tensor<string, []>("op_14050_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14050_cast = einsum(equation = var_14050_equation_0, values = (var_13622_cast, var_13935_cast))[name = tensor<string, []>("op_14050_cast")];
            tensor<string, []> var_14052_equation_0 = const()[name = tensor<string, []>("op_14052_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14052_cast = einsum(equation = var_14052_equation_0, values = (var_13622_cast, var_13936_cast))[name = tensor<string, []>("op_14052_cast")];
            tensor<string, []> var_14054_equation_0 = const()[name = tensor<string, []>("op_14054_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14054_cast = einsum(equation = var_14054_equation_0, values = (var_13622_cast, var_13937_cast))[name = tensor<string, []>("op_14054_cast")];
            tensor<string, []> var_14056_equation_0 = const()[name = tensor<string, []>("op_14056_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14056_cast = einsum(equation = var_14056_equation_0, values = (var_13622_cast, var_13938_cast))[name = tensor<string, []>("op_14056_cast")];
            tensor<string, []> var_14058_equation_0 = const()[name = tensor<string, []>("op_14058_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14058_cast = einsum(equation = var_14058_equation_0, values = (var_13622_cast, var_13939_cast))[name = tensor<string, []>("op_14058_cast")];
            tensor<string, []> var_14060_equation_0 = const()[name = tensor<string, []>("op_14060_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14060_cast = einsum(equation = var_14060_equation_0, values = (var_13622_cast, var_13940_cast))[name = tensor<string, []>("op_14060_cast")];
            tensor<string, []> var_14062_equation_0 = const()[name = tensor<string, []>("op_14062_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14062_cast = einsum(equation = var_14062_equation_0, values = (var_13626_cast, var_13941_cast))[name = tensor<string, []>("op_14062_cast")];
            tensor<string, []> var_14064_equation_0 = const()[name = tensor<string, []>("op_14064_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14064_cast = einsum(equation = var_14064_equation_0, values = (var_13626_cast, var_13942_cast))[name = tensor<string, []>("op_14064_cast")];
            tensor<string, []> var_14066_equation_0 = const()[name = tensor<string, []>("op_14066_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14066_cast = einsum(equation = var_14066_equation_0, values = (var_13626_cast, var_13943_cast))[name = tensor<string, []>("op_14066_cast")];
            tensor<string, []> var_14068_equation_0 = const()[name = tensor<string, []>("op_14068_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14068_cast = einsum(equation = var_14068_equation_0, values = (var_13626_cast, var_13944_cast))[name = tensor<string, []>("op_14068_cast")];
            tensor<string, []> var_14070_equation_0 = const()[name = tensor<string, []>("op_14070_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14070_cast = einsum(equation = var_14070_equation_0, values = (var_13626_cast, var_13945_cast))[name = tensor<string, []>("op_14070_cast")];
            tensor<string, []> var_14072_equation_0 = const()[name = tensor<string, []>("op_14072_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14072_cast = einsum(equation = var_14072_equation_0, values = (var_13626_cast, var_13946_cast))[name = tensor<string, []>("op_14072_cast")];
            tensor<string, []> var_14074_equation_0 = const()[name = tensor<string, []>("op_14074_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14074_cast = einsum(equation = var_14074_equation_0, values = (var_13626_cast, var_13947_cast))[name = tensor<string, []>("op_14074_cast")];
            tensor<string, []> var_14076_equation_0 = const()[name = tensor<string, []>("op_14076_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14076_cast = einsum(equation = var_14076_equation_0, values = (var_13626_cast, var_13948_cast))[name = tensor<string, []>("op_14076_cast")];
            tensor<bool, []> var_14078_interleave_0 = const()[name = tensor<string, []>("op_14078_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_14078_cast = concat(axis = var_10347, interleave = var_14078_interleave_0, values = (var_13950_cast, var_13952_cast, var_13954_cast, var_13956_cast, var_13958_cast, var_13960_cast, var_13962_cast, var_13964_cast))[name = tensor<string, []>("op_14078_cast")];
            tensor<bool, []> var_14080_interleave_0 = const()[name = tensor<string, []>("op_14080_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_14080_cast = concat(axis = var_10347, interleave = var_14080_interleave_0, values = (var_13966_cast, var_13968_cast, var_13970_cast, var_13972_cast, var_13974_cast, var_13976_cast, var_13978_cast, var_13980_cast))[name = tensor<string, []>("op_14080_cast")];
            tensor<bool, []> var_14082_interleave_0 = const()[name = tensor<string, []>("op_14082_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_14082_cast = concat(axis = var_10347, interleave = var_14082_interleave_0, values = (var_13982_cast, var_13984_cast, var_13986_cast, var_13988_cast, var_13990_cast, var_13992_cast, var_13994_cast, var_13996_cast))[name = tensor<string, []>("op_14082_cast")];
            tensor<bool, []> var_14084_interleave_0 = const()[name = tensor<string, []>("op_14084_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_14084_cast = concat(axis = var_10347, interleave = var_14084_interleave_0, values = (var_13998_cast, var_14000_cast, var_14002_cast, var_14004_cast, var_14006_cast, var_14008_cast, var_14010_cast, var_14012_cast))[name = tensor<string, []>("op_14084_cast")];
            tensor<bool, []> var_14086_interleave_0 = const()[name = tensor<string, []>("op_14086_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_14086_cast = concat(axis = var_10347, interleave = var_14086_interleave_0, values = (var_14014_cast, var_14016_cast, var_14018_cast, var_14020_cast, var_14022_cast, var_14024_cast, var_14026_cast, var_14028_cast))[name = tensor<string, []>("op_14086_cast")];
            tensor<bool, []> var_14088_interleave_0 = const()[name = tensor<string, []>("op_14088_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_14088_cast = concat(axis = var_10347, interleave = var_14088_interleave_0, values = (var_14030_cast, var_14032_cast, var_14034_cast, var_14036_cast, var_14038_cast, var_14040_cast, var_14042_cast, var_14044_cast))[name = tensor<string, []>("op_14088_cast")];
            tensor<bool, []> var_14090_interleave_0 = const()[name = tensor<string, []>("op_14090_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_14090_cast = concat(axis = var_10347, interleave = var_14090_interleave_0, values = (var_14046_cast, var_14048_cast, var_14050_cast, var_14052_cast, var_14054_cast, var_14056_cast, var_14058_cast, var_14060_cast))[name = tensor<string, []>("op_14090_cast")];
            tensor<bool, []> var_14092_interleave_0 = const()[name = tensor<string, []>("op_14092_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_14092_cast = concat(axis = var_10347, interleave = var_14092_interleave_0, values = (var_14062_cast, var_14064_cast, var_14066_cast, var_14068_cast, var_14070_cast, var_14072_cast, var_14074_cast, var_14076_cast))[name = tensor<string, []>("op_14092_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_10375, interleave = input_517_interleave_0, values = (var_14078_cast, var_14080_cast, var_14082_cast, var_14084_cast, var_14086_cast, var_14088_cast, var_14090_cast, var_14092_cast))[name = tensor<string, []>("input_517_cast")];
            tensor<int32, [2]> var_14098 = const()[name = tensor<string, []>("op_14098"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14100 = const()[name = tensor<string, []>("op_14100"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_14102_pad_type_0 = const()[name = tensor<string, []>("op_14102_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_14102_pad_0 = const()[name = tensor<string, []>("op_14102_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643550016))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643626880))), name = tensor<string, []>("up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            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, []>(643627072)))];
            tensor<fp16, [2, 320, 1, 4096]> var_14102_cast = conv(bias = up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_14100, groups = var_10375, pad = var_14102_pad_0, pad_type = var_14102_pad_type_0, strides = var_14098, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_517_cast)[name = tensor<string, []>("op_14102_cast")];
            tensor<fp16, [2, 320, 1, 4096]> inputs_93_cast = add(x = var_14102_cast, y = inputs_91_cast)[name = tensor<string, []>("inputs_93_cast")];
            tensor<int32, [1]> var_14106 = const()[name = tensor<string, []>("op_14106"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_93_cast = reduce_mean(axes = var_14106, keep_dims = var_10370, x = inputs_93_cast)[name = tensor<string, []>("channels_mean_93_cast")];
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_93_cast = sub(x = inputs_93_cast, y = channels_mean_93_cast)[name = tensor<string, []>("zero_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)[name = tensor<string, []>("zero_mean_sq_93_cast")];
            tensor<int32, [1]> var_14110 = const()[name = tensor<string, []>("op_14110"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_14111_cast = reduce_mean(axes = var_14110, keep_dims = var_10370, x = zero_mean_sq_93_cast)[name = tensor<string, []>("op_14111_cast")];
            tensor<fp16, []> var_14112_to_fp16 = const()[name = tensor<string, []>("op_14112_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_14113_cast = add(x = var_14111_cast, y = var_14112_to_fp16)[name = tensor<string, []>("op_14113_cast")];
            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_14113_cast)[name = tensor<string, []>("denom_93_cast")];
            tensor<fp16, [2, 320, 1, 4096]> out_93_cast = mul(x = zero_mean_93_cast, y = denom_93_cast)[name = tensor<string, []>("out_93_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_14117_to_fp16 = const()[name = tensor<string, []>("op_14117_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643627776)))];
            tensor<fp16, [2, 320, 1, 4096]> var_14118_cast = add(x = out_93_cast, y = var_14117_to_fp16)[name = tensor<string, []>("op_14118_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_14120_to_fp16 = const()[name = tensor<string, []>("op_14120_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643628480)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_329_cast = mul(x = var_14118_cast, y = var_14120_to_fp16)[name = tensor<string, []>("hidden_states_329_cast")];
            tensor<int32, [2]> var_14127 = const()[name = tensor<string, []>("op_14127"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14129 = const()[name = tensor<string, []>("op_14129"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643629184))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643706048))), name = tensor<string, []>("up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            tensor<fp16, [2, 320, 1, 4096]> q_cast = conv(dilations = var_14129, groups = var_10375, pad = q_pad_0, pad_type = q_pad_type_0, strides = var_14127, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_329_cast)[name = tensor<string, []>("q_cast")];
            tensor<int32, [2]> var_14133 = const()[name = tensor<string, []>("op_14133"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14135 = const()[name = tensor<string, []>("op_14135"), 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, 768, 1, 1]> up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [184320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643706240))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643890624))), name = tensor<string, []>("up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 768, 1, 1])];
            tensor<fp16, [2, 320, 1, 77]> k_125_cast = conv(dilations = var_14135, groups = var_10375, pad = k_125_pad_0, pad_type = k_125_pad_type_0, strides = var_14133, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("k_125_cast")];
            tensor<int32, [2]> var_14139 = const()[name = tensor<string, []>("op_14139"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14141 = const()[name = tensor<string, []>("op_14141"), 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, 768, 1, 1]> up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [184320]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643890816))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(644075200))), name = tensor<string, []>("up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 768, 1, 1])];
            tensor<fp16, [2, 320, 1, 77]> v_cast = conv(dilations = var_14141, groups = var_10375, pad = v_pad_0, pad_type = v_pad_type_0, strides = var_14139, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor<string, []>("v_cast")];
            tensor<int32, [4]> var_14145_begin_0 = const()[name = tensor<string, []>("op_14145_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14145_end_0 = const()[name = tensor<string, []>("op_14145_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_14145_end_mask_0 = const()[name = tensor<string, []>("op_14145_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_14145_cast = slice_by_index(begin = var_14145_begin_0, end = var_14145_end_0, end_mask = var_14145_end_mask_0, x = q_cast)[name = tensor<string, []>("op_14145_cast")];
            tensor<int32, [4]> var_14149_begin_0 = const()[name = tensor<string, []>("op_14149_begin_0"), val = tensor<int32, [4]>([0, 40, 0, 0])];
            tensor<int32, [4]> var_14149_end_0 = const()[name = tensor<string, []>("op_14149_end_0"), val = tensor<int32, [4]>([2, 80, 1, 4096])];
            tensor<bool, [4]> var_14149_end_mask_0 = const()[name = tensor<string, []>("op_14149_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_14149_cast = slice_by_index(begin = var_14149_begin_0, end = var_14149_end_0, end_mask = var_14149_end_mask_0, x = q_cast)[name = tensor<string, []>("op_14149_cast")];
            tensor<int32, [4]> var_14153_begin_0 = const()[name = tensor<string, []>("op_14153_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_14153_end_0 = const()[name = tensor<string, []>("op_14153_end_0"), val = tensor<int32, [4]>([2, 120, 1, 4096])];
            tensor<bool, [4]> var_14153_end_mask_0 = const()[name = tensor<string, []>("op_14153_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_14153_cast = slice_by_index(begin = var_14153_begin_0, end = var_14153_end_0, end_mask = var_14153_end_mask_0, x = q_cast)[name = tensor<string, []>("op_14153_cast")];
            tensor<int32, [4]> var_14157_begin_0 = const()[name = tensor<string, []>("op_14157_begin_0"), val = tensor<int32, [4]>([0, 120, 0, 0])];
            tensor<int32, [4]> var_14157_end_0 = const()[name = tensor<string, []>("op_14157_end_0"), val = tensor<int32, [4]>([2, 160, 1, 4096])];
            tensor<bool, [4]> var_14157_end_mask_0 = const()[name = tensor<string, []>("op_14157_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_14157_cast = slice_by_index(begin = var_14157_begin_0, end = var_14157_end_0, end_mask = var_14157_end_mask_0, x = q_cast)[name = tensor<string, []>("op_14157_cast")];
            tensor<int32, [4]> var_14161_begin_0 = const()[name = tensor<string, []>("op_14161_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_14161_end_0 = const()[name = tensor<string, []>("op_14161_end_0"), val = tensor<int32, [4]>([2, 200, 1, 4096])];
            tensor<bool, [4]> var_14161_end_mask_0 = const()[name = tensor<string, []>("op_14161_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_14161_cast = slice_by_index(begin = var_14161_begin_0, end = var_14161_end_0, end_mask = var_14161_end_mask_0, x = q_cast)[name = tensor<string, []>("op_14161_cast")];
            tensor<int32, [4]> var_14165_begin_0 = const()[name = tensor<string, []>("op_14165_begin_0"), val = tensor<int32, [4]>([0, 200, 0, 0])];
            tensor<int32, [4]> var_14165_end_0 = const()[name = tensor<string, []>("op_14165_end_0"), val = tensor<int32, [4]>([2, 240, 1, 4096])];
            tensor<bool, [4]> var_14165_end_mask_0 = const()[name = tensor<string, []>("op_14165_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_14165_cast = slice_by_index(begin = var_14165_begin_0, end = var_14165_end_0, end_mask = var_14165_end_mask_0, x = q_cast)[name = tensor<string, []>("op_14165_cast")];
            tensor<int32, [4]> var_14169_begin_0 = const()[name = tensor<string, []>("op_14169_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_14169_end_0 = const()[name = tensor<string, []>("op_14169_end_0"), val = tensor<int32, [4]>([2, 280, 1, 4096])];
            tensor<bool, [4]> var_14169_end_mask_0 = const()[name = tensor<string, []>("op_14169_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_14169_cast = slice_by_index(begin = var_14169_begin_0, end = var_14169_end_0, end_mask = var_14169_end_mask_0, x = q_cast)[name = tensor<string, []>("op_14169_cast")];
            tensor<int32, [4]> var_14173_begin_0 = const()[name = tensor<string, []>("op_14173_begin_0"), val = tensor<int32, [4]>([0, 280, 0, 0])];
            tensor<int32, [4]> var_14173_end_0 = const()[name = tensor<string, []>("op_14173_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_14173_end_mask_0 = const()[name = tensor<string, []>("op_14173_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 4096]> var_14173_cast = slice_by_index(begin = var_14173_begin_0, end = var_14173_end_0, end_mask = var_14173_end_mask_0, x = q_cast)[name = tensor<string, []>("op_14173_cast")];
            tensor<int32, [4]> var_14176_begin_0 = const()[name = tensor<string, []>("op_14176_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14176_end_0 = const()[name = tensor<string, []>("op_14176_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_14176_end_mask_0 = const()[name = tensor<string, []>("op_14176_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14176_cast = slice_by_index(begin = var_14176_begin_0, end = var_14176_end_0, end_mask = var_14176_end_mask_0, x = var_14145_cast)[name = tensor<string, []>("op_14176_cast")];
            tensor<int32, [4]> var_14177_begin_0 = const()[name = tensor<string, []>("op_14177_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_14177_end_0 = const()[name = tensor<string, []>("op_14177_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_14177_end_mask_0 = const()[name = tensor<string, []>("op_14177_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14177_cast = slice_by_index(begin = var_14177_begin_0, end = var_14177_end_0, end_mask = var_14177_end_mask_0, x = var_14145_cast)[name = tensor<string, []>("op_14177_cast")];
            tensor<int32, [4]> var_14178_begin_0 = const()[name = tensor<string, []>("op_14178_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_14178_end_0 = const()[name = tensor<string, []>("op_14178_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_14178_end_mask_0 = const()[name = tensor<string, []>("op_14178_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14178_cast = slice_by_index(begin = var_14178_begin_0, end = var_14178_end_0, end_mask = var_14178_end_mask_0, x = var_14145_cast)[name = tensor<string, []>("op_14178_cast")];
            tensor<int32, [4]> var_14179_begin_0 = const()[name = tensor<string, []>("op_14179_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_14179_end_0 = const()[name = tensor<string, []>("op_14179_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_14179_end_mask_0 = const()[name = tensor<string, []>("op_14179_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14179_cast = slice_by_index(begin = var_14179_begin_0, end = var_14179_end_0, end_mask = var_14179_end_mask_0, x = var_14145_cast)[name = tensor<string, []>("op_14179_cast")];
            tensor<int32, [4]> var_14180_begin_0 = const()[name = tensor<string, []>("op_14180_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_14180_end_0 = const()[name = tensor<string, []>("op_14180_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_14180_end_mask_0 = const()[name = tensor<string, []>("op_14180_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14180_cast = slice_by_index(begin = var_14180_begin_0, end = var_14180_end_0, end_mask = var_14180_end_mask_0, x = var_14145_cast)[name = tensor<string, []>("op_14180_cast")];
            tensor<int32, [4]> var_14181_begin_0 = const()[name = tensor<string, []>("op_14181_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_14181_end_0 = const()[name = tensor<string, []>("op_14181_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_14181_end_mask_0 = const()[name = tensor<string, []>("op_14181_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14181_cast = slice_by_index(begin = var_14181_begin_0, end = var_14181_end_0, end_mask = var_14181_end_mask_0, x = var_14145_cast)[name = tensor<string, []>("op_14181_cast")];
            tensor<int32, [4]> var_14182_begin_0 = const()[name = tensor<string, []>("op_14182_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_14182_end_0 = const()[name = tensor<string, []>("op_14182_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_14182_end_mask_0 = const()[name = tensor<string, []>("op_14182_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14182_cast = slice_by_index(begin = var_14182_begin_0, end = var_14182_end_0, end_mask = var_14182_end_mask_0, x = var_14145_cast)[name = tensor<string, []>("op_14182_cast")];
            tensor<int32, [4]> var_14183_begin_0 = const()[name = tensor<string, []>("op_14183_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_14183_end_0 = const()[name = tensor<string, []>("op_14183_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_14183_end_mask_0 = const()[name = tensor<string, []>("op_14183_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14183_cast = slice_by_index(begin = var_14183_begin_0, end = var_14183_end_0, end_mask = var_14183_end_mask_0, x = var_14145_cast)[name = tensor<string, []>("op_14183_cast")];
            tensor<int32, [4]> var_14184_begin_0 = const()[name = tensor<string, []>("op_14184_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14184_end_0 = const()[name = tensor<string, []>("op_14184_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_14184_end_mask_0 = const()[name = tensor<string, []>("op_14184_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14184_cast = slice_by_index(begin = var_14184_begin_0, end = var_14184_end_0, end_mask = var_14184_end_mask_0, x = var_14149_cast)[name = tensor<string, []>("op_14184_cast")];
            tensor<int32, [4]> var_14185_begin_0 = const()[name = tensor<string, []>("op_14185_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_14185_end_0 = const()[name = tensor<string, []>("op_14185_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_14185_end_mask_0 = const()[name = tensor<string, []>("op_14185_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14185_cast = slice_by_index(begin = var_14185_begin_0, end = var_14185_end_0, end_mask = var_14185_end_mask_0, x = var_14149_cast)[name = tensor<string, []>("op_14185_cast")];
            tensor<int32, [4]> var_14186_begin_0 = const()[name = tensor<string, []>("op_14186_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_14186_end_0 = const()[name = tensor<string, []>("op_14186_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_14186_end_mask_0 = const()[name = tensor<string, []>("op_14186_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14186_cast = slice_by_index(begin = var_14186_begin_0, end = var_14186_end_0, end_mask = var_14186_end_mask_0, x = var_14149_cast)[name = tensor<string, []>("op_14186_cast")];
            tensor<int32, [4]> var_14187_begin_0 = const()[name = tensor<string, []>("op_14187_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_14187_end_0 = const()[name = tensor<string, []>("op_14187_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_14187_end_mask_0 = const()[name = tensor<string, []>("op_14187_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14187_cast = slice_by_index(begin = var_14187_begin_0, end = var_14187_end_0, end_mask = var_14187_end_mask_0, x = var_14149_cast)[name = tensor<string, []>("op_14187_cast")];
            tensor<int32, [4]> var_14188_begin_0 = const()[name = tensor<string, []>("op_14188_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_14188_end_0 = const()[name = tensor<string, []>("op_14188_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_14188_end_mask_0 = const()[name = tensor<string, []>("op_14188_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14188_cast = slice_by_index(begin = var_14188_begin_0, end = var_14188_end_0, end_mask = var_14188_end_mask_0, x = var_14149_cast)[name = tensor<string, []>("op_14188_cast")];
            tensor<int32, [4]> var_14189_begin_0 = const()[name = tensor<string, []>("op_14189_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_14189_end_0 = const()[name = tensor<string, []>("op_14189_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_14189_end_mask_0 = const()[name = tensor<string, []>("op_14189_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14189_cast = slice_by_index(begin = var_14189_begin_0, end = var_14189_end_0, end_mask = var_14189_end_mask_0, x = var_14149_cast)[name = tensor<string, []>("op_14189_cast")];
            tensor<int32, [4]> var_14190_begin_0 = const()[name = tensor<string, []>("op_14190_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_14190_end_0 = const()[name = tensor<string, []>("op_14190_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_14190_end_mask_0 = const()[name = tensor<string, []>("op_14190_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14190_cast = slice_by_index(begin = var_14190_begin_0, end = var_14190_end_0, end_mask = var_14190_end_mask_0, x = var_14149_cast)[name = tensor<string, []>("op_14190_cast")];
            tensor<int32, [4]> var_14191_begin_0 = const()[name = tensor<string, []>("op_14191_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_14191_end_0 = const()[name = tensor<string, []>("op_14191_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_14191_end_mask_0 = const()[name = tensor<string, []>("op_14191_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14191_cast = slice_by_index(begin = var_14191_begin_0, end = var_14191_end_0, end_mask = var_14191_end_mask_0, x = var_14149_cast)[name = tensor<string, []>("op_14191_cast")];
            tensor<int32, [4]> var_14192_begin_0 = const()[name = tensor<string, []>("op_14192_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14192_end_0 = const()[name = tensor<string, []>("op_14192_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_14192_end_mask_0 = const()[name = tensor<string, []>("op_14192_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14192_cast = slice_by_index(begin = var_14192_begin_0, end = var_14192_end_0, end_mask = var_14192_end_mask_0, x = var_14153_cast)[name = tensor<string, []>("op_14192_cast")];
            tensor<int32, [4]> var_14193_begin_0 = const()[name = tensor<string, []>("op_14193_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_14193_end_0 = const()[name = tensor<string, []>("op_14193_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_14193_end_mask_0 = const()[name = tensor<string, []>("op_14193_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14193_cast = slice_by_index(begin = var_14193_begin_0, end = var_14193_end_0, end_mask = var_14193_end_mask_0, x = var_14153_cast)[name = tensor<string, []>("op_14193_cast")];
            tensor<int32, [4]> var_14194_begin_0 = const()[name = tensor<string, []>("op_14194_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_14194_end_0 = const()[name = tensor<string, []>("op_14194_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_14194_end_mask_0 = const()[name = tensor<string, []>("op_14194_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14194_cast = slice_by_index(begin = var_14194_begin_0, end = var_14194_end_0, end_mask = var_14194_end_mask_0, x = var_14153_cast)[name = tensor<string, []>("op_14194_cast")];
            tensor<int32, [4]> var_14195_begin_0 = const()[name = tensor<string, []>("op_14195_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_14195_end_0 = const()[name = tensor<string, []>("op_14195_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_14195_end_mask_0 = const()[name = tensor<string, []>("op_14195_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14195_cast = slice_by_index(begin = var_14195_begin_0, end = var_14195_end_0, end_mask = var_14195_end_mask_0, x = var_14153_cast)[name = tensor<string, []>("op_14195_cast")];
            tensor<int32, [4]> var_14196_begin_0 = const()[name = tensor<string, []>("op_14196_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_14196_end_0 = const()[name = tensor<string, []>("op_14196_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_14196_end_mask_0 = const()[name = tensor<string, []>("op_14196_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14196_cast = slice_by_index(begin = var_14196_begin_0, end = var_14196_end_0, end_mask = var_14196_end_mask_0, x = var_14153_cast)[name = tensor<string, []>("op_14196_cast")];
            tensor<int32, [4]> var_14197_begin_0 = const()[name = tensor<string, []>("op_14197_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_14197_end_0 = const()[name = tensor<string, []>("op_14197_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_14197_end_mask_0 = const()[name = tensor<string, []>("op_14197_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14197_cast = slice_by_index(begin = var_14197_begin_0, end = var_14197_end_0, end_mask = var_14197_end_mask_0, x = var_14153_cast)[name = tensor<string, []>("op_14197_cast")];
            tensor<int32, [4]> var_14198_begin_0 = const()[name = tensor<string, []>("op_14198_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_14198_end_0 = const()[name = tensor<string, []>("op_14198_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_14198_end_mask_0 = const()[name = tensor<string, []>("op_14198_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14198_cast = slice_by_index(begin = var_14198_begin_0, end = var_14198_end_0, end_mask = var_14198_end_mask_0, x = var_14153_cast)[name = tensor<string, []>("op_14198_cast")];
            tensor<int32, [4]> var_14199_begin_0 = const()[name = tensor<string, []>("op_14199_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_14199_end_0 = const()[name = tensor<string, []>("op_14199_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_14199_end_mask_0 = const()[name = tensor<string, []>("op_14199_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14199_cast = slice_by_index(begin = var_14199_begin_0, end = var_14199_end_0, end_mask = var_14199_end_mask_0, x = var_14153_cast)[name = tensor<string, []>("op_14199_cast")];
            tensor<int32, [4]> var_14200_begin_0 = const()[name = tensor<string, []>("op_14200_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14200_end_0 = const()[name = tensor<string, []>("op_14200_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_14200_end_mask_0 = const()[name = tensor<string, []>("op_14200_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14200_cast = slice_by_index(begin = var_14200_begin_0, end = var_14200_end_0, end_mask = var_14200_end_mask_0, x = var_14157_cast)[name = tensor<string, []>("op_14200_cast")];
            tensor<int32, [4]> var_14201_begin_0 = const()[name = tensor<string, []>("op_14201_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_14201_end_0 = const()[name = tensor<string, []>("op_14201_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_14201_end_mask_0 = const()[name = tensor<string, []>("op_14201_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14201_cast = slice_by_index(begin = var_14201_begin_0, end = var_14201_end_0, end_mask = var_14201_end_mask_0, x = var_14157_cast)[name = tensor<string, []>("op_14201_cast")];
            tensor<int32, [4]> var_14202_begin_0 = const()[name = tensor<string, []>("op_14202_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_14202_end_0 = const()[name = tensor<string, []>("op_14202_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_14202_end_mask_0 = const()[name = tensor<string, []>("op_14202_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14202_cast = slice_by_index(begin = var_14202_begin_0, end = var_14202_end_0, end_mask = var_14202_end_mask_0, x = var_14157_cast)[name = tensor<string, []>("op_14202_cast")];
            tensor<int32, [4]> var_14203_begin_0 = const()[name = tensor<string, []>("op_14203_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_14203_end_0 = const()[name = tensor<string, []>("op_14203_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_14203_end_mask_0 = const()[name = tensor<string, []>("op_14203_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14203_cast = slice_by_index(begin = var_14203_begin_0, end = var_14203_end_0, end_mask = var_14203_end_mask_0, x = var_14157_cast)[name = tensor<string, []>("op_14203_cast")];
            tensor<int32, [4]> var_14204_begin_0 = const()[name = tensor<string, []>("op_14204_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_14204_end_0 = const()[name = tensor<string, []>("op_14204_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_14204_end_mask_0 = const()[name = tensor<string, []>("op_14204_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14204_cast = slice_by_index(begin = var_14204_begin_0, end = var_14204_end_0, end_mask = var_14204_end_mask_0, x = var_14157_cast)[name = tensor<string, []>("op_14204_cast")];
            tensor<int32, [4]> var_14205_begin_0 = const()[name = tensor<string, []>("op_14205_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_14205_end_0 = const()[name = tensor<string, []>("op_14205_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_14205_end_mask_0 = const()[name = tensor<string, []>("op_14205_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14205_cast = slice_by_index(begin = var_14205_begin_0, end = var_14205_end_0, end_mask = var_14205_end_mask_0, x = var_14157_cast)[name = tensor<string, []>("op_14205_cast")];
            tensor<int32, [4]> var_14206_begin_0 = const()[name = tensor<string, []>("op_14206_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_14206_end_0 = const()[name = tensor<string, []>("op_14206_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_14206_end_mask_0 = const()[name = tensor<string, []>("op_14206_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14206_cast = slice_by_index(begin = var_14206_begin_0, end = var_14206_end_0, end_mask = var_14206_end_mask_0, x = var_14157_cast)[name = tensor<string, []>("op_14206_cast")];
            tensor<int32, [4]> var_14207_begin_0 = const()[name = tensor<string, []>("op_14207_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_14207_end_0 = const()[name = tensor<string, []>("op_14207_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_14207_end_mask_0 = const()[name = tensor<string, []>("op_14207_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14207_cast = slice_by_index(begin = var_14207_begin_0, end = var_14207_end_0, end_mask = var_14207_end_mask_0, x = var_14157_cast)[name = tensor<string, []>("op_14207_cast")];
            tensor<int32, [4]> var_14208_begin_0 = const()[name = tensor<string, []>("op_14208_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14208_end_0 = const()[name = tensor<string, []>("op_14208_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_14208_end_mask_0 = const()[name = tensor<string, []>("op_14208_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14208_cast = slice_by_index(begin = var_14208_begin_0, end = var_14208_end_0, end_mask = var_14208_end_mask_0, x = var_14161_cast)[name = tensor<string, []>("op_14208_cast")];
            tensor<int32, [4]> var_14209_begin_0 = const()[name = tensor<string, []>("op_14209_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_14209_end_0 = const()[name = tensor<string, []>("op_14209_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_14209_end_mask_0 = const()[name = tensor<string, []>("op_14209_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14209_cast = slice_by_index(begin = var_14209_begin_0, end = var_14209_end_0, end_mask = var_14209_end_mask_0, x = var_14161_cast)[name = tensor<string, []>("op_14209_cast")];
            tensor<int32, [4]> var_14210_begin_0 = const()[name = tensor<string, []>("op_14210_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_14210_end_0 = const()[name = tensor<string, []>("op_14210_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_14210_end_mask_0 = const()[name = tensor<string, []>("op_14210_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14210_cast = slice_by_index(begin = var_14210_begin_0, end = var_14210_end_0, end_mask = var_14210_end_mask_0, x = var_14161_cast)[name = tensor<string, []>("op_14210_cast")];
            tensor<int32, [4]> var_14211_begin_0 = const()[name = tensor<string, []>("op_14211_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_14211_end_0 = const()[name = tensor<string, []>("op_14211_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_14211_end_mask_0 = const()[name = tensor<string, []>("op_14211_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14211_cast = slice_by_index(begin = var_14211_begin_0, end = var_14211_end_0, end_mask = var_14211_end_mask_0, x = var_14161_cast)[name = tensor<string, []>("op_14211_cast")];
            tensor<int32, [4]> var_14212_begin_0 = const()[name = tensor<string, []>("op_14212_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_14212_end_0 = const()[name = tensor<string, []>("op_14212_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_14212_end_mask_0 = const()[name = tensor<string, []>("op_14212_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14212_cast = slice_by_index(begin = var_14212_begin_0, end = var_14212_end_0, end_mask = var_14212_end_mask_0, x = var_14161_cast)[name = tensor<string, []>("op_14212_cast")];
            tensor<int32, [4]> var_14213_begin_0 = const()[name = tensor<string, []>("op_14213_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_14213_end_0 = const()[name = tensor<string, []>("op_14213_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_14213_end_mask_0 = const()[name = tensor<string, []>("op_14213_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14213_cast = slice_by_index(begin = var_14213_begin_0, end = var_14213_end_0, end_mask = var_14213_end_mask_0, x = var_14161_cast)[name = tensor<string, []>("op_14213_cast")];
            tensor<int32, [4]> var_14214_begin_0 = const()[name = tensor<string, []>("op_14214_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_14214_end_0 = const()[name = tensor<string, []>("op_14214_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_14214_end_mask_0 = const()[name = tensor<string, []>("op_14214_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14214_cast = slice_by_index(begin = var_14214_begin_0, end = var_14214_end_0, end_mask = var_14214_end_mask_0, x = var_14161_cast)[name = tensor<string, []>("op_14214_cast")];
            tensor<int32, [4]> var_14215_begin_0 = const()[name = tensor<string, []>("op_14215_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_14215_end_0 = const()[name = tensor<string, []>("op_14215_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_14215_end_mask_0 = const()[name = tensor<string, []>("op_14215_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14215_cast = slice_by_index(begin = var_14215_begin_0, end = var_14215_end_0, end_mask = var_14215_end_mask_0, x = var_14161_cast)[name = tensor<string, []>("op_14215_cast")];
            tensor<int32, [4]> var_14216_begin_0 = const()[name = tensor<string, []>("op_14216_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14216_end_0 = const()[name = tensor<string, []>("op_14216_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_14216_end_mask_0 = const()[name = tensor<string, []>("op_14216_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14216_cast = slice_by_index(begin = var_14216_begin_0, end = var_14216_end_0, end_mask = var_14216_end_mask_0, x = var_14165_cast)[name = tensor<string, []>("op_14216_cast")];
            tensor<int32, [4]> var_14217_begin_0 = const()[name = tensor<string, []>("op_14217_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_14217_end_0 = const()[name = tensor<string, []>("op_14217_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_14217_end_mask_0 = const()[name = tensor<string, []>("op_14217_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14217_cast = slice_by_index(begin = var_14217_begin_0, end = var_14217_end_0, end_mask = var_14217_end_mask_0, x = var_14165_cast)[name = tensor<string, []>("op_14217_cast")];
            tensor<int32, [4]> var_14218_begin_0 = const()[name = tensor<string, []>("op_14218_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_14218_end_0 = const()[name = tensor<string, []>("op_14218_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_14218_end_mask_0 = const()[name = tensor<string, []>("op_14218_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14218_cast = slice_by_index(begin = var_14218_begin_0, end = var_14218_end_0, end_mask = var_14218_end_mask_0, x = var_14165_cast)[name = tensor<string, []>("op_14218_cast")];
            tensor<int32, [4]> var_14219_begin_0 = const()[name = tensor<string, []>("op_14219_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_14219_end_0 = const()[name = tensor<string, []>("op_14219_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_14219_end_mask_0 = const()[name = tensor<string, []>("op_14219_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14219_cast = slice_by_index(begin = var_14219_begin_0, end = var_14219_end_0, end_mask = var_14219_end_mask_0, x = var_14165_cast)[name = tensor<string, []>("op_14219_cast")];
            tensor<int32, [4]> var_14220_begin_0 = const()[name = tensor<string, []>("op_14220_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_14220_end_0 = const()[name = tensor<string, []>("op_14220_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_14220_end_mask_0 = const()[name = tensor<string, []>("op_14220_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14220_cast = slice_by_index(begin = var_14220_begin_0, end = var_14220_end_0, end_mask = var_14220_end_mask_0, x = var_14165_cast)[name = tensor<string, []>("op_14220_cast")];
            tensor<int32, [4]> var_14221_begin_0 = const()[name = tensor<string, []>("op_14221_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_14221_end_0 = const()[name = tensor<string, []>("op_14221_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_14221_end_mask_0 = const()[name = tensor<string, []>("op_14221_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14221_cast = slice_by_index(begin = var_14221_begin_0, end = var_14221_end_0, end_mask = var_14221_end_mask_0, x = var_14165_cast)[name = tensor<string, []>("op_14221_cast")];
            tensor<int32, [4]> var_14222_begin_0 = const()[name = tensor<string, []>("op_14222_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_14222_end_0 = const()[name = tensor<string, []>("op_14222_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_14222_end_mask_0 = const()[name = tensor<string, []>("op_14222_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14222_cast = slice_by_index(begin = var_14222_begin_0, end = var_14222_end_0, end_mask = var_14222_end_mask_0, x = var_14165_cast)[name = tensor<string, []>("op_14222_cast")];
            tensor<int32, [4]> var_14223_begin_0 = const()[name = tensor<string, []>("op_14223_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_14223_end_0 = const()[name = tensor<string, []>("op_14223_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_14223_end_mask_0 = const()[name = tensor<string, []>("op_14223_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14223_cast = slice_by_index(begin = var_14223_begin_0, end = var_14223_end_0, end_mask = var_14223_end_mask_0, x = var_14165_cast)[name = tensor<string, []>("op_14223_cast")];
            tensor<int32, [4]> var_14224_begin_0 = const()[name = tensor<string, []>("op_14224_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14224_end_0 = const()[name = tensor<string, []>("op_14224_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_14224_end_mask_0 = const()[name = tensor<string, []>("op_14224_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14224_cast = slice_by_index(begin = var_14224_begin_0, end = var_14224_end_0, end_mask = var_14224_end_mask_0, x = var_14169_cast)[name = tensor<string, []>("op_14224_cast")];
            tensor<int32, [4]> var_14225_begin_0 = const()[name = tensor<string, []>("op_14225_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_14225_end_0 = const()[name = tensor<string, []>("op_14225_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_14225_end_mask_0 = const()[name = tensor<string, []>("op_14225_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14225_cast = slice_by_index(begin = var_14225_begin_0, end = var_14225_end_0, end_mask = var_14225_end_mask_0, x = var_14169_cast)[name = tensor<string, []>("op_14225_cast")];
            tensor<int32, [4]> var_14226_begin_0 = const()[name = tensor<string, []>("op_14226_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_14226_end_0 = const()[name = tensor<string, []>("op_14226_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_14226_end_mask_0 = const()[name = tensor<string, []>("op_14226_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14226_cast = slice_by_index(begin = var_14226_begin_0, end = var_14226_end_0, end_mask = var_14226_end_mask_0, x = var_14169_cast)[name = tensor<string, []>("op_14226_cast")];
            tensor<int32, [4]> var_14227_begin_0 = const()[name = tensor<string, []>("op_14227_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_14227_end_0 = const()[name = tensor<string, []>("op_14227_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_14227_end_mask_0 = const()[name = tensor<string, []>("op_14227_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14227_cast = slice_by_index(begin = var_14227_begin_0, end = var_14227_end_0, end_mask = var_14227_end_mask_0, x = var_14169_cast)[name = tensor<string, []>("op_14227_cast")];
            tensor<int32, [4]> var_14228_begin_0 = const()[name = tensor<string, []>("op_14228_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_14228_end_0 = const()[name = tensor<string, []>("op_14228_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_14228_end_mask_0 = const()[name = tensor<string, []>("op_14228_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14228_cast = slice_by_index(begin = var_14228_begin_0, end = var_14228_end_0, end_mask = var_14228_end_mask_0, x = var_14169_cast)[name = tensor<string, []>("op_14228_cast")];
            tensor<int32, [4]> var_14229_begin_0 = const()[name = tensor<string, []>("op_14229_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_14229_end_0 = const()[name = tensor<string, []>("op_14229_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_14229_end_mask_0 = const()[name = tensor<string, []>("op_14229_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14229_cast = slice_by_index(begin = var_14229_begin_0, end = var_14229_end_0, end_mask = var_14229_end_mask_0, x = var_14169_cast)[name = tensor<string, []>("op_14229_cast")];
            tensor<int32, [4]> var_14230_begin_0 = const()[name = tensor<string, []>("op_14230_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_14230_end_0 = const()[name = tensor<string, []>("op_14230_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_14230_end_mask_0 = const()[name = tensor<string, []>("op_14230_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14230_cast = slice_by_index(begin = var_14230_begin_0, end = var_14230_end_0, end_mask = var_14230_end_mask_0, x = var_14169_cast)[name = tensor<string, []>("op_14230_cast")];
            tensor<int32, [4]> var_14231_begin_0 = const()[name = tensor<string, []>("op_14231_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_14231_end_0 = const()[name = tensor<string, []>("op_14231_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_14231_end_mask_0 = const()[name = tensor<string, []>("op_14231_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14231_cast = slice_by_index(begin = var_14231_begin_0, end = var_14231_end_0, end_mask = var_14231_end_mask_0, x = var_14169_cast)[name = tensor<string, []>("op_14231_cast")];
            tensor<int32, [4]> var_14232_begin_0 = const()[name = tensor<string, []>("op_14232_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14232_end_0 = const()[name = tensor<string, []>("op_14232_end_0"), val = tensor<int32, [4]>([2, 40, 1, 512])];
            tensor<bool, [4]> var_14232_end_mask_0 = const()[name = tensor<string, []>("op_14232_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14232_cast = slice_by_index(begin = var_14232_begin_0, end = var_14232_end_0, end_mask = var_14232_end_mask_0, x = var_14173_cast)[name = tensor<string, []>("op_14232_cast")];
            tensor<int32, [4]> var_14233_begin_0 = const()[name = tensor<string, []>("op_14233_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_14233_end_0 = const()[name = tensor<string, []>("op_14233_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1024])];
            tensor<bool, [4]> var_14233_end_mask_0 = const()[name = tensor<string, []>("op_14233_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14233_cast = slice_by_index(begin = var_14233_begin_0, end = var_14233_end_0, end_mask = var_14233_end_mask_0, x = var_14173_cast)[name = tensor<string, []>("op_14233_cast")];
            tensor<int32, [4]> var_14234_begin_0 = const()[name = tensor<string, []>("op_14234_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_14234_end_0 = const()[name = tensor<string, []>("op_14234_end_0"), val = tensor<int32, [4]>([2, 40, 1, 1536])];
            tensor<bool, [4]> var_14234_end_mask_0 = const()[name = tensor<string, []>("op_14234_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14234_cast = slice_by_index(begin = var_14234_begin_0, end = var_14234_end_0, end_mask = var_14234_end_mask_0, x = var_14173_cast)[name = tensor<string, []>("op_14234_cast")];
            tensor<int32, [4]> var_14235_begin_0 = const()[name = tensor<string, []>("op_14235_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_14235_end_0 = const()[name = tensor<string, []>("op_14235_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2048])];
            tensor<bool, [4]> var_14235_end_mask_0 = const()[name = tensor<string, []>("op_14235_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14235_cast = slice_by_index(begin = var_14235_begin_0, end = var_14235_end_0, end_mask = var_14235_end_mask_0, x = var_14173_cast)[name = tensor<string, []>("op_14235_cast")];
            tensor<int32, [4]> var_14236_begin_0 = const()[name = tensor<string, []>("op_14236_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_14236_end_0 = const()[name = tensor<string, []>("op_14236_end_0"), val = tensor<int32, [4]>([2, 40, 1, 2560])];
            tensor<bool, [4]> var_14236_end_mask_0 = const()[name = tensor<string, []>("op_14236_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14236_cast = slice_by_index(begin = var_14236_begin_0, end = var_14236_end_0, end_mask = var_14236_end_mask_0, x = var_14173_cast)[name = tensor<string, []>("op_14236_cast")];
            tensor<int32, [4]> var_14237_begin_0 = const()[name = tensor<string, []>("op_14237_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_14237_end_0 = const()[name = tensor<string, []>("op_14237_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3072])];
            tensor<bool, [4]> var_14237_end_mask_0 = const()[name = tensor<string, []>("op_14237_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14237_cast = slice_by_index(begin = var_14237_begin_0, end = var_14237_end_0, end_mask = var_14237_end_mask_0, x = var_14173_cast)[name = tensor<string, []>("op_14237_cast")];
            tensor<int32, [4]> var_14238_begin_0 = const()[name = tensor<string, []>("op_14238_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_14238_end_0 = const()[name = tensor<string, []>("op_14238_end_0"), val = tensor<int32, [4]>([2, 40, 1, 3584])];
            tensor<bool, [4]> var_14238_end_mask_0 = const()[name = tensor<string, []>("op_14238_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14238_cast = slice_by_index(begin = var_14238_begin_0, end = var_14238_end_0, end_mask = var_14238_end_mask_0, x = var_14173_cast)[name = tensor<string, []>("op_14238_cast")];
            tensor<int32, [4]> var_14239_begin_0 = const()[name = tensor<string, []>("op_14239_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_14239_end_0 = const()[name = tensor<string, []>("op_14239_end_0"), val = tensor<int32, [4]>([2, 40, 1, 4096])];
            tensor<bool, [4]> var_14239_end_mask_0 = const()[name = tensor<string, []>("op_14239_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 40, 1, 512]> var_14239_cast = slice_by_index(begin = var_14239_begin_0, end = var_14239_end_0, end_mask = var_14239_end_mask_0, x = var_14173_cast)[name = tensor<string, []>("op_14239_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_14244_begin_0 = const()[name = tensor<string, []>("op_14244_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14244_end_0 = const()[name = tensor<string, []>("op_14244_end_0"), val = tensor<int32, [4]>([2, 77, 1, 40])];
            tensor<bool, [4]> var_14244_end_mask_0 = const()[name = tensor<string, []>("op_14244_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)[name = tensor<string, []>("transpose_0")];
            tensor<fp16, [2, 77, 1, 40]> var_14244_cast = slice_by_index(begin = var_14244_begin_0, end = var_14244_end_0, end_mask = var_14244_end_mask_0, x = transpose_0)[name = tensor<string, []>("op_14244_cast")];
            tensor<int32, [4]> var_14248_begin_0 = const()[name = tensor<string, []>("op_14248_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 40])];
            tensor<int32, [4]> var_14248_end_0 = const()[name = tensor<string, []>("op_14248_end_0"), val = tensor<int32, [4]>([2, 77, 1, 80])];
            tensor<bool, [4]> var_14248_end_mask_0 = const()[name = tensor<string, []>("op_14248_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 40]> var_14248_cast = slice_by_index(begin = var_14248_begin_0, end = var_14248_end_0, end_mask = var_14248_end_mask_0, x = transpose_0)[name = tensor<string, []>("op_14248_cast")];
            tensor<int32, [4]> var_14252_begin_0 = const()[name = tensor<string, []>("op_14252_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 80])];
            tensor<int32, [4]> var_14252_end_0 = const()[name = tensor<string, []>("op_14252_end_0"), val = tensor<int32, [4]>([2, 77, 1, 120])];
            tensor<bool, [4]> var_14252_end_mask_0 = const()[name = tensor<string, []>("op_14252_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 40]> var_14252_cast = slice_by_index(begin = var_14252_begin_0, end = var_14252_end_0, end_mask = var_14252_end_mask_0, x = transpose_0)[name = tensor<string, []>("op_14252_cast")];
            tensor<int32, [4]> var_14256_begin_0 = const()[name = tensor<string, []>("op_14256_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 120])];
            tensor<int32, [4]> var_14256_end_0 = const()[name = tensor<string, []>("op_14256_end_0"), val = tensor<int32, [4]>([2, 77, 1, 160])];
            tensor<bool, [4]> var_14256_end_mask_0 = const()[name = tensor<string, []>("op_14256_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 40]> var_14256_cast = slice_by_index(begin = var_14256_begin_0, end = var_14256_end_0, end_mask = var_14256_end_mask_0, x = transpose_0)[name = tensor<string, []>("op_14256_cast")];
            tensor<int32, [4]> var_14260_begin_0 = const()[name = tensor<string, []>("op_14260_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 160])];
            tensor<int32, [4]> var_14260_end_0 = const()[name = tensor<string, []>("op_14260_end_0"), val = tensor<int32, [4]>([2, 77, 1, 200])];
            tensor<bool, [4]> var_14260_end_mask_0 = const()[name = tensor<string, []>("op_14260_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 40]> var_14260_cast = slice_by_index(begin = var_14260_begin_0, end = var_14260_end_0, end_mask = var_14260_end_mask_0, x = transpose_0)[name = tensor<string, []>("op_14260_cast")];
            tensor<int32, [4]> var_14264_begin_0 = const()[name = tensor<string, []>("op_14264_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 200])];
            tensor<int32, [4]> var_14264_end_0 = const()[name = tensor<string, []>("op_14264_end_0"), val = tensor<int32, [4]>([2, 77, 1, 240])];
            tensor<bool, [4]> var_14264_end_mask_0 = const()[name = tensor<string, []>("op_14264_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 40]> var_14264_cast = slice_by_index(begin = var_14264_begin_0, end = var_14264_end_0, end_mask = var_14264_end_mask_0, x = transpose_0)[name = tensor<string, []>("op_14264_cast")];
            tensor<int32, [4]> var_14268_begin_0 = const()[name = tensor<string, []>("op_14268_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 240])];
            tensor<int32, [4]> var_14268_end_0 = const()[name = tensor<string, []>("op_14268_end_0"), val = tensor<int32, [4]>([2, 77, 1, 280])];
            tensor<bool, [4]> var_14268_end_mask_0 = const()[name = tensor<string, []>("op_14268_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 40]> var_14268_cast = slice_by_index(begin = var_14268_begin_0, end = var_14268_end_0, end_mask = var_14268_end_mask_0, x = transpose_0)[name = tensor<string, []>("op_14268_cast")];
            tensor<int32, [4]> var_14272_begin_0 = const()[name = tensor<string, []>("op_14272_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 280])];
            tensor<int32, [4]> var_14272_end_0 = const()[name = tensor<string, []>("op_14272_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_14272_end_mask_0 = const()[name = tensor<string, []>("op_14272_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 40]> var_14272_cast = slice_by_index(begin = var_14272_begin_0, end = var_14272_end_0, end_mask = var_14272_end_mask_0, x = transpose_0)[name = tensor<string, []>("op_14272_cast")];
            tensor<int32, [4]> var_14274_begin_0 = const()[name = tensor<string, []>("op_14274_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14274_end_0 = const()[name = tensor<string, []>("op_14274_end_0"), val = tensor<int32, [4]>([2, 40, 1, 77])];
            tensor<bool, [4]> var_14274_end_mask_0 = const()[name = tensor<string, []>("op_14274_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_14274_cast = slice_by_index(begin = var_14274_begin_0, end = var_14274_end_0, end_mask = var_14274_end_mask_0, x = v_cast)[name = tensor<string, []>("op_14274_cast")];
            tensor<int32, [4]> var_14278_begin_0 = const()[name = tensor<string, []>("op_14278_begin_0"), val = tensor<int32, [4]>([0, 40, 0, 0])];
            tensor<int32, [4]> var_14278_end_0 = const()[name = tensor<string, []>("op_14278_end_0"), val = tensor<int32, [4]>([2, 80, 1, 77])];
            tensor<bool, [4]> var_14278_end_mask_0 = const()[name = tensor<string, []>("op_14278_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_14278_cast = slice_by_index(begin = var_14278_begin_0, end = var_14278_end_0, end_mask = var_14278_end_mask_0, x = v_cast)[name = tensor<string, []>("op_14278_cast")];
            tensor<int32, [4]> var_14282_begin_0 = const()[name = tensor<string, []>("op_14282_begin_0"), val = tensor<int32, [4]>([0, 80, 0, 0])];
            tensor<int32, [4]> var_14282_end_0 = const()[name = tensor<string, []>("op_14282_end_0"), val = tensor<int32, [4]>([2, 120, 1, 77])];
            tensor<bool, [4]> var_14282_end_mask_0 = const()[name = tensor<string, []>("op_14282_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_14282_cast = slice_by_index(begin = var_14282_begin_0, end = var_14282_end_0, end_mask = var_14282_end_mask_0, x = v_cast)[name = tensor<string, []>("op_14282_cast")];
            tensor<int32, [4]> var_14286_begin_0 = const()[name = tensor<string, []>("op_14286_begin_0"), val = tensor<int32, [4]>([0, 120, 0, 0])];
            tensor<int32, [4]> var_14286_end_0 = const()[name = tensor<string, []>("op_14286_end_0"), val = tensor<int32, [4]>([2, 160, 1, 77])];
            tensor<bool, [4]> var_14286_end_mask_0 = const()[name = tensor<string, []>("op_14286_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_14286_cast = slice_by_index(begin = var_14286_begin_0, end = var_14286_end_0, end_mask = var_14286_end_mask_0, x = v_cast)[name = tensor<string, []>("op_14286_cast")];
            tensor<int32, [4]> var_14290_begin_0 = const()[name = tensor<string, []>("op_14290_begin_0"), val = tensor<int32, [4]>([0, 160, 0, 0])];
            tensor<int32, [4]> var_14290_end_0 = const()[name = tensor<string, []>("op_14290_end_0"), val = tensor<int32, [4]>([2, 200, 1, 77])];
            tensor<bool, [4]> var_14290_end_mask_0 = const()[name = tensor<string, []>("op_14290_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_14290_cast = slice_by_index(begin = var_14290_begin_0, end = var_14290_end_0, end_mask = var_14290_end_mask_0, x = v_cast)[name = tensor<string, []>("op_14290_cast")];
            tensor<int32, [4]> var_14294_begin_0 = const()[name = tensor<string, []>("op_14294_begin_0"), val = tensor<int32, [4]>([0, 200, 0, 0])];
            tensor<int32, [4]> var_14294_end_0 = const()[name = tensor<string, []>("op_14294_end_0"), val = tensor<int32, [4]>([2, 240, 1, 77])];
            tensor<bool, [4]> var_14294_end_mask_0 = const()[name = tensor<string, []>("op_14294_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_14294_cast = slice_by_index(begin = var_14294_begin_0, end = var_14294_end_0, end_mask = var_14294_end_mask_0, x = v_cast)[name = tensor<string, []>("op_14294_cast")];
            tensor<int32, [4]> var_14298_begin_0 = const()[name = tensor<string, []>("op_14298_begin_0"), val = tensor<int32, [4]>([0, 240, 0, 0])];
            tensor<int32, [4]> var_14298_end_0 = const()[name = tensor<string, []>("op_14298_end_0"), val = tensor<int32, [4]>([2, 280, 1, 77])];
            tensor<bool, [4]> var_14298_end_mask_0 = const()[name = tensor<string, []>("op_14298_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_14298_cast = slice_by_index(begin = var_14298_begin_0, end = var_14298_end_0, end_mask = var_14298_end_mask_0, x = v_cast)[name = tensor<string, []>("op_14298_cast")];
            tensor<int32, [4]> var_14302_begin_0 = const()[name = tensor<string, []>("op_14302_begin_0"), val = tensor<int32, [4]>([0, 280, 0, 0])];
            tensor<int32, [4]> var_14302_end_0 = const()[name = tensor<string, []>("op_14302_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_14302_end_mask_0 = const()[name = tensor<string, []>("op_14302_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 40, 1, 77]> var_14302_cast = slice_by_index(begin = var_14302_begin_0, end = var_14302_end_0, end_mask = var_14302_end_mask_0, x = v_cast)[name = tensor<string, []>("op_14302_cast")];
            tensor<string, []> var_14306_equation_0 = const()[name = tensor<string, []>("op_14306_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14306_cast = einsum(equation = var_14306_equation_0, values = (var_14244_cast, var_14176_cast))[name = tensor<string, []>("op_14306_cast")];
            tensor<fp16, []> var_14307_to_fp16 = const()[name = tensor<string, []>("op_14307_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1473_cast = mul(x = var_14306_cast, y = var_14307_to_fp16)[name = tensor<string, []>("aw_chunk_1473_cast")];
            tensor<string, []> var_14310_equation_0 = const()[name = tensor<string, []>("op_14310_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14310_cast = einsum(equation = var_14310_equation_0, values = (var_14244_cast, var_14177_cast))[name = tensor<string, []>("op_14310_cast")];
            tensor<fp16, []> var_14311_to_fp16 = const()[name = tensor<string, []>("op_14311_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1475_cast = mul(x = var_14310_cast, y = var_14311_to_fp16)[name = tensor<string, []>("aw_chunk_1475_cast")];
            tensor<string, []> var_14314_equation_0 = const()[name = tensor<string, []>("op_14314_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14314_cast = einsum(equation = var_14314_equation_0, values = (var_14244_cast, var_14178_cast))[name = tensor<string, []>("op_14314_cast")];
            tensor<fp16, []> var_14315_to_fp16 = const()[name = tensor<string, []>("op_14315_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1477_cast = mul(x = var_14314_cast, y = var_14315_to_fp16)[name = tensor<string, []>("aw_chunk_1477_cast")];
            tensor<string, []> var_14318_equation_0 = const()[name = tensor<string, []>("op_14318_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14318_cast = einsum(equation = var_14318_equation_0, values = (var_14244_cast, var_14179_cast))[name = tensor<string, []>("op_14318_cast")];
            tensor<fp16, []> var_14319_to_fp16 = const()[name = tensor<string, []>("op_14319_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1479_cast = mul(x = var_14318_cast, y = var_14319_to_fp16)[name = tensor<string, []>("aw_chunk_1479_cast")];
            tensor<string, []> var_14322_equation_0 = const()[name = tensor<string, []>("op_14322_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14322_cast = einsum(equation = var_14322_equation_0, values = (var_14244_cast, var_14180_cast))[name = tensor<string, []>("op_14322_cast")];
            tensor<fp16, []> var_14323_to_fp16 = const()[name = tensor<string, []>("op_14323_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1481_cast = mul(x = var_14322_cast, y = var_14323_to_fp16)[name = tensor<string, []>("aw_chunk_1481_cast")];
            tensor<string, []> var_14326_equation_0 = const()[name = tensor<string, []>("op_14326_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14326_cast = einsum(equation = var_14326_equation_0, values = (var_14244_cast, var_14181_cast))[name = tensor<string, []>("op_14326_cast")];
            tensor<fp16, []> var_14327_to_fp16 = const()[name = tensor<string, []>("op_14327_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1483_cast = mul(x = var_14326_cast, y = var_14327_to_fp16)[name = tensor<string, []>("aw_chunk_1483_cast")];
            tensor<string, []> var_14330_equation_0 = const()[name = tensor<string, []>("op_14330_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14330_cast = einsum(equation = var_14330_equation_0, values = (var_14244_cast, var_14182_cast))[name = tensor<string, []>("op_14330_cast")];
            tensor<fp16, []> var_14331_to_fp16 = const()[name = tensor<string, []>("op_14331_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1485_cast = mul(x = var_14330_cast, y = var_14331_to_fp16)[name = tensor<string, []>("aw_chunk_1485_cast")];
            tensor<string, []> var_14334_equation_0 = const()[name = tensor<string, []>("op_14334_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14334_cast = einsum(equation = var_14334_equation_0, values = (var_14244_cast, var_14183_cast))[name = tensor<string, []>("op_14334_cast")];
            tensor<fp16, []> var_14335_to_fp16 = const()[name = tensor<string, []>("op_14335_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1487_cast = mul(x = var_14334_cast, y = var_14335_to_fp16)[name = tensor<string, []>("aw_chunk_1487_cast")];
            tensor<string, []> var_14338_equation_0 = const()[name = tensor<string, []>("op_14338_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14338_cast = einsum(equation = var_14338_equation_0, values = (var_14248_cast, var_14184_cast))[name = tensor<string, []>("op_14338_cast")];
            tensor<fp16, []> var_14339_to_fp16 = const()[name = tensor<string, []>("op_14339_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1489_cast = mul(x = var_14338_cast, y = var_14339_to_fp16)[name = tensor<string, []>("aw_chunk_1489_cast")];
            tensor<string, []> var_14342_equation_0 = const()[name = tensor<string, []>("op_14342_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14342_cast = einsum(equation = var_14342_equation_0, values = (var_14248_cast, var_14185_cast))[name = tensor<string, []>("op_14342_cast")];
            tensor<fp16, []> var_14343_to_fp16 = const()[name = tensor<string, []>("op_14343_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1491_cast = mul(x = var_14342_cast, y = var_14343_to_fp16)[name = tensor<string, []>("aw_chunk_1491_cast")];
            tensor<string, []> var_14346_equation_0 = const()[name = tensor<string, []>("op_14346_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14346_cast = einsum(equation = var_14346_equation_0, values = (var_14248_cast, var_14186_cast))[name = tensor<string, []>("op_14346_cast")];
            tensor<fp16, []> var_14347_to_fp16 = const()[name = tensor<string, []>("op_14347_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1493_cast = mul(x = var_14346_cast, y = var_14347_to_fp16)[name = tensor<string, []>("aw_chunk_1493_cast")];
            tensor<string, []> var_14350_equation_0 = const()[name = tensor<string, []>("op_14350_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14350_cast = einsum(equation = var_14350_equation_0, values = (var_14248_cast, var_14187_cast))[name = tensor<string, []>("op_14350_cast")];
            tensor<fp16, []> var_14351_to_fp16 = const()[name = tensor<string, []>("op_14351_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1495_cast = mul(x = var_14350_cast, y = var_14351_to_fp16)[name = tensor<string, []>("aw_chunk_1495_cast")];
            tensor<string, []> var_14354_equation_0 = const()[name = tensor<string, []>("op_14354_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14354_cast = einsum(equation = var_14354_equation_0, values = (var_14248_cast, var_14188_cast))[name = tensor<string, []>("op_14354_cast")];
            tensor<fp16, []> var_14355_to_fp16 = const()[name = tensor<string, []>("op_14355_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1497_cast = mul(x = var_14354_cast, y = var_14355_to_fp16)[name = tensor<string, []>("aw_chunk_1497_cast")];
            tensor<string, []> var_14358_equation_0 = const()[name = tensor<string, []>("op_14358_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14358_cast = einsum(equation = var_14358_equation_0, values = (var_14248_cast, var_14189_cast))[name = tensor<string, []>("op_14358_cast")];
            tensor<fp16, []> var_14359_to_fp16 = const()[name = tensor<string, []>("op_14359_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1499_cast = mul(x = var_14358_cast, y = var_14359_to_fp16)[name = tensor<string, []>("aw_chunk_1499_cast")];
            tensor<string, []> var_14362_equation_0 = const()[name = tensor<string, []>("op_14362_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14362_cast = einsum(equation = var_14362_equation_0, values = (var_14248_cast, var_14190_cast))[name = tensor<string, []>("op_14362_cast")];
            tensor<fp16, []> var_14363_to_fp16 = const()[name = tensor<string, []>("op_14363_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1501_cast = mul(x = var_14362_cast, y = var_14363_to_fp16)[name = tensor<string, []>("aw_chunk_1501_cast")];
            tensor<string, []> var_14366_equation_0 = const()[name = tensor<string, []>("op_14366_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14366_cast = einsum(equation = var_14366_equation_0, values = (var_14248_cast, var_14191_cast))[name = tensor<string, []>("op_14366_cast")];
            tensor<fp16, []> var_14367_to_fp16 = const()[name = tensor<string, []>("op_14367_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1503_cast = mul(x = var_14366_cast, y = var_14367_to_fp16)[name = tensor<string, []>("aw_chunk_1503_cast")];
            tensor<string, []> var_14370_equation_0 = const()[name = tensor<string, []>("op_14370_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14370_cast = einsum(equation = var_14370_equation_0, values = (var_14252_cast, var_14192_cast))[name = tensor<string, []>("op_14370_cast")];
            tensor<fp16, []> var_14371_to_fp16 = const()[name = tensor<string, []>("op_14371_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1505_cast = mul(x = var_14370_cast, y = var_14371_to_fp16)[name = tensor<string, []>("aw_chunk_1505_cast")];
            tensor<string, []> var_14374_equation_0 = const()[name = tensor<string, []>("op_14374_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14374_cast = einsum(equation = var_14374_equation_0, values = (var_14252_cast, var_14193_cast))[name = tensor<string, []>("op_14374_cast")];
            tensor<fp16, []> var_14375_to_fp16 = const()[name = tensor<string, []>("op_14375_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1507_cast = mul(x = var_14374_cast, y = var_14375_to_fp16)[name = tensor<string, []>("aw_chunk_1507_cast")];
            tensor<string, []> var_14378_equation_0 = const()[name = tensor<string, []>("op_14378_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14378_cast = einsum(equation = var_14378_equation_0, values = (var_14252_cast, var_14194_cast))[name = tensor<string, []>("op_14378_cast")];
            tensor<fp16, []> var_14379_to_fp16 = const()[name = tensor<string, []>("op_14379_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1509_cast = mul(x = var_14378_cast, y = var_14379_to_fp16)[name = tensor<string, []>("aw_chunk_1509_cast")];
            tensor<string, []> var_14382_equation_0 = const()[name = tensor<string, []>("op_14382_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14382_cast = einsum(equation = var_14382_equation_0, values = (var_14252_cast, var_14195_cast))[name = tensor<string, []>("op_14382_cast")];
            tensor<fp16, []> var_14383_to_fp16 = const()[name = tensor<string, []>("op_14383_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1511_cast = mul(x = var_14382_cast, y = var_14383_to_fp16)[name = tensor<string, []>("aw_chunk_1511_cast")];
            tensor<string, []> var_14386_equation_0 = const()[name = tensor<string, []>("op_14386_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14386_cast = einsum(equation = var_14386_equation_0, values = (var_14252_cast, var_14196_cast))[name = tensor<string, []>("op_14386_cast")];
            tensor<fp16, []> var_14387_to_fp16 = const()[name = tensor<string, []>("op_14387_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1513_cast = mul(x = var_14386_cast, y = var_14387_to_fp16)[name = tensor<string, []>("aw_chunk_1513_cast")];
            tensor<string, []> var_14390_equation_0 = const()[name = tensor<string, []>("op_14390_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14390_cast = einsum(equation = var_14390_equation_0, values = (var_14252_cast, var_14197_cast))[name = tensor<string, []>("op_14390_cast")];
            tensor<fp16, []> var_14391_to_fp16 = const()[name = tensor<string, []>("op_14391_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1515_cast = mul(x = var_14390_cast, y = var_14391_to_fp16)[name = tensor<string, []>("aw_chunk_1515_cast")];
            tensor<string, []> var_14394_equation_0 = const()[name = tensor<string, []>("op_14394_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14394_cast = einsum(equation = var_14394_equation_0, values = (var_14252_cast, var_14198_cast))[name = tensor<string, []>("op_14394_cast")];
            tensor<fp16, []> var_14395_to_fp16 = const()[name = tensor<string, []>("op_14395_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1517_cast = mul(x = var_14394_cast, y = var_14395_to_fp16)[name = tensor<string, []>("aw_chunk_1517_cast")];
            tensor<string, []> var_14398_equation_0 = const()[name = tensor<string, []>("op_14398_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14398_cast = einsum(equation = var_14398_equation_0, values = (var_14252_cast, var_14199_cast))[name = tensor<string, []>("op_14398_cast")];
            tensor<fp16, []> var_14399_to_fp16 = const()[name = tensor<string, []>("op_14399_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1519_cast = mul(x = var_14398_cast, y = var_14399_to_fp16)[name = tensor<string, []>("aw_chunk_1519_cast")];
            tensor<string, []> var_14402_equation_0 = const()[name = tensor<string, []>("op_14402_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14402_cast = einsum(equation = var_14402_equation_0, values = (var_14256_cast, var_14200_cast))[name = tensor<string, []>("op_14402_cast")];
            tensor<fp16, []> var_14403_to_fp16 = const()[name = tensor<string, []>("op_14403_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1521_cast = mul(x = var_14402_cast, y = var_14403_to_fp16)[name = tensor<string, []>("aw_chunk_1521_cast")];
            tensor<string, []> var_14406_equation_0 = const()[name = tensor<string, []>("op_14406_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14406_cast = einsum(equation = var_14406_equation_0, values = (var_14256_cast, var_14201_cast))[name = tensor<string, []>("op_14406_cast")];
            tensor<fp16, []> var_14407_to_fp16 = const()[name = tensor<string, []>("op_14407_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1523_cast = mul(x = var_14406_cast, y = var_14407_to_fp16)[name = tensor<string, []>("aw_chunk_1523_cast")];
            tensor<string, []> var_14410_equation_0 = const()[name = tensor<string, []>("op_14410_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14410_cast = einsum(equation = var_14410_equation_0, values = (var_14256_cast, var_14202_cast))[name = tensor<string, []>("op_14410_cast")];
            tensor<fp16, []> var_14411_to_fp16 = const()[name = tensor<string, []>("op_14411_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1525_cast = mul(x = var_14410_cast, y = var_14411_to_fp16)[name = tensor<string, []>("aw_chunk_1525_cast")];
            tensor<string, []> var_14414_equation_0 = const()[name = tensor<string, []>("op_14414_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14414_cast = einsum(equation = var_14414_equation_0, values = (var_14256_cast, var_14203_cast))[name = tensor<string, []>("op_14414_cast")];
            tensor<fp16, []> var_14415_to_fp16 = const()[name = tensor<string, []>("op_14415_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1527_cast = mul(x = var_14414_cast, y = var_14415_to_fp16)[name = tensor<string, []>("aw_chunk_1527_cast")];
            tensor<string, []> var_14418_equation_0 = const()[name = tensor<string, []>("op_14418_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14418_cast = einsum(equation = var_14418_equation_0, values = (var_14256_cast, var_14204_cast))[name = tensor<string, []>("op_14418_cast")];
            tensor<fp16, []> var_14419_to_fp16 = const()[name = tensor<string, []>("op_14419_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1529_cast = mul(x = var_14418_cast, y = var_14419_to_fp16)[name = tensor<string, []>("aw_chunk_1529_cast")];
            tensor<string, []> var_14422_equation_0 = const()[name = tensor<string, []>("op_14422_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14422_cast = einsum(equation = var_14422_equation_0, values = (var_14256_cast, var_14205_cast))[name = tensor<string, []>("op_14422_cast")];
            tensor<fp16, []> var_14423_to_fp16 = const()[name = tensor<string, []>("op_14423_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1531_cast = mul(x = var_14422_cast, y = var_14423_to_fp16)[name = tensor<string, []>("aw_chunk_1531_cast")];
            tensor<string, []> var_14426_equation_0 = const()[name = tensor<string, []>("op_14426_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14426_cast = einsum(equation = var_14426_equation_0, values = (var_14256_cast, var_14206_cast))[name = tensor<string, []>("op_14426_cast")];
            tensor<fp16, []> var_14427_to_fp16 = const()[name = tensor<string, []>("op_14427_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1533_cast = mul(x = var_14426_cast, y = var_14427_to_fp16)[name = tensor<string, []>("aw_chunk_1533_cast")];
            tensor<string, []> var_14430_equation_0 = const()[name = tensor<string, []>("op_14430_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14430_cast = einsum(equation = var_14430_equation_0, values = (var_14256_cast, var_14207_cast))[name = tensor<string, []>("op_14430_cast")];
            tensor<fp16, []> var_14431_to_fp16 = const()[name = tensor<string, []>("op_14431_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1535_cast = mul(x = var_14430_cast, y = var_14431_to_fp16)[name = tensor<string, []>("aw_chunk_1535_cast")];
            tensor<string, []> var_14434_equation_0 = const()[name = tensor<string, []>("op_14434_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14434_cast = einsum(equation = var_14434_equation_0, values = (var_14260_cast, var_14208_cast))[name = tensor<string, []>("op_14434_cast")];
            tensor<fp16, []> var_14435_to_fp16 = const()[name = tensor<string, []>("op_14435_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1537_cast = mul(x = var_14434_cast, y = var_14435_to_fp16)[name = tensor<string, []>("aw_chunk_1537_cast")];
            tensor<string, []> var_14438_equation_0 = const()[name = tensor<string, []>("op_14438_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14438_cast = einsum(equation = var_14438_equation_0, values = (var_14260_cast, var_14209_cast))[name = tensor<string, []>("op_14438_cast")];
            tensor<fp16, []> var_14439_to_fp16 = const()[name = tensor<string, []>("op_14439_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1539_cast = mul(x = var_14438_cast, y = var_14439_to_fp16)[name = tensor<string, []>("aw_chunk_1539_cast")];
            tensor<string, []> var_14442_equation_0 = const()[name = tensor<string, []>("op_14442_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14442_cast = einsum(equation = var_14442_equation_0, values = (var_14260_cast, var_14210_cast))[name = tensor<string, []>("op_14442_cast")];
            tensor<fp16, []> var_14443_to_fp16 = const()[name = tensor<string, []>("op_14443_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1541_cast = mul(x = var_14442_cast, y = var_14443_to_fp16)[name = tensor<string, []>("aw_chunk_1541_cast")];
            tensor<string, []> var_14446_equation_0 = const()[name = tensor<string, []>("op_14446_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14446_cast = einsum(equation = var_14446_equation_0, values = (var_14260_cast, var_14211_cast))[name = tensor<string, []>("op_14446_cast")];
            tensor<fp16, []> var_14447_to_fp16 = const()[name = tensor<string, []>("op_14447_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1543_cast = mul(x = var_14446_cast, y = var_14447_to_fp16)[name = tensor<string, []>("aw_chunk_1543_cast")];
            tensor<string, []> var_14450_equation_0 = const()[name = tensor<string, []>("op_14450_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14450_cast = einsum(equation = var_14450_equation_0, values = (var_14260_cast, var_14212_cast))[name = tensor<string, []>("op_14450_cast")];
            tensor<fp16, []> var_14451_to_fp16 = const()[name = tensor<string, []>("op_14451_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1545_cast = mul(x = var_14450_cast, y = var_14451_to_fp16)[name = tensor<string, []>("aw_chunk_1545_cast")];
            tensor<string, []> var_14454_equation_0 = const()[name = tensor<string, []>("op_14454_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14454_cast = einsum(equation = var_14454_equation_0, values = (var_14260_cast, var_14213_cast))[name = tensor<string, []>("op_14454_cast")];
            tensor<fp16, []> var_14455_to_fp16 = const()[name = tensor<string, []>("op_14455_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1547_cast = mul(x = var_14454_cast, y = var_14455_to_fp16)[name = tensor<string, []>("aw_chunk_1547_cast")];
            tensor<string, []> var_14458_equation_0 = const()[name = tensor<string, []>("op_14458_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14458_cast = einsum(equation = var_14458_equation_0, values = (var_14260_cast, var_14214_cast))[name = tensor<string, []>("op_14458_cast")];
            tensor<fp16, []> var_14459_to_fp16 = const()[name = tensor<string, []>("op_14459_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1549_cast = mul(x = var_14458_cast, y = var_14459_to_fp16)[name = tensor<string, []>("aw_chunk_1549_cast")];
            tensor<string, []> var_14462_equation_0 = const()[name = tensor<string, []>("op_14462_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14462_cast = einsum(equation = var_14462_equation_0, values = (var_14260_cast, var_14215_cast))[name = tensor<string, []>("op_14462_cast")];
            tensor<fp16, []> var_14463_to_fp16 = const()[name = tensor<string, []>("op_14463_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1551_cast = mul(x = var_14462_cast, y = var_14463_to_fp16)[name = tensor<string, []>("aw_chunk_1551_cast")];
            tensor<string, []> var_14466_equation_0 = const()[name = tensor<string, []>("op_14466_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14466_cast = einsum(equation = var_14466_equation_0, values = (var_14264_cast, var_14216_cast))[name = tensor<string, []>("op_14466_cast")];
            tensor<fp16, []> var_14467_to_fp16 = const()[name = tensor<string, []>("op_14467_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1553_cast = mul(x = var_14466_cast, y = var_14467_to_fp16)[name = tensor<string, []>("aw_chunk_1553_cast")];
            tensor<string, []> var_14470_equation_0 = const()[name = tensor<string, []>("op_14470_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14470_cast = einsum(equation = var_14470_equation_0, values = (var_14264_cast, var_14217_cast))[name = tensor<string, []>("op_14470_cast")];
            tensor<fp16, []> var_14471_to_fp16 = const()[name = tensor<string, []>("op_14471_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1555_cast = mul(x = var_14470_cast, y = var_14471_to_fp16)[name = tensor<string, []>("aw_chunk_1555_cast")];
            tensor<string, []> var_14474_equation_0 = const()[name = tensor<string, []>("op_14474_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14474_cast = einsum(equation = var_14474_equation_0, values = (var_14264_cast, var_14218_cast))[name = tensor<string, []>("op_14474_cast")];
            tensor<fp16, []> var_14475_to_fp16 = const()[name = tensor<string, []>("op_14475_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1557_cast = mul(x = var_14474_cast, y = var_14475_to_fp16)[name = tensor<string, []>("aw_chunk_1557_cast")];
            tensor<string, []> var_14478_equation_0 = const()[name = tensor<string, []>("op_14478_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14478_cast = einsum(equation = var_14478_equation_0, values = (var_14264_cast, var_14219_cast))[name = tensor<string, []>("op_14478_cast")];
            tensor<fp16, []> var_14479_to_fp16 = const()[name = tensor<string, []>("op_14479_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1559_cast = mul(x = var_14478_cast, y = var_14479_to_fp16)[name = tensor<string, []>("aw_chunk_1559_cast")];
            tensor<string, []> var_14482_equation_0 = const()[name = tensor<string, []>("op_14482_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14482_cast = einsum(equation = var_14482_equation_0, values = (var_14264_cast, var_14220_cast))[name = tensor<string, []>("op_14482_cast")];
            tensor<fp16, []> var_14483_to_fp16 = const()[name = tensor<string, []>("op_14483_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1561_cast = mul(x = var_14482_cast, y = var_14483_to_fp16)[name = tensor<string, []>("aw_chunk_1561_cast")];
            tensor<string, []> var_14486_equation_0 = const()[name = tensor<string, []>("op_14486_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14486_cast = einsum(equation = var_14486_equation_0, values = (var_14264_cast, var_14221_cast))[name = tensor<string, []>("op_14486_cast")];
            tensor<fp16, []> var_14487_to_fp16 = const()[name = tensor<string, []>("op_14487_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1563_cast = mul(x = var_14486_cast, y = var_14487_to_fp16)[name = tensor<string, []>("aw_chunk_1563_cast")];
            tensor<string, []> var_14490_equation_0 = const()[name = tensor<string, []>("op_14490_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14490_cast = einsum(equation = var_14490_equation_0, values = (var_14264_cast, var_14222_cast))[name = tensor<string, []>("op_14490_cast")];
            tensor<fp16, []> var_14491_to_fp16 = const()[name = tensor<string, []>("op_14491_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1565_cast = mul(x = var_14490_cast, y = var_14491_to_fp16)[name = tensor<string, []>("aw_chunk_1565_cast")];
            tensor<string, []> var_14494_equation_0 = const()[name = tensor<string, []>("op_14494_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14494_cast = einsum(equation = var_14494_equation_0, values = (var_14264_cast, var_14223_cast))[name = tensor<string, []>("op_14494_cast")];
            tensor<fp16, []> var_14495_to_fp16 = const()[name = tensor<string, []>("op_14495_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1567_cast = mul(x = var_14494_cast, y = var_14495_to_fp16)[name = tensor<string, []>("aw_chunk_1567_cast")];
            tensor<string, []> var_14498_equation_0 = const()[name = tensor<string, []>("op_14498_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14498_cast = einsum(equation = var_14498_equation_0, values = (var_14268_cast, var_14224_cast))[name = tensor<string, []>("op_14498_cast")];
            tensor<fp16, []> var_14499_to_fp16 = const()[name = tensor<string, []>("op_14499_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1569_cast = mul(x = var_14498_cast, y = var_14499_to_fp16)[name = tensor<string, []>("aw_chunk_1569_cast")];
            tensor<string, []> var_14502_equation_0 = const()[name = tensor<string, []>("op_14502_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14502_cast = einsum(equation = var_14502_equation_0, values = (var_14268_cast, var_14225_cast))[name = tensor<string, []>("op_14502_cast")];
            tensor<fp16, []> var_14503_to_fp16 = const()[name = tensor<string, []>("op_14503_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1571_cast = mul(x = var_14502_cast, y = var_14503_to_fp16)[name = tensor<string, []>("aw_chunk_1571_cast")];
            tensor<string, []> var_14506_equation_0 = const()[name = tensor<string, []>("op_14506_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14506_cast = einsum(equation = var_14506_equation_0, values = (var_14268_cast, var_14226_cast))[name = tensor<string, []>("op_14506_cast")];
            tensor<fp16, []> var_14507_to_fp16 = const()[name = tensor<string, []>("op_14507_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1573_cast = mul(x = var_14506_cast, y = var_14507_to_fp16)[name = tensor<string, []>("aw_chunk_1573_cast")];
            tensor<string, []> var_14510_equation_0 = const()[name = tensor<string, []>("op_14510_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14510_cast = einsum(equation = var_14510_equation_0, values = (var_14268_cast, var_14227_cast))[name = tensor<string, []>("op_14510_cast")];
            tensor<fp16, []> var_14511_to_fp16 = const()[name = tensor<string, []>("op_14511_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1575_cast = mul(x = var_14510_cast, y = var_14511_to_fp16)[name = tensor<string, []>("aw_chunk_1575_cast")];
            tensor<string, []> var_14514_equation_0 = const()[name = tensor<string, []>("op_14514_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14514_cast = einsum(equation = var_14514_equation_0, values = (var_14268_cast, var_14228_cast))[name = tensor<string, []>("op_14514_cast")];
            tensor<fp16, []> var_14515_to_fp16 = const()[name = tensor<string, []>("op_14515_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1577_cast = mul(x = var_14514_cast, y = var_14515_to_fp16)[name = tensor<string, []>("aw_chunk_1577_cast")];
            tensor<string, []> var_14518_equation_0 = const()[name = tensor<string, []>("op_14518_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14518_cast = einsum(equation = var_14518_equation_0, values = (var_14268_cast, var_14229_cast))[name = tensor<string, []>("op_14518_cast")];
            tensor<fp16, []> var_14519_to_fp16 = const()[name = tensor<string, []>("op_14519_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1579_cast = mul(x = var_14518_cast, y = var_14519_to_fp16)[name = tensor<string, []>("aw_chunk_1579_cast")];
            tensor<string, []> var_14522_equation_0 = const()[name = tensor<string, []>("op_14522_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14522_cast = einsum(equation = var_14522_equation_0, values = (var_14268_cast, var_14230_cast))[name = tensor<string, []>("op_14522_cast")];
            tensor<fp16, []> var_14523_to_fp16 = const()[name = tensor<string, []>("op_14523_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1581_cast = mul(x = var_14522_cast, y = var_14523_to_fp16)[name = tensor<string, []>("aw_chunk_1581_cast")];
            tensor<string, []> var_14526_equation_0 = const()[name = tensor<string, []>("op_14526_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14526_cast = einsum(equation = var_14526_equation_0, values = (var_14268_cast, var_14231_cast))[name = tensor<string, []>("op_14526_cast")];
            tensor<fp16, []> var_14527_to_fp16 = const()[name = tensor<string, []>("op_14527_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1583_cast = mul(x = var_14526_cast, y = var_14527_to_fp16)[name = tensor<string, []>("aw_chunk_1583_cast")];
            tensor<string, []> var_14530_equation_0 = const()[name = tensor<string, []>("op_14530_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14530_cast = einsum(equation = var_14530_equation_0, values = (var_14272_cast, var_14232_cast))[name = tensor<string, []>("op_14530_cast")];
            tensor<fp16, []> var_14531_to_fp16 = const()[name = tensor<string, []>("op_14531_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1585_cast = mul(x = var_14530_cast, y = var_14531_to_fp16)[name = tensor<string, []>("aw_chunk_1585_cast")];
            tensor<string, []> var_14534_equation_0 = const()[name = tensor<string, []>("op_14534_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14534_cast = einsum(equation = var_14534_equation_0, values = (var_14272_cast, var_14233_cast))[name = tensor<string, []>("op_14534_cast")];
            tensor<fp16, []> var_14535_to_fp16 = const()[name = tensor<string, []>("op_14535_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1587_cast = mul(x = var_14534_cast, y = var_14535_to_fp16)[name = tensor<string, []>("aw_chunk_1587_cast")];
            tensor<string, []> var_14538_equation_0 = const()[name = tensor<string, []>("op_14538_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14538_cast = einsum(equation = var_14538_equation_0, values = (var_14272_cast, var_14234_cast))[name = tensor<string, []>("op_14538_cast")];
            tensor<fp16, []> var_14539_to_fp16 = const()[name = tensor<string, []>("op_14539_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1589_cast = mul(x = var_14538_cast, y = var_14539_to_fp16)[name = tensor<string, []>("aw_chunk_1589_cast")];
            tensor<string, []> var_14542_equation_0 = const()[name = tensor<string, []>("op_14542_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14542_cast = einsum(equation = var_14542_equation_0, values = (var_14272_cast, var_14235_cast))[name = tensor<string, []>("op_14542_cast")];
            tensor<fp16, []> var_14543_to_fp16 = const()[name = tensor<string, []>("op_14543_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1591_cast = mul(x = var_14542_cast, y = var_14543_to_fp16)[name = tensor<string, []>("aw_chunk_1591_cast")];
            tensor<string, []> var_14546_equation_0 = const()[name = tensor<string, []>("op_14546_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14546_cast = einsum(equation = var_14546_equation_0, values = (var_14272_cast, var_14236_cast))[name = tensor<string, []>("op_14546_cast")];
            tensor<fp16, []> var_14547_to_fp16 = const()[name = tensor<string, []>("op_14547_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1593_cast = mul(x = var_14546_cast, y = var_14547_to_fp16)[name = tensor<string, []>("aw_chunk_1593_cast")];
            tensor<string, []> var_14550_equation_0 = const()[name = tensor<string, []>("op_14550_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14550_cast = einsum(equation = var_14550_equation_0, values = (var_14272_cast, var_14237_cast))[name = tensor<string, []>("op_14550_cast")];
            tensor<fp16, []> var_14551_to_fp16 = const()[name = tensor<string, []>("op_14551_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1595_cast = mul(x = var_14550_cast, y = var_14551_to_fp16)[name = tensor<string, []>("aw_chunk_1595_cast")];
            tensor<string, []> var_14554_equation_0 = const()[name = tensor<string, []>("op_14554_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14554_cast = einsum(equation = var_14554_equation_0, values = (var_14272_cast, var_14238_cast))[name = tensor<string, []>("op_14554_cast")];
            tensor<fp16, []> var_14555_to_fp16 = const()[name = tensor<string, []>("op_14555_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1597_cast = mul(x = var_14554_cast, y = var_14555_to_fp16)[name = tensor<string, []>("aw_chunk_1597_cast")];
            tensor<string, []> var_14558_equation_0 = const()[name = tensor<string, []>("op_14558_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14558_cast = einsum(equation = var_14558_equation_0, values = (var_14272_cast, var_14239_cast))[name = tensor<string, []>("op_14558_cast")];
            tensor<fp16, []> var_14559_to_fp16 = const()[name = tensor<string, []>("op_14559_to_fp16"), val = tensor<fp16, []>(0x1.43cp-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_cast = mul(x = var_14558_cast, y = var_14559_to_fp16)[name = tensor<string, []>("aw_chunk_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14561_cast = softmax(axis = var_10375, x = aw_chunk_1473_cast)[name = tensor<string, []>("op_14561_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14562_cast = softmax(axis = var_10375, x = aw_chunk_1475_cast)[name = tensor<string, []>("op_14562_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14563_cast = softmax(axis = var_10375, x = aw_chunk_1477_cast)[name = tensor<string, []>("op_14563_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14564_cast = softmax(axis = var_10375, x = aw_chunk_1479_cast)[name = tensor<string, []>("op_14564_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14565_cast = softmax(axis = var_10375, x = aw_chunk_1481_cast)[name = tensor<string, []>("op_14565_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14566_cast = softmax(axis = var_10375, x = aw_chunk_1483_cast)[name = tensor<string, []>("op_14566_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14567_cast = softmax(axis = var_10375, x = aw_chunk_1485_cast)[name = tensor<string, []>("op_14567_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14568_cast = softmax(axis = var_10375, x = aw_chunk_1487_cast)[name = tensor<string, []>("op_14568_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14569_cast = softmax(axis = var_10375, x = aw_chunk_1489_cast)[name = tensor<string, []>("op_14569_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14570_cast = softmax(axis = var_10375, x = aw_chunk_1491_cast)[name = tensor<string, []>("op_14570_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14571_cast = softmax(axis = var_10375, x = aw_chunk_1493_cast)[name = tensor<string, []>("op_14571_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14572_cast = softmax(axis = var_10375, x = aw_chunk_1495_cast)[name = tensor<string, []>("op_14572_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14573_cast = softmax(axis = var_10375, x = aw_chunk_1497_cast)[name = tensor<string, []>("op_14573_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14574_cast = softmax(axis = var_10375, x = aw_chunk_1499_cast)[name = tensor<string, []>("op_14574_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14575_cast = softmax(axis = var_10375, x = aw_chunk_1501_cast)[name = tensor<string, []>("op_14575_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14576_cast = softmax(axis = var_10375, x = aw_chunk_1503_cast)[name = tensor<string, []>("op_14576_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14577_cast = softmax(axis = var_10375, x = aw_chunk_1505_cast)[name = tensor<string, []>("op_14577_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14578_cast = softmax(axis = var_10375, x = aw_chunk_1507_cast)[name = tensor<string, []>("op_14578_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14579_cast = softmax(axis = var_10375, x = aw_chunk_1509_cast)[name = tensor<string, []>("op_14579_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14580_cast = softmax(axis = var_10375, x = aw_chunk_1511_cast)[name = tensor<string, []>("op_14580_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14581_cast = softmax(axis = var_10375, x = aw_chunk_1513_cast)[name = tensor<string, []>("op_14581_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14582_cast = softmax(axis = var_10375, x = aw_chunk_1515_cast)[name = tensor<string, []>("op_14582_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14583_cast = softmax(axis = var_10375, x = aw_chunk_1517_cast)[name = tensor<string, []>("op_14583_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14584_cast = softmax(axis = var_10375, x = aw_chunk_1519_cast)[name = tensor<string, []>("op_14584_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14585_cast = softmax(axis = var_10375, x = aw_chunk_1521_cast)[name = tensor<string, []>("op_14585_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14586_cast = softmax(axis = var_10375, x = aw_chunk_1523_cast)[name = tensor<string, []>("op_14586_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14587_cast = softmax(axis = var_10375, x = aw_chunk_1525_cast)[name = tensor<string, []>("op_14587_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14588_cast = softmax(axis = var_10375, x = aw_chunk_1527_cast)[name = tensor<string, []>("op_14588_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14589_cast = softmax(axis = var_10375, x = aw_chunk_1529_cast)[name = tensor<string, []>("op_14589_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14590_cast = softmax(axis = var_10375, x = aw_chunk_1531_cast)[name = tensor<string, []>("op_14590_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14591_cast = softmax(axis = var_10375, x = aw_chunk_1533_cast)[name = tensor<string, []>("op_14591_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14592_cast = softmax(axis = var_10375, x = aw_chunk_1535_cast)[name = tensor<string, []>("op_14592_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14593_cast = softmax(axis = var_10375, x = aw_chunk_1537_cast)[name = tensor<string, []>("op_14593_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14594_cast = softmax(axis = var_10375, x = aw_chunk_1539_cast)[name = tensor<string, []>("op_14594_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14595_cast = softmax(axis = var_10375, x = aw_chunk_1541_cast)[name = tensor<string, []>("op_14595_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14596_cast = softmax(axis = var_10375, x = aw_chunk_1543_cast)[name = tensor<string, []>("op_14596_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14597_cast = softmax(axis = var_10375, x = aw_chunk_1545_cast)[name = tensor<string, []>("op_14597_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14598_cast = softmax(axis = var_10375, x = aw_chunk_1547_cast)[name = tensor<string, []>("op_14598_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14599_cast = softmax(axis = var_10375, x = aw_chunk_1549_cast)[name = tensor<string, []>("op_14599_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14600_cast = softmax(axis = var_10375, x = aw_chunk_1551_cast)[name = tensor<string, []>("op_14600_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14601_cast = softmax(axis = var_10375, x = aw_chunk_1553_cast)[name = tensor<string, []>("op_14601_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14602_cast = softmax(axis = var_10375, x = aw_chunk_1555_cast)[name = tensor<string, []>("op_14602_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14603_cast = softmax(axis = var_10375, x = aw_chunk_1557_cast)[name = tensor<string, []>("op_14603_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14604_cast = softmax(axis = var_10375, x = aw_chunk_1559_cast)[name = tensor<string, []>("op_14604_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14605_cast = softmax(axis = var_10375, x = aw_chunk_1561_cast)[name = tensor<string, []>("op_14605_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14606_cast = softmax(axis = var_10375, x = aw_chunk_1563_cast)[name = tensor<string, []>("op_14606_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14607_cast = softmax(axis = var_10375, x = aw_chunk_1565_cast)[name = tensor<string, []>("op_14607_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14608_cast = softmax(axis = var_10375, x = aw_chunk_1567_cast)[name = tensor<string, []>("op_14608_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14609_cast = softmax(axis = var_10375, x = aw_chunk_1569_cast)[name = tensor<string, []>("op_14609_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14610_cast = softmax(axis = var_10375, x = aw_chunk_1571_cast)[name = tensor<string, []>("op_14610_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14611_cast = softmax(axis = var_10375, x = aw_chunk_1573_cast)[name = tensor<string, []>("op_14611_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14612_cast = softmax(axis = var_10375, x = aw_chunk_1575_cast)[name = tensor<string, []>("op_14612_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14613_cast = softmax(axis = var_10375, x = aw_chunk_1577_cast)[name = tensor<string, []>("op_14613_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14614_cast = softmax(axis = var_10375, x = aw_chunk_1579_cast)[name = tensor<string, []>("op_14614_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14615_cast = softmax(axis = var_10375, x = aw_chunk_1581_cast)[name = tensor<string, []>("op_14615_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14616_cast = softmax(axis = var_10375, x = aw_chunk_1583_cast)[name = tensor<string, []>("op_14616_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14617_cast = softmax(axis = var_10375, x = aw_chunk_1585_cast)[name = tensor<string, []>("op_14617_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14618_cast = softmax(axis = var_10375, x = aw_chunk_1587_cast)[name = tensor<string, []>("op_14618_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14619_cast = softmax(axis = var_10375, x = aw_chunk_1589_cast)[name = tensor<string, []>("op_14619_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14620_cast = softmax(axis = var_10375, x = aw_chunk_1591_cast)[name = tensor<string, []>("op_14620_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14621_cast = softmax(axis = var_10375, x = aw_chunk_1593_cast)[name = tensor<string, []>("op_14621_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14622_cast = softmax(axis = var_10375, x = aw_chunk_1595_cast)[name = tensor<string, []>("op_14622_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14623_cast = softmax(axis = var_10375, x = aw_chunk_1597_cast)[name = tensor<string, []>("op_14623_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14624_cast = softmax(axis = var_10375, x = aw_chunk_cast)[name = tensor<string, []>("op_14624_cast")];
            tensor<string, []> var_14626_equation_0 = const()[name = tensor<string, []>("op_14626_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14626_cast = einsum(equation = var_14626_equation_0, values = (var_14274_cast, var_14561_cast))[name = tensor<string, []>("op_14626_cast")];
            tensor<string, []> var_14628_equation_0 = const()[name = tensor<string, []>("op_14628_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14628_cast = einsum(equation = var_14628_equation_0, values = (var_14274_cast, var_14562_cast))[name = tensor<string, []>("op_14628_cast")];
            tensor<string, []> var_14630_equation_0 = const()[name = tensor<string, []>("op_14630_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14630_cast = einsum(equation = var_14630_equation_0, values = (var_14274_cast, var_14563_cast))[name = tensor<string, []>("op_14630_cast")];
            tensor<string, []> var_14632_equation_0 = const()[name = tensor<string, []>("op_14632_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14632_cast = einsum(equation = var_14632_equation_0, values = (var_14274_cast, var_14564_cast))[name = tensor<string, []>("op_14632_cast")];
            tensor<string, []> var_14634_equation_0 = const()[name = tensor<string, []>("op_14634_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14634_cast = einsum(equation = var_14634_equation_0, values = (var_14274_cast, var_14565_cast))[name = tensor<string, []>("op_14634_cast")];
            tensor<string, []> var_14636_equation_0 = const()[name = tensor<string, []>("op_14636_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14636_cast = einsum(equation = var_14636_equation_0, values = (var_14274_cast, var_14566_cast))[name = tensor<string, []>("op_14636_cast")];
            tensor<string, []> var_14638_equation_0 = const()[name = tensor<string, []>("op_14638_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14638_cast = einsum(equation = var_14638_equation_0, values = (var_14274_cast, var_14567_cast))[name = tensor<string, []>("op_14638_cast")];
            tensor<string, []> var_14640_equation_0 = const()[name = tensor<string, []>("op_14640_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14640_cast = einsum(equation = var_14640_equation_0, values = (var_14274_cast, var_14568_cast))[name = tensor<string, []>("op_14640_cast")];
            tensor<string, []> var_14642_equation_0 = const()[name = tensor<string, []>("op_14642_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14642_cast = einsum(equation = var_14642_equation_0, values = (var_14278_cast, var_14569_cast))[name = tensor<string, []>("op_14642_cast")];
            tensor<string, []> var_14644_equation_0 = const()[name = tensor<string, []>("op_14644_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14644_cast = einsum(equation = var_14644_equation_0, values = (var_14278_cast, var_14570_cast))[name = tensor<string, []>("op_14644_cast")];
            tensor<string, []> var_14646_equation_0 = const()[name = tensor<string, []>("op_14646_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14646_cast = einsum(equation = var_14646_equation_0, values = (var_14278_cast, var_14571_cast))[name = tensor<string, []>("op_14646_cast")];
            tensor<string, []> var_14648_equation_0 = const()[name = tensor<string, []>("op_14648_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14648_cast = einsum(equation = var_14648_equation_0, values = (var_14278_cast, var_14572_cast))[name = tensor<string, []>("op_14648_cast")];
            tensor<string, []> var_14650_equation_0 = const()[name = tensor<string, []>("op_14650_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14650_cast = einsum(equation = var_14650_equation_0, values = (var_14278_cast, var_14573_cast))[name = tensor<string, []>("op_14650_cast")];
            tensor<string, []> var_14652_equation_0 = const()[name = tensor<string, []>("op_14652_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14652_cast = einsum(equation = var_14652_equation_0, values = (var_14278_cast, var_14574_cast))[name = tensor<string, []>("op_14652_cast")];
            tensor<string, []> var_14654_equation_0 = const()[name = tensor<string, []>("op_14654_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14654_cast = einsum(equation = var_14654_equation_0, values = (var_14278_cast, var_14575_cast))[name = tensor<string, []>("op_14654_cast")];
            tensor<string, []> var_14656_equation_0 = const()[name = tensor<string, []>("op_14656_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14656_cast = einsum(equation = var_14656_equation_0, values = (var_14278_cast, var_14576_cast))[name = tensor<string, []>("op_14656_cast")];
            tensor<string, []> var_14658_equation_0 = const()[name = tensor<string, []>("op_14658_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14658_cast = einsum(equation = var_14658_equation_0, values = (var_14282_cast, var_14577_cast))[name = tensor<string, []>("op_14658_cast")];
            tensor<string, []> var_14660_equation_0 = const()[name = tensor<string, []>("op_14660_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14660_cast = einsum(equation = var_14660_equation_0, values = (var_14282_cast, var_14578_cast))[name = tensor<string, []>("op_14660_cast")];
            tensor<string, []> var_14662_equation_0 = const()[name = tensor<string, []>("op_14662_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14662_cast = einsum(equation = var_14662_equation_0, values = (var_14282_cast, var_14579_cast))[name = tensor<string, []>("op_14662_cast")];
            tensor<string, []> var_14664_equation_0 = const()[name = tensor<string, []>("op_14664_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14664_cast = einsum(equation = var_14664_equation_0, values = (var_14282_cast, var_14580_cast))[name = tensor<string, []>("op_14664_cast")];
            tensor<string, []> var_14666_equation_0 = const()[name = tensor<string, []>("op_14666_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14666_cast = einsum(equation = var_14666_equation_0, values = (var_14282_cast, var_14581_cast))[name = tensor<string, []>("op_14666_cast")];
            tensor<string, []> var_14668_equation_0 = const()[name = tensor<string, []>("op_14668_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14668_cast = einsum(equation = var_14668_equation_0, values = (var_14282_cast, var_14582_cast))[name = tensor<string, []>("op_14668_cast")];
            tensor<string, []> var_14670_equation_0 = const()[name = tensor<string, []>("op_14670_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14670_cast = einsum(equation = var_14670_equation_0, values = (var_14282_cast, var_14583_cast))[name = tensor<string, []>("op_14670_cast")];
            tensor<string, []> var_14672_equation_0 = const()[name = tensor<string, []>("op_14672_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14672_cast = einsum(equation = var_14672_equation_0, values = (var_14282_cast, var_14584_cast))[name = tensor<string, []>("op_14672_cast")];
            tensor<string, []> var_14674_equation_0 = const()[name = tensor<string, []>("op_14674_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14674_cast = einsum(equation = var_14674_equation_0, values = (var_14286_cast, var_14585_cast))[name = tensor<string, []>("op_14674_cast")];
            tensor<string, []> var_14676_equation_0 = const()[name = tensor<string, []>("op_14676_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14676_cast = einsum(equation = var_14676_equation_0, values = (var_14286_cast, var_14586_cast))[name = tensor<string, []>("op_14676_cast")];
            tensor<string, []> var_14678_equation_0 = const()[name = tensor<string, []>("op_14678_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14678_cast = einsum(equation = var_14678_equation_0, values = (var_14286_cast, var_14587_cast))[name = tensor<string, []>("op_14678_cast")];
            tensor<string, []> var_14680_equation_0 = const()[name = tensor<string, []>("op_14680_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14680_cast = einsum(equation = var_14680_equation_0, values = (var_14286_cast, var_14588_cast))[name = tensor<string, []>("op_14680_cast")];
            tensor<string, []> var_14682_equation_0 = const()[name = tensor<string, []>("op_14682_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14682_cast = einsum(equation = var_14682_equation_0, values = (var_14286_cast, var_14589_cast))[name = tensor<string, []>("op_14682_cast")];
            tensor<string, []> var_14684_equation_0 = const()[name = tensor<string, []>("op_14684_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14684_cast = einsum(equation = var_14684_equation_0, values = (var_14286_cast, var_14590_cast))[name = tensor<string, []>("op_14684_cast")];
            tensor<string, []> var_14686_equation_0 = const()[name = tensor<string, []>("op_14686_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14686_cast = einsum(equation = var_14686_equation_0, values = (var_14286_cast, var_14591_cast))[name = tensor<string, []>("op_14686_cast")];
            tensor<string, []> var_14688_equation_0 = const()[name = tensor<string, []>("op_14688_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14688_cast = einsum(equation = var_14688_equation_0, values = (var_14286_cast, var_14592_cast))[name = tensor<string, []>("op_14688_cast")];
            tensor<string, []> var_14690_equation_0 = const()[name = tensor<string, []>("op_14690_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14690_cast = einsum(equation = var_14690_equation_0, values = (var_14290_cast, var_14593_cast))[name = tensor<string, []>("op_14690_cast")];
            tensor<string, []> var_14692_equation_0 = const()[name = tensor<string, []>("op_14692_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14692_cast = einsum(equation = var_14692_equation_0, values = (var_14290_cast, var_14594_cast))[name = tensor<string, []>("op_14692_cast")];
            tensor<string, []> var_14694_equation_0 = const()[name = tensor<string, []>("op_14694_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14694_cast = einsum(equation = var_14694_equation_0, values = (var_14290_cast, var_14595_cast))[name = tensor<string, []>("op_14694_cast")];
            tensor<string, []> var_14696_equation_0 = const()[name = tensor<string, []>("op_14696_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14696_cast = einsum(equation = var_14696_equation_0, values = (var_14290_cast, var_14596_cast))[name = tensor<string, []>("op_14696_cast")];
            tensor<string, []> var_14698_equation_0 = const()[name = tensor<string, []>("op_14698_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14698_cast = einsum(equation = var_14698_equation_0, values = (var_14290_cast, var_14597_cast))[name = tensor<string, []>("op_14698_cast")];
            tensor<string, []> var_14700_equation_0 = const()[name = tensor<string, []>("op_14700_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14700_cast = einsum(equation = var_14700_equation_0, values = (var_14290_cast, var_14598_cast))[name = tensor<string, []>("op_14700_cast")];
            tensor<string, []> var_14702_equation_0 = const()[name = tensor<string, []>("op_14702_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14702_cast = einsum(equation = var_14702_equation_0, values = (var_14290_cast, var_14599_cast))[name = tensor<string, []>("op_14702_cast")];
            tensor<string, []> var_14704_equation_0 = const()[name = tensor<string, []>("op_14704_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14704_cast = einsum(equation = var_14704_equation_0, values = (var_14290_cast, var_14600_cast))[name = tensor<string, []>("op_14704_cast")];
            tensor<string, []> var_14706_equation_0 = const()[name = tensor<string, []>("op_14706_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14706_cast = einsum(equation = var_14706_equation_0, values = (var_14294_cast, var_14601_cast))[name = tensor<string, []>("op_14706_cast")];
            tensor<string, []> var_14708_equation_0 = const()[name = tensor<string, []>("op_14708_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14708_cast = einsum(equation = var_14708_equation_0, values = (var_14294_cast, var_14602_cast))[name = tensor<string, []>("op_14708_cast")];
            tensor<string, []> var_14710_equation_0 = const()[name = tensor<string, []>("op_14710_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14710_cast = einsum(equation = var_14710_equation_0, values = (var_14294_cast, var_14603_cast))[name = tensor<string, []>("op_14710_cast")];
            tensor<string, []> var_14712_equation_0 = const()[name = tensor<string, []>("op_14712_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14712_cast = einsum(equation = var_14712_equation_0, values = (var_14294_cast, var_14604_cast))[name = tensor<string, []>("op_14712_cast")];
            tensor<string, []> var_14714_equation_0 = const()[name = tensor<string, []>("op_14714_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14714_cast = einsum(equation = var_14714_equation_0, values = (var_14294_cast, var_14605_cast))[name = tensor<string, []>("op_14714_cast")];
            tensor<string, []> var_14716_equation_0 = const()[name = tensor<string, []>("op_14716_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14716_cast = einsum(equation = var_14716_equation_0, values = (var_14294_cast, var_14606_cast))[name = tensor<string, []>("op_14716_cast")];
            tensor<string, []> var_14718_equation_0 = const()[name = tensor<string, []>("op_14718_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14718_cast = einsum(equation = var_14718_equation_0, values = (var_14294_cast, var_14607_cast))[name = tensor<string, []>("op_14718_cast")];
            tensor<string, []> var_14720_equation_0 = const()[name = tensor<string, []>("op_14720_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14720_cast = einsum(equation = var_14720_equation_0, values = (var_14294_cast, var_14608_cast))[name = tensor<string, []>("op_14720_cast")];
            tensor<string, []> var_14722_equation_0 = const()[name = tensor<string, []>("op_14722_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14722_cast = einsum(equation = var_14722_equation_0, values = (var_14298_cast, var_14609_cast))[name = tensor<string, []>("op_14722_cast")];
            tensor<string, []> var_14724_equation_0 = const()[name = tensor<string, []>("op_14724_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14724_cast = einsum(equation = var_14724_equation_0, values = (var_14298_cast, var_14610_cast))[name = tensor<string, []>("op_14724_cast")];
            tensor<string, []> var_14726_equation_0 = const()[name = tensor<string, []>("op_14726_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14726_cast = einsum(equation = var_14726_equation_0, values = (var_14298_cast, var_14611_cast))[name = tensor<string, []>("op_14726_cast")];
            tensor<string, []> var_14728_equation_0 = const()[name = tensor<string, []>("op_14728_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14728_cast = einsum(equation = var_14728_equation_0, values = (var_14298_cast, var_14612_cast))[name = tensor<string, []>("op_14728_cast")];
            tensor<string, []> var_14730_equation_0 = const()[name = tensor<string, []>("op_14730_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14730_cast = einsum(equation = var_14730_equation_0, values = (var_14298_cast, var_14613_cast))[name = tensor<string, []>("op_14730_cast")];
            tensor<string, []> var_14732_equation_0 = const()[name = tensor<string, []>("op_14732_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14732_cast = einsum(equation = var_14732_equation_0, values = (var_14298_cast, var_14614_cast))[name = tensor<string, []>("op_14732_cast")];
            tensor<string, []> var_14734_equation_0 = const()[name = tensor<string, []>("op_14734_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14734_cast = einsum(equation = var_14734_equation_0, values = (var_14298_cast, var_14615_cast))[name = tensor<string, []>("op_14734_cast")];
            tensor<string, []> var_14736_equation_0 = const()[name = tensor<string, []>("op_14736_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14736_cast = einsum(equation = var_14736_equation_0, values = (var_14298_cast, var_14616_cast))[name = tensor<string, []>("op_14736_cast")];
            tensor<string, []> var_14738_equation_0 = const()[name = tensor<string, []>("op_14738_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14738_cast = einsum(equation = var_14738_equation_0, values = (var_14302_cast, var_14617_cast))[name = tensor<string, []>("op_14738_cast")];
            tensor<string, []> var_14740_equation_0 = const()[name = tensor<string, []>("op_14740_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14740_cast = einsum(equation = var_14740_equation_0, values = (var_14302_cast, var_14618_cast))[name = tensor<string, []>("op_14740_cast")];
            tensor<string, []> var_14742_equation_0 = const()[name = tensor<string, []>("op_14742_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14742_cast = einsum(equation = var_14742_equation_0, values = (var_14302_cast, var_14619_cast))[name = tensor<string, []>("op_14742_cast")];
            tensor<string, []> var_14744_equation_0 = const()[name = tensor<string, []>("op_14744_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14744_cast = einsum(equation = var_14744_equation_0, values = (var_14302_cast, var_14620_cast))[name = tensor<string, []>("op_14744_cast")];
            tensor<string, []> var_14746_equation_0 = const()[name = tensor<string, []>("op_14746_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14746_cast = einsum(equation = var_14746_equation_0, values = (var_14302_cast, var_14621_cast))[name = tensor<string, []>("op_14746_cast")];
            tensor<string, []> var_14748_equation_0 = const()[name = tensor<string, []>("op_14748_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14748_cast = einsum(equation = var_14748_equation_0, values = (var_14302_cast, var_14622_cast))[name = tensor<string, []>("op_14748_cast")];
            tensor<string, []> var_14750_equation_0 = const()[name = tensor<string, []>("op_14750_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14750_cast = einsum(equation = var_14750_equation_0, values = (var_14302_cast, var_14623_cast))[name = tensor<string, []>("op_14750_cast")];
            tensor<string, []> var_14752_equation_0 = const()[name = tensor<string, []>("op_14752_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 40, 1, 512]> var_14752_cast = einsum(equation = var_14752_equation_0, values = (var_14302_cast, var_14624_cast))[name = tensor<string, []>("op_14752_cast")];
            tensor<bool, []> var_14754_interleave_0 = const()[name = tensor<string, []>("op_14754_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_14754_cast = concat(axis = var_10347, interleave = var_14754_interleave_0, values = (var_14626_cast, var_14628_cast, var_14630_cast, var_14632_cast, var_14634_cast, var_14636_cast, var_14638_cast, var_14640_cast))[name = tensor<string, []>("op_14754_cast")];
            tensor<bool, []> var_14756_interleave_0 = const()[name = tensor<string, []>("op_14756_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_14756_cast = concat(axis = var_10347, interleave = var_14756_interleave_0, values = (var_14642_cast, var_14644_cast, var_14646_cast, var_14648_cast, var_14650_cast, var_14652_cast, var_14654_cast, var_14656_cast))[name = tensor<string, []>("op_14756_cast")];
            tensor<bool, []> var_14758_interleave_0 = const()[name = tensor<string, []>("op_14758_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_14758_cast = concat(axis = var_10347, interleave = var_14758_interleave_0, values = (var_14658_cast, var_14660_cast, var_14662_cast, var_14664_cast, var_14666_cast, var_14668_cast, var_14670_cast, var_14672_cast))[name = tensor<string, []>("op_14758_cast")];
            tensor<bool, []> var_14760_interleave_0 = const()[name = tensor<string, []>("op_14760_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_14760_cast = concat(axis = var_10347, interleave = var_14760_interleave_0, values = (var_14674_cast, var_14676_cast, var_14678_cast, var_14680_cast, var_14682_cast, var_14684_cast, var_14686_cast, var_14688_cast))[name = tensor<string, []>("op_14760_cast")];
            tensor<bool, []> var_14762_interleave_0 = const()[name = tensor<string, []>("op_14762_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_14762_cast = concat(axis = var_10347, interleave = var_14762_interleave_0, values = (var_14690_cast, var_14692_cast, var_14694_cast, var_14696_cast, var_14698_cast, var_14700_cast, var_14702_cast, var_14704_cast))[name = tensor<string, []>("op_14762_cast")];
            tensor<bool, []> var_14764_interleave_0 = const()[name = tensor<string, []>("op_14764_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_14764_cast = concat(axis = var_10347, interleave = var_14764_interleave_0, values = (var_14706_cast, var_14708_cast, var_14710_cast, var_14712_cast, var_14714_cast, var_14716_cast, var_14718_cast, var_14720_cast))[name = tensor<string, []>("op_14764_cast")];
            tensor<bool, []> var_14766_interleave_0 = const()[name = tensor<string, []>("op_14766_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_14766_cast = concat(axis = var_10347, interleave = var_14766_interleave_0, values = (var_14722_cast, var_14724_cast, var_14726_cast, var_14728_cast, var_14730_cast, var_14732_cast, var_14734_cast, var_14736_cast))[name = tensor<string, []>("op_14766_cast")];
            tensor<bool, []> var_14768_interleave_0 = const()[name = tensor<string, []>("op_14768_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 40, 1, 4096]> var_14768_cast = concat(axis = var_10347, interleave = var_14768_interleave_0, values = (var_14738_cast, var_14740_cast, var_14742_cast, var_14744_cast, var_14746_cast, var_14748_cast, var_14750_cast, var_14752_cast))[name = tensor<string, []>("op_14768_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_10375, interleave = input_519_interleave_0, values = (var_14754_cast, var_14756_cast, var_14758_cast, var_14760_cast, var_14762_cast, var_14764_cast, var_14766_cast, var_14768_cast))[name = tensor<string, []>("input_519_cast")];
            tensor<int32, [2]> var_14774 = const()[name = tensor<string, []>("op_14774"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14776 = const()[name = tensor<string, []>("op_14776"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_14778_pad_type_0 = const()[name = tensor<string, []>("op_14778_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_14778_pad_0 = const()[name = tensor<string, []>("op_14778_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(644075392))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(644152256))), name = tensor<string, []>("up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            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, []>(644152448)))];
            tensor<fp16, [2, 320, 1, 4096]> var_14778_cast = conv(bias = up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_14776, groups = var_10375, pad = var_14778_pad_0, pad_type = var_14778_pad_type_0, strides = var_14774, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_519_cast)[name = tensor<string, []>("op_14778_cast")];
            tensor<fp16, [2, 320, 1, 4096]> inputs_cast = add(x = var_14778_cast, y = inputs_93_cast)[name = tensor<string, []>("inputs_cast")];
            tensor<int32, [1]> var_14782 = const()[name = tensor<string, []>("op_14782"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_cast = reduce_mean(axes = var_14782, keep_dims = var_10370, x = inputs_cast)[name = tensor<string, []>("channels_mean_cast")];
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_cast = sub(x = inputs_cast, y = channels_mean_cast)[name = tensor<string, []>("zero_mean_cast")];
            tensor<fp16, [2, 320, 1, 4096]> zero_mean_sq_cast = mul(x = zero_mean_cast, y = zero_mean_cast)[name = tensor<string, []>("zero_mean_sq_cast")];
            tensor<int32, [1]> var_14786 = const()[name = tensor<string, []>("op_14786"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_14787_cast = reduce_mean(axes = var_14786, keep_dims = var_10370, x = zero_mean_sq_cast)[name = tensor<string, []>("op_14787_cast")];
            tensor<fp16, []> var_14788_to_fp16 = const()[name = tensor<string, []>("op_14788_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_14789_cast = add(x = var_14787_cast, y = var_14788_to_fp16)[name = tensor<string, []>("op_14789_cast")];
            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_14789_cast)[name = tensor<string, []>("denom_cast")];
            tensor<fp16, [2, 320, 1, 4096]> out_cast = mul(x = zero_mean_cast, y = denom_cast)[name = tensor<string, []>("out_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_14793_to_fp16 = const()[name = tensor<string, []>("op_14793_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(644153152)))];
            tensor<fp16, [2, 320, 1, 4096]> var_14794_cast = add(x = out_cast, y = var_14793_to_fp16)[name = tensor<string, []>("op_14794_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_14796_to_fp16 = const()[name = tensor<string, []>("op_14796_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(644153856)))];
            tensor<fp16, [2, 320, 1, 4096]> input_521_cast = mul(x = var_14794_cast, y = var_14796_to_fp16)[name = tensor<string, []>("input_521_cast")];
            tensor<int32, [2]> var_14804 = const()[name = tensor<string, []>("op_14804"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14806 = const()[name = tensor<string, []>("op_14806"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_14808_pad_type_0 = const()[name = tensor<string, []>("op_14808_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_14808_pad_0 = const()[name = tensor<string, []>("op_14808_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [614400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(644154560))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(644769024))), name = tensor<string, []>("up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([2560, 320, 1, 1])];
            tensor<fp16, [2560]> up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(644769216))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(644771200))), name = tensor<string, []>("up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([2560])];
            tensor<fp16, [2, 2560, 1, 4096]> var_14808_cast = conv(bias = up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_14806, groups = var_10375, pad = var_14808_pad_0, pad_type = var_14808_pad_type_0, strides = var_14804, weight = up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_521_cast)[name = tensor<string, []>("op_14808_cast")];
            tensor<int32, [2]> var_14809_split_sizes_0 = const()[name = tensor<string, []>("op_14809_split_sizes_0"), val = tensor<int32, [2]>([1280, 1280])];
            tensor<int32, []> var_14809_axis_0 = const()[name = tensor<string, []>("op_14809_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 1280, 1, 4096]> var_14809_cast_0, tensor<fp16, [2, 1280, 1, 4096]> var_14809_cast_1 = split(axis = var_14809_axis_0, split_sizes = var_14809_split_sizes_0, x = var_14808_cast)[name = tensor<string, []>("op_14809_cast")];
            tensor<string, []> var_14811_mode_0 = const()[name = tensor<string, []>("op_14811_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 1280, 1, 4096]> var_14811_cast = gelu(mode = var_14811_mode_0, x = var_14809_cast_1)[name = tensor<string, []>("op_14811_cast")];
            tensor<fp16, [2, 1280, 1, 4096]> input_523_cast = mul(x = var_14809_cast_0, y = var_14811_cast)[name = tensor<string, []>("input_523_cast")];
            tensor<int32, [2]> var_14815 = const()[name = tensor<string, []>("op_14815"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14817 = const()[name = tensor<string, []>("op_14817"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_14819_pad_type_0 = const()[name = tensor<string, []>("op_14819_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_14819_pad_0 = const()[name = tensor<string, []>("op_14819_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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(644771392))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(645078656))), name = tensor<string, []>("up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 1280, 1, 1])];
            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, []>(645078848)))];
            tensor<fp16, [2, 320, 1, 4096]> var_14819_cast = conv(bias = up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_14817, groups = var_10375, pad = var_14819_pad_0, pad_type = var_14819_pad_type_0, strides = var_14815, weight = up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_523_cast)[name = tensor<string, []>("op_14819_cast")];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_333_cast = add(x = var_14819_cast, y = inputs_cast)[name = tensor<string, []>("hidden_states_333_cast")];
            tensor<int32, [4]> var_14821 = const()[name = tensor<string, []>("op_14821"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> input_525_cast = reshape(shape = var_14821, x = hidden_states_333_cast)[name = tensor<string, []>("input_525_cast")];
            tensor<int32, [2]> var_14825 = const()[name = tensor<string, []>("op_14825"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14827 = const()[name = tensor<string, []>("op_14827"), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(645079552))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(645156416))), name = tensor<string, []>("up_blocks_3_attentions_2_proj_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 320, 1, 1])];
            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, []>(645156608)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_cast = conv(bias = up_blocks_3_attentions_2_proj_out_bias_to_fp16, dilations = var_14827, groups = var_10375, pad = hidden_states_pad_0, pad_type = hidden_states_pad_type_0, strides = var_14825, weight = up_blocks_3_attentions_2_proj_out_weight_to_fp16_palettized, x = input_525_cast)[name = tensor<string, []>("hidden_states_cast")];
            tensor<fp16, [2, 320, 64, 64]> input_527_cast = add(x = hidden_states_cast, y = hidden_states_323_cast)[name = tensor<string, []>("input_527_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)[name = tensor<string, []>("reshape_240_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)[name = tensor<string, []>("reduce_mean_180_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> sub_120_cast = sub(x = reshape_240_cast, y = reduce_mean_180_cast)[name = tensor<string, []>("sub_120_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> square_60_cast = square(x = sub_120_cast)[name = tensor<string, []>("square_60_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)[name = tensor<string, []>("reduce_mean_182_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)[name = tensor<string, []>("add_120_cast")];
            tensor<fp16, [2, 32, 1, 1, 1]> sqrt_60_cast = sqrt(x = add_120_cast)[name = tensor<string, []>("sqrt_60_cast")];
            tensor<fp16, [2, 32, 10, 64, 64]> real_div_60_cast = real_div(x = sub_120_cast, y = sqrt_60_cast)[name = tensor<string, []>("real_div_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)[name = tensor<string, []>("reshape_241_cast")];
            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, []>(645157312)))];
            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, []>(645158016)))];
            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_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_241_cast)[name = tensor<string, []>("add_121_cast")];
            tensor<fp16, [2, 320, 64, 64]> input_cast = silu(x = add_121_cast)[name = tensor<string, []>("input_cast")];
            tensor<int32, []> var_14841 = const()[name = tensor<string, []>("op_14841"), val = tensor<int32, []>(1)];
            tensor<int32, [2]> var_14844 = const()[name = tensor<string, []>("op_14844"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14846 = const()[name = tensor<string, []>("op_14846"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_14848_pad_type_0 = const()[name = tensor<string, []>("op_14848_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_14848_pad_0 = const()[name = tensor<string, []>("op_14848_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<fp16, [4, 320, 3, 3]> conv_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [8640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(645158720))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(645167424))), name = tensor<string, []>("conv_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([4, 320, 3, 3])];
            tensor<fp16, [4]> conv_out_bias_to_fp16 = const()[name = tensor<string, []>("conv_out_bias_to_fp16"), val = tensor<fp16, [4]>([-0x1.c54p-10, -0x1.88p-10, -0x1.834p-13, -0x1.bfp-9])];
            tensor<fp16, [2, 4, 64, 64]> var_14848_cast = conv(bias = conv_out_bias_to_fp16, dilations = var_14846, groups = var_14841, pad = var_14848_pad_0, pad_type = var_14848_pad_type_0, strides = var_14844, weight = conv_out_weight_to_fp16_palettized, x = input_cast)[name = tensor<string, []>("op_14848_cast")];
            tensor<string, []> var_14848_cast_to_fp32_dtype_0 = const()[name = tensor<string, []>("op_14848_cast_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
            tensor<fp32, [2, 4, 64, 64]> noise_pred = cast(dtype = var_14848_cast_to_fp32_dtype_0, x = var_14848_cast)[name = tensor<string, []>("cast_0")];
        } -> (noise_pred);
}